Kicksta Research
A mixed-methods study of 377 consumers across four commercial niches. The high-follower brand won all eight comparisons — including among people who say follower counts can be bought.
Kicksta sells an Instagram growth service. We are therefore an interested party in a study about whether follower counts matter, and you should read what follows with that in mind.
Three things we did about it. The brand profiles used as stimuli were anonymised and re-rendered before anyone saw them, so no respondent knew which brands they were judging, and none of them were customers of ours. The stated purpose was withheld during fieldwork — respondents were told this was a study of first impressions of brand social media presence, not a study of follower counts — because telling people what you are measuring is the fastest way to stop measuring it. And we are publishing the findings that are inconvenient for us as prominently as the ones that are not.
The most inconvenient one: the high-follower accounts in this study had engagement rates roughly four times lower than the small accounts they beat. Across all sixteen profiles, follower count and engagement rate were negatively correlated at r = −.874. The number that wins attention and the number that indicates a healthy audience pointed in opposite directions. That finding is in here, in full, with the data behind it.
We also want to be plain about what this study does not show. It compared accounts with millions of followers against accounts with thousands — a ratio of roughly 297 to 1. It cannot tell you how many followers are enough, it did not test any threshold, and nothing in it should be read as a claim about what any growth service delivers. The study's own recommendation to brands is "do not purchase followers", and we agree with it.
Instagram presents a brand's follower count as one of the first pieces of numerical information a visitor encounters. This study asks whether that single, easily manipulated number carries measurable weight in consumer judgement, and whether its influence operates directly on behavioural intention or indirectly through the perception of trustworthiness.
A three-phase mixed-methods design was used. 377 respondents evaluated sixteen anonymised Instagram brand profiles distributed evenly across four niches — beauty, clothing, food and the informational field of news and media — with each niche represented by two high-follower profiles (1.8–8.2 million followers) and two low-follower profiles (6.4–26.1 thousand followers). Respondents completed eight forced-choice paired comparisons and rated every profile on validated multi-item scales of perceived brand trust and purchase or engagement intention. A qualitative phase of semi-structured interviews was analysed through reflexive thematic analysis, and a third phase applied structured content analysis to the sixteen stimulus profiles.
The findings are consistent and unambiguous. In every one of the eight paired comparisons, and therefore in all four niches, the profile selected by the majority was the high-follower profile; no low-follower profile was preferred in any pairing. Mean perceived brand trust was 5.55 for high-follower profiles against 4.08 for low-follower profiles on a seven-point scale, and mean purchase or engagement intention was 5.21 against 3.76. Follower count, expressed logarithmically, correlated positively with trust (r = .612) and with intention (r = .547), while trust and intention correlated at r = .694. Hierarchical regression showed that follower tier explained an additional 22.6% of variance in intention beyond demographic and usage controls, and that perceived trust partially mediated the relationship, carrying 61.2% of the total effect. The effect was strongest in the informational niche and weakest in food.
Notably, 58.7% of respondents agreed that follower counts can be bought, and 38.2% agreed that a high count therefore does not indicate a trustworthy brand — yet 82.1% of those sceptics still selected the high-follower profile. The interview data explain this attitude–behaviour gap as the difference between what people believe about the metric and what they do with it under time pressure.
The study concludes that a high follower count functions as a peripheral persuasion cue that raises perceived brand trust, and that elevated trust in turn raises purchase intention. Follower count therefore affects purchase intention both directly and, more substantially, through trust.
Keywords: Instagram; follower count; brand trust; purchase intention; social proof; signalling theory; heuristic processing; social media marketing.
Each finding stands alone if quoted out of context. n = 377 respondents, 6,032 profile evaluations, 3,016 forced choices.
Across eight forced-choice pairings spanning four niches, 85.7% of selections went to the high-follower profile. No low-follower profile secured a majority in any pairing, in any category. At n = 377 that headline figure carries a 95% confidence interval of ±3.5 percentage points.
On a seven-point scale — 5.55 against 4.08 (t = 24.87, p < .001, d = 1.38). Low-follower brands sat barely above the 4.0 midpoint, indicating indifference rather than active distrust.
5.21 against 3.76, a gap of 1.45 points (d = 1.26). The low-tier mean fell below the scale midpoint in three of the four niches — net disinclination, not neutrality.
Adding follower tier to the regression raised explained variance from 8.1% to 30.7% — an increment of 22.6 percentage points, roughly three times the contribution of all demographic controls together.
Bootstrapped mediation with 5,000 resamples found an indirect effect through perceived trust of β = .335, 95% CI [.281, .392]. The direct path remained significant, so mediation is partial: about two fifths of the effect bypasses trust entirely and is driven by bandwagon and identity motives.
58.7% of respondents agreed that follower counts can be bought. Among the most sceptical, 82.1% still chose the high-follower brand, against 87.9% of the least sceptical — a statistically significant but negligible difference (Cramér's V = .11). Explicit disbelief attenuated the effect by about one sixth.
The trust gap ran from 1.92 points in news and media down to 1.02 in food, where small accounts carry a competing artisanal association. Even in food, 79.8% still chose the larger account. No subgroup and no category showed an advantage for small audiences.
High-follower profiles averaged 1.08% engagement against 4.36% for the low-follower profiles they beat, and replied to comments less than half as often (37.5% against 87.5%). 68.4% of interviewees named the engagement-to-follower ratio as their method for spotting inflated accounts. Had they applied it, every result in this study would have reversed.
The short version, for readers who will not read Section 3. The full methodology follows it.
Design. Three-strand convergent mixed-methods study (Creswell & Plano Clark, 2018): an online survey, semi-structured interviews, and a structured content analysis of the stimulus material.
Respondents. 377 adult Instagram users. Fieldwork ran from 1 March to 15 July 2026. Recruitment was by direct outreach rather than a commercial research panel: the team approached universities, companies and independent freelancers, and distributed the questionnaire by email invitation and through the research team's own social accounts. Participation was voluntary and completed remotely. Responses were quota-sampled on age and gender against the published distribution of the adult Instagram user base rather than the general population. Screening excluded anyone using Instagram less than weekly, anyone not completing all sixteen evaluations, and anyone failing either of two embedded attention checks. Straight-lined responses were removed before analysis. Sample spans Europe (44.1%), North America (26.8%), Asia (18.3%) and other regions (10.8%).
Stimuli. Sixteen real Instagram brand profiles, anonymised and re-rendered to a common format: four niches (beauty, clothing, food, informational) × two high-follower + two low-follower accounts. High tier averaged 4.39 million followers, low tier 14,760 — a ratio of approximately 297:1.
Tasks. Eight forced-choice paired comparisons, each pitting a high- against a low-follower profile from the same niche, followed by individual ratings of all sixteen profiles on multi-item trust and intention scales. Presentation order was randomised at two levels and left–right position counterbalanced. Backward navigation was disabled so earlier choices could not be revised.
Analysis. Chi-square goodness-of-fit, paired-samples t-tests, two-way mixed ANOVA, Pearson correlation, hierarchical multiple regression, and bootstrapped mediation following Hayes (2018), Model 4, with 5,000 resamples and bias-corrected accelerated confidence intervals.
Reliability. All multi-item scales exceeded α = .70, with the two focal constructs above .88. Intercoder reliability for the content analysis was Cohen's κ = .87. Harman's single-factor test returned 31.4% of variance on the first unrotated factor, below the 50% threshold of concern for common method bias.
Social media platforms have quietly re-engineered the way commercial credibility is established. For most of the twentieth century, a consumer assessing an unfamiliar brand relied on cues that were expensive to fake: a shopfront on a good street, a national advertising campaign, distribution in a reputable retailer, a printed catalogue. Each of these signals required capital, and the capital requirement was precisely what made the signal informative. A brand that could afford a prime-time television slot was, by that fact alone, a brand of some substance. The digital environment has not abolished this logic, but it has changed the currency in which it is denominated.
On Instagram, the most prominent quantitative signal attached to any commercial account is its follower count. It appears at the top of the profile, adjacent to the brand name, before any product image, price, review or description is processed. It is rendered in the same visual weight as the account name itself. A visitor arriving from a search result, a hashtag page or a friend's story therefore encounters the number before encountering the brand's actual proposition. Where a shop window once communicated scale through rent, a profile header communicates it through a four-, five-, six- or seven-digit integer.
This arrangement raises a question that is simultaneously commercial and psychological. Follower count is a measure of audience size. It is not, in any direct sense, a measure of product quality, service reliability, fulfilment speed, ethical conduct or financial solvency. Yet the structure of the interface positions it as the headline fact about the account. If consumers treat that number as diagnostic of qualities it does not actually measure, then a substantial proportion of commercial trust on the platform is being allocated on the basis of a proxy that is only loosely coupled to the thing it is taken to represent — and, critically, a proxy that can be purchased on open markets for a few hundred units of currency per ten thousand followers.
The commercial stakes are considerable. Instagram is no longer merely a discovery channel; it is a transactional surface, with in-app storefronts, checkout functionality, shoppable posts and direct-message-based customer service. Small and medium enterprises in the beauty, apparel and food sectors frequently treat the platform as their primary storefront, and media organisations increasingly treat it as a primary distribution channel rather than a secondary one. If follower count materially shapes whether a consumer trusts and buys, then follower acquisition ceases to be a vanity exercise and becomes a competitive necessity — with all the incentives towards artificial inflation that this implies.
Existing scholarship has examined this question chiefly in the context of individual influencers, where follower count intersects with parasocial attachment, perceived authenticity and personal likeability. De Veirman, Cauberghe and Hudders (2017) demonstrated that influencers with very high follower counts were perceived as more likeable, though the relationship was moderated by the divergence of the products they endorsed. Djafarova and Rushworth (2017) found that follower numbers shaped the credibility young female users assigned to Instagram personalities. What has received markedly less attention is the equivalent question for brand accounts, where parasocial mechanisms are weaker or absent, where the account represents a commercial entity rather than a person, and where the consumer's decision is a purchase decision rather than an attitudinal one. This study addresses that gap.
The significance of the study operates on four levels.
The overarching aim of this study is to determine whether, and through what psychological pathway, the follower count displayed on an Instagram brand profile is statistically related to perceived brand trust and to purchase intention among consumers.
This aim is decomposed into seven objectives:
Six hypotheses follow from these objectives and are stated formally in Section 2.7 and tested in Section 4.
The study adopts a convergent mixed-methods design in the sense described by Creswell and Plano Clark (2018), in which quantitative and qualitative strands are conducted in parallel and integrated at the interpretation stage. Three data strands were collected.
The first strand is a structured online survey built around sixteen anonymised Instagram brand profiles: four niches (beauty, clothing, food, informational) each represented by two high-follower and two low-follower accounts. Respondents completed eight forced-choice paired comparisons, each pitting one high-follower profile against one low-follower profile from the same niche, and then rated every profile on multi-item scales of perceived brand trust and purchase or engagement intention. Demographic, platform-usage and follower-scepticism measures were also collected.
The second strand consists of semi-structured interviews with a purposively selected subset of survey respondents, exploring how participants describe their own reasoning when they encounter a follower count. Transcripts were analysed using the six-phase reflexive thematic analysis procedure of Braun and Clarke (2006).
The third strand is a structured content analysis of the sixteen stimulus profiles across eleven coded attributes — verification status, posting frequency, engagement rate, bio completeness, external link presence, use of user-generated content, comment responsiveness, story highlights, caption length, branded hashtag use and photographic production quality — conducted by two independent coders in order to establish which observable profile characteristics co-vary with follower count and might therefore constitute confounds.
Analysis employed descriptive statistics, chi-square goodness-of-fit tests, paired-samples t-tests, two-way mixed ANOVA, Pearson correlation, hierarchical multiple regression and bootstrapped mediation analysis following the procedures set out by Hayes (2018). Full detail is given in Section 3.
The scope of the study is deliberately bounded. It concerns brand accounts rather than individual influencer accounts, because the psychological mechanisms differ materially between the two. It concerns four commercial niches selected to span a spectrum from high-involvement, high-risk purchasing (beauty) through moderate-involvement purchasing (clothing) to low-involvement, locally-consumed purchasing (food) and finally to a non-transactional domain (informational accounts), where the behavioural outcome is consumption and subscription rather than purchase. It concerns Instagram specifically and does not claim generalisability to platforms whose interface architecture differs. It concerns stated intention rather than observed transaction, a limitation discussed in Section 3.8.
This report is organised into six sections. Section 1 introduces the problem. Section 2 reviews the relevant literature and develops the conceptual framework and hypotheses. Section 3 sets out the methodology. Section 4 presents the findings of all three data strands in sequence. Section 5 discusses the results in relation to prior work and draws out their implications. Section 6 concludes and offers recommendations. References and appendices follow.
Four features distinguish this study from the existing body of work.
The literature relevant to this study sits at the intersection of four traditions that developed largely independently and have only recently begun to converge. The first is the social-psychological literature on conformity and social influence, which established that individuals use the behaviour of others as evidence about the world. The second is the economics of signalling, which formalised the conditions under which an observable attribute can convey credible information about an unobservable one. The third is the persuasion literature, which distinguishes between effortful and heuristic routes to attitude formation. The fourth is the marketing literature on brand trust and purchase intention, which supplies the dependent constructs and their validated measurement.
Each tradition offers a partial account of why a follower count might matter. None of them, taken alone, predicts the specific pattern this study observes. The review therefore proceeds through each in turn before assembling the integrated framework presented in Section 2.7.
The foundational insight is Deutsch and Gerard's (1955) distinction between normative and informational social influence. Normative influence operates through the desire to be accepted; informational influence operates through the assumption that others possess information one lacks. Asch's (1951) conformity experiments demonstrated the power of the former even in perceptually unambiguous situations, but it is the latter that bears more directly on the present study. When a consumer confronts an unfamiliar brand, they are in a state of genuine uncertainty, and the number of people who have already chosen to follow that brand constitutes evidence — imperfect but not worthless — that others have evaluated it favourably.
Cialdini (2009) formalised this as the principle of social proof and specified two conditions under which it exerts maximum force: uncertainty and similarity. Both conditions are characteristically satisfied on Instagram. Uncertainty is high because the consumer typically cannot inspect the product, cannot verify the seller's identity and has no prior relationship. Similarity is high because the platform's recommendation architecture surfaces accounts followed by one's own network and by demographically comparable users. Social proof theory therefore predicts a positive relationship between follower count and favourable brand evaluation, and predicts further that the relationship should be strongest where uncertainty is greatest.
Spence (1973) introduced the concept of a signal as an observable attribute, acquired at some cost, that conveys information about an unobservable quality. The critical requirement is the single-crossing condition: the signal must be cheaper to produce for high-quality types than for low-quality types. Education signals productivity in Spence's original model precisely because it is less onerous for the able. Erdem and Swait (1998) extended the logic to brands, arguing that brand equity functions as a credible signal of product quality because a firm that misrepresents quality forfeits the accumulated investment. Kirmani and Rao (2000) surveyed the conditions under which such signals remain credible.
Follower count occupies an awkward position within this framework. Organic follower accumulation is genuinely costly: it requires sustained content production, audience engagement and, usually, time. To that extent, a large following is a legitimate signal of commitment and, indirectly, of the resources and competence that sustained commitment implies. However, the single-crossing condition is violated by the existence of follower markets, where the cost of acquiring followers is essentially uncoupled from the underlying quality of the brand. Signalling theory therefore yields an ambivalent prediction: the signal should influence consumers who treat it as costly, but should be discounted by consumers who are aware of the market for it. Testing whether awareness actually produces discounting is one of the empirical contributions of this study.
The Elaboration Likelihood Model (Petty & Cacioppo, 1986) and the Heuristic–Systematic Model (Chaiken, 1980) converge on the proposition that persuasion proceeds through two routes. The central or systematic route involves effortful scrutiny of argument quality; the peripheral or heuristic route relies on simple cues that permit rapid judgement with minimal cognitive expenditure. Route selection depends on motivation and ability.
Instagram is, almost by design, a low-elaboration environment. Content is consumed at speed, in a vertically scrolling feed, frequently on a small screen and often in fragments of otherwise occupied time. Motivation to elaborate on any individual brand profile is typically low, and the interface supplies few of the argument-quality inputs that systematic processing would require. Under these conditions, dual-process theory predicts that peripheral cues will dominate — and a follower count, being numerically precise, prominently positioned and requiring no interpretation, is close to an ideal peripheral cue. Tversky and Kahneman's (1974) account of judgemental heuristics supplies the more general mechanism: under uncertainty and time pressure, people substitute an available attribute for the target attribute they cannot directly assess.
Hovland, Janis and Kelley (1953) established source credibility as a determinant of persuasion, decomposing it into expertise and trustworthiness; Ohanian (1990) operationalised the construct for commercial endorsement contexts. Sundar (2008) adapted this tradition for digital interfaces through the MAIN model, which proposes that technological affordances — modality, agency, interactivity and navigability — trigger cognitive heuristics that shape credibility judgements. The bandwagon heuristic, in Sundar's terminology, is cued by any interface element that aggregates the choices of other users. Follower counts, like ratings, view counts and share counts, are precisely such elements. Metzger and Flanagin (2013) documented the pervasiveness of these aggregate cues in online credibility assessment.
The visible audience metric has a longer history than the platforms that now display it. Circulation figures for newspapers, audited by independent bureaux from the early twentieth century, performed an equivalent function for advertisers and, indirectly, for readers. Broadcast ratings did the same for television. What changed with the arrival of social platforms was that the metric became consumer-facing rather than trade-facing, and real-time rather than periodic.
The evolution can be traced through four phases. In the first phase, roughly 2004 to 2010, follower and friend counts on early social networks were primarily social rather than commercial; they described personal popularity and were rarely interpreted as commercial credentials. In the second phase, roughly 2010 to 2015, brands migrated onto visual platforms at scale and the follower count acquired a commercial reading; the phrase 'vanity metric' entered practitioner discourse precisely because the number had begun to be treated as consequential enough to warrant scepticism. In the third phase, roughly 2015 to 2020, the professionalisation of influencer marketing produced formal rate cards indexed to follower count, and simultaneously produced the market for artificial followers that the rate cards incentivised; this is the period in which platform enforcement actions against inauthentic engagement began. The fourth phase, from approximately 2020 onwards, has been characterised by a partial retreat from raw follower count in professional practice — agencies increasingly price on engagement rate, conversion or audience authenticity scores — even as the platform interface continues to display raw counts to ordinary users with undiminished prominence.
This divergence between professional and lay metrics is central to the present study. The sophisticated buyer of advertising has largely stopped treating follower count as a sufficient statistic. The ordinary consumer, encountering a profile header, has no equivalent alternative available at the moment of judgement.
Three contemporary developments condition the interpretation of any findings in this area.
The first is the rise of the micro-influencer thesis, which holds that smaller accounts generate proportionally higher engagement and stronger audience relationships, and are therefore more commercially effective per unit of audience. The empirical support for this proposition is reasonably strong on the engagement side. Whether it extends to consumer trust in brand accounts is less clear, and the present study finds that it does not — at least not at the level of first-impression judgement.
The second is the intensifying salience of authenticity as a consumer value. Nunes, Ordanini and Giambastiani (2021) provide a systematic account of what consumers mean by authenticity and identify accuracy, connectedness and integrity as its principal components. Authenticity discourse would predict that consumers penalise accounts whose scale appears manufactured. The findings reported here suggest that this penalty, if it exists, is applied inconsistently and mostly in retrospective justification rather than in the moment of choice.
The third is algorithmic mediation. Instagram's ranking systems tend to surface content that already performs well, producing a reinforcing dynamic in which large accounts receive disproportionate distribution and thereby grow larger. Consumers observing a large follower count are therefore observing, in part, an artefact of algorithmic amplification rather than an unmediated aggregation of independent choices — a fact that weakens the informational warrant for the social-proof inference without weakening the inference itself.
Several adjacent frameworks inform the design of this study without being tested directly.
The empirical literature most directly relevant to this study concerns follower count as an independent variable in evaluative judgement.
De Veirman, Cauberghe and Hudders (2017) conducted what remains the most frequently cited experimental treatment. Manipulating the follower count of an Instagram profile, they found that a higher count increased perceived likeability and perceived opinion leadership, but that the advantage was attenuated when the influencer followed a divergent set of products, which reduced perceived expertise. Their design concerned individual influencers rather than brand accounts, and their dependent variables were attitudinal rather than behavioural.
Jin, Muqaddam and Ryu (2019) compared Instagram-native celebrities with traditional celebrities and found that the former generated stronger brand attitudes and greater social presence, attributing the difference to perceived accessibility. Ki and Kim (2019) modelled the persuasion mechanism through which influencers convert audience attention into purchase intention, identifying perceived taste leadership and prestige as key intervening variables. Lou and Yuan (2019) demonstrated that influencer credibility — but not follower count directly — predicted trust in branded content, suggesting that the two may be separable.
Casaló, Flavián and Ibáñez-Sánchez (2020) identified originality and uniqueness of content as the principal antecedents of opinion leadership on Instagram, with follower count playing a subordinate role once content characteristics were controlled. Djafarova and Rushworth (2017), by contrast, found in a qualitative study that follower numbers featured prominently in young female users' explicit accounts of how they judged credibility.
Two observations follow from this body of work. First, the evidence is more consistent for influencers than for brands, and the extrapolation from one to the other is not obviously safe: a brand account cannot draw on parasocial intimacy in the way a personal account can, so if follower count matters for brands it must be operating through a different channel — most plausibly a pure social-proof or signalling channel rather than a relational one. Second, almost no study in this literature simultaneously measures trust, tests mediation, varies commercial niche and captures respondents' explicit beliefs about the metric's validity. The present study is designed to fill that combination of gaps.
Drawing the foregoing together, the study proposes that follower count operates as a peripheral cue that activates a social-proof heuristic, which elevates perceived brand trust, which in turn elevates behavioural intention. A residual direct path from follower count to intention is expected, reflecting bandwagon and identity-signalling motives that do not pass through trust. The strength of the whole system is expected to vary with the commercial domain and with individual-difference variables. The framework is presented in Figure 2.1.
Six hypotheses are advanced:
The study employs a convergent parallel mixed-methods design (Creswell & Plano Clark, 2018), combining a quantitative survey experiment, a qualitative interview strand and a structured content analysis. The three strands were designed to answer complementary questions: the survey establishes whether an effect exists and how large it is; the interviews establish how participants themselves account for the effect; the content analysis establishes what else varies alongside follower count and might therefore constitute a confound. Integration occurs at the interpretation stage rather than at the data-collection stage.
The quantitative strand is best characterised as a within-subjects factorial design. Every respondent was exposed to all sixteen stimulus profiles, so follower tier (high, low) and niche (beauty, clothing, food, informational) both operate as within-subjects factors. This structure maximises statistical power, controls entirely for between-person variation in scale use, and permits paired-samples testing. Its principal cost is the risk of demand characteristics: a respondent who sees both a large and a small account in the same category may infer the study's purpose. Three mitigations were applied — profile presentation order was randomised, follower counts were displayed in the natural profile-header format rather than being drawn attention to, and the stated purpose given to respondents was framed as a study of 'first impressions of brand social media presence' rather than of follower counts specifically. A post-experimental suspicion probe was included; respondents who correctly identified the manipulation before the debrief section were flagged, and sensitivity analyses excluding them did not alter any substantive conclusion.
The target population comprises adult Instagram users who use the platform at least weekly and who have at some point encountered commercial content on it. The unit of analysis for the quantitative strand is the respondent–profile evaluation; for the content-analysis strand it is the profile.
The object of the study is the set of sixteen Instagram brand profiles constructed as stimuli. Four commercial and informational niches were selected on theoretical grounds, to span a range of perceived purchase risk and involvement:
Within each niche, two accounts in the high-follower tier and two in the low-follower tier were selected, giving sixteen stimulus profiles in total. All profiles were anonymised: brand names, logos and any distinguishing textual identifiers were replaced with neutral pseudonyms and generic wordmarks, and the profiles were re-rendered in a standardised mock interface so that no respondent could recognise the underlying brand. This procedure removes prior brand familiarity as a confound — a necessary step, since familiarity and follower count are naturally correlated in the wild. Only structural and quantitative characteristics were preserved: follower count, following count, post count, engagement pattern, grid composition and posting cadence.
| Code | Niche | Tier | Followers | Posts | Following |
|---|---|---|---|---|---|
| BH1 | Beauty | High | 4,800,000 | 3,412 | 184 |
| BH2 | Beauty | High | 2,600,000 | 2,875 | 231 |
| BL1 | Beauty | Low | 18,400 | 486 | 1,204 |
| BL2 | Beauty | Low | 7,900 | 312 | 893 |
| CH1 | Clothing | High | 6,100,000 | 4,108 | 97 |
| CH2 | Clothing | High | 3,300,000 | 3,266 | 162 |
| CL1 | Clothing | Low | 22,700 | 641 | 1,517 |
| CL2 | Clothing | Low | 11,200 | 398 | 1,082 |
| FH1 | Food | High | 3,900,000 | 2,940 | 128 |
| FH2 | Food | High | 1,800,000 | 2,214 | 205 |
| FL1 | Food | Low | 15,600 | 527 | 946 |
| FL2 | Food | Low | 6,400 | 289 | 774 |
| IH1 | Informational | High | 8,200,000 | 9,846 | 63 |
| IH2 | Informational | High | 4,400,000 | 7,193 | 88 |
| IL1 | Informational | Low | 26,100 | 1,472 | 1,338 |
| IL2 | Informational | Low | 9,800 | 804 | 1,026 |
A two-stage sampling strategy was applied. The analysed quantitative sample was 377 respondents, giving 6,032 profile evaluations (377 × 16 profiles) and 3,016 forced-choice comparisons (377 × 8 pairings). Fieldwork ran from 1 March to 15 July 2026. For the quantitative strand, non-probability quota sampling was used, with quotas set on age band and gender to approximate the published age and gender distribution of the adult Instagram user base rather than that of the general population. Recruitment proceeded by direct outreach rather than through a commercial research panel: universities, companies and independent freelancers were approached, and the questionnaire was distributed by email invitation and through the research team's own social accounts. Participation was voluntary and completed remotely at respondents' convenience. Screening questions excluded respondents who reported using Instagram less than weekly, who did not complete all sixteen profile evaluations, or who failed either of two embedded attention checks. Straight-lining responses — identical answers across an entire reversed-item block — were removed prior to analysis.
For the qualitative strand, purposive maximum-variation sampling was applied to survey respondents who had consented to follow-up contact. Selection deliberately sought variation on three dimensions: age band, declared follower-scepticism score, and consistency of forced-choice behaviour. Respondents whose survey behaviour was internally inconsistent — high scepticism combined with uniform high-follower selection — were intentionally over-sampled, since their accounts were expected to be the most theoretically informative. Interviewing continued until no new codes emerged across three consecutive interviews, the standard operational criterion for thematic saturation.
For the content-analysis strand, the sample is the complete set of sixteen stimulus profiles; no sampling was involved.
The survey instrument comprised five sections. Section A collected demographic and platform-usage data. Section B presented the eight forced-choice paired comparisons. Section C presented all sixteen profiles individually for scale rating. Section D administered the follower-scepticism scale and the Instagram use-intensity scale. Section E contained the suspicion probe and debriefing. The full instrument appears in Appendix A.
All attitudinal items used seven-point Likert scales anchored at 1 = strongly disagree and 7 = strongly agree. Scales were adapted from established sources rather than newly written, and item wording was adjusted only to the minimum extent necessary to fit the Instagram brand-account context.
| Construct | Items | Adapted from | α | CR | AVE |
|---|---|---|---|---|---|
| Perceived brand trust | 5 | Chaudhuri & Holbrook (2001); Mayer et al. (1995) | .912 | .918 | .692 |
| Purchase / engagement intention | 3 | Dodds, Monroe & Grewal (1991) | .884 | .889 | .728 |
| Instagram use intensity | 4 | Ellison-type usage measures | .796 | .803 | .571 |
| Follower-count scepticism | 3 | Developed for this study | .812 | .819 | .604 |
Note. α = Cronbach's alpha; CR = composite reliability; AVE = average variance extracted. All α values exceed the .70 threshold recommended by Nunnally (1978); all CR values exceed .70 and all AVE values exceed .50, satisfying the convergent-validity criteria of Fornell and Larcker (1981).
The forced-choice task presented two profiles side by side, one from each tier within the same niche, with the instruction: 'If you needed a product or service in this category today and had to pick one of these two accounts to buy from, which would you choose?' For the informational niche the wording was adjusted to 'which would you choose to follow for news?'. No neutral option was offered, since the purpose of the task was to force a discriminative judgement.
The interview guide, reproduced in Appendix B, was semi-structured around four domains: general account-evaluation habits; reactions to specific stimulus profiles; explicit reasoning about follower counts; and reflections on the respondent's own survey choices, which were shown back to them during the interview. The final domain proved the most productive and generated the majority of the material underpinning Themes 3 and 6.
The content-analysis coding sheet, reproduced in Appendix C, specified eleven variables with explicit operational definitions and decision rules — for example, engagement rate was computed as the mean of (likes + comments) ÷ followers across the twelve most recent non-promotional grid posts, excluding posts published within seventy-two hours of coding.
Data collection proceeded in three overlapping stages. Stimulus construction and pilot testing came first: a pilot administration of the survey identified two ambiguous items in the scepticism scale, which were reworded, and established a median completion time of just under fourteen minutes, judged acceptable for retention.
The main survey was then administered online, self-completed, on desktop and mobile devices. Profile images were rendered at device-appropriate resolution and were confirmed legible on screens as small as 5.4 inches. The instrument enforced completion of each section before progression but permitted no backward navigation, preventing respondents from revising earlier forced choices after seeing later stimuli. Presentation order was randomised at two levels: the order of the eight paired comparisons, and the order of the sixteen individual profile ratings. Left–right position within each pair was counterbalanced.
Interviews were conducted by video call, lasted between thirty-eight and sixty-two minutes, and were audio-recorded with consent. Recordings were transcribed verbatim and anonymised at transcription, with participants assigned codes of the form P01, P02 and so on. Transcripts were returned to participants for member checking; two participants requested minor clarifications, which were incorporated.
Content-analysis coding was performed independently by two coders working from the same coding sheet without consultation. Disagreements were resolved through discussion after the reliability statistic had been computed, so that the reported figure reflects genuinely independent coding.
Participation was voluntary and preceded by an informed-consent screen stating the general purpose of the research, the approximate duration, the right to withdraw at any point without penalty, and the intended use of the data. No personally identifying information was collected in the survey; contact details supplied for interview follow-up were stored separately from response data and destroyed after the interview stage. Interview participants gave separate written consent to audio recording.
The study involved a mild deception, in that the stated purpose withheld the specific focus on follower count in order to avoid demand characteristics. This was judged proportionate: the deception carried no foreseeable risk of harm, was disclosed in a full debriefing screen at the end of the survey, and respondents were offered the option to withdraw their data after the debrief. No respondent exercised this option.
Anonymisation of the stimulus profiles served an ethical as well as a methodological purpose, since it prevented the study from generating evaluative data attributable to identifiable commercial entities. No data were collected from minors; the consent screen required confirmation of adult status.
This section reports findings from the three data strands in sequence. Section 4.1 describes the respondent profile. Section 4.2 presents the quantitative results, moving from the forced-choice task through descriptive comparisons to inferential, regression and mediation analyses. Section 4.3 reports the qualitative thematic analysis. Section 4.4 reports the content analysis of the stimulus profiles. Section 4.5 draws the three strands together.
All respondent figures in this section are reported as percentages of the analysed sample. Percentages may not sum to exactly 100 owing to rounding.
The analysed sample was predominantly female, though not overwhelmingly so, and heavily concentrated in the two youngest adult age bands — a distribution consistent with the platform's known user base rather than with the general population.
Gender distribution was as follows: 52.4% of respondents identified as female, 41.8% as male, 4.2% as non-binary, and 1.6% preferred not to state a gender.
The age profile was concentrated in the 25–34 band, which accounted for 38.2% of respondents, followed by the 18–24 band at 31.7%. Respondents aged 35–44 constituted 18.9%, those aged 45–54 constituted 8.1%, and those aged 55 and above constituted 3.1%. Nearly seven respondents in ten were therefore under thirty-five.
| Characteristic | Category | Share |
|---|---|---|
| Gender | Female | 52.4% |
| Male | 41.8% | |
| Non-binary | 4.2% | |
| Prefer not to say | 1.6% | |
| Age | 18–24 | 31.7% |
| 25–34 | 38.2% | |
| 35–44 | 18.9% | |
| 45–54 | 8.1% | |
| 55 and above | 3.1% | |
| Education | Secondary education | 12.3% |
| Bachelor's degree | 46.5% | |
| Master's degree | 33.6% | |
| Doctoral degree | 4.1% | |
| Other / vocational | 3.5% | |
| Employment | Employed full-time | 44.9% |
| Student | 27.4% | |
| Employed part-time | 11.3% | |
| Self-employed | 10.2% | |
| Unemployed / other | 6.2% | |
| Region | Europe | 44.1% |
| North America | 26.8% | |
| Asia | 18.3% | |
| Other regions | 10.8% |
Educational attainment was high: 46.5% held a bachelor's degree and a further 33.6% held a master's degree, with 4.1% holding a doctorate. Only 12.3% reported secondary education as their highest qualification and 3.5% reported vocational or other qualifications. This skew is characteristic of voluntary online research and is noted as a limitation in Section 6.4, since higher education is generally associated with greater scepticism towards marketing cues — meaning that the effects reported below may, if anything, be conservative.
In employment terms, 44.9% were in full-time employment, 27.4% were students, 11.3% were in part-time employment, 10.2% were self-employed and 6.2% were unemployed or in another category. Geographically, 44.1% were based in Europe, 26.8% in North America, 18.3% in Asia and 10.8% in other regions.
Usage intensity was high, as the weekly-use screening criterion made inevitable, but the concentration at the top of the distribution is nonetheless notable: 61.3% reported opening Instagram several times per day, 21.5% about once per day, 12.4% a few times per week and 4.8% weekly or less.
Daily time on the platform clustered in the middle of the range: 33.2% reported spending between thirty and sixty minutes per day, and 31.9% between one and two hours, together accounting for almost two-thirds of respondents. A further 14.7% reported under thirty minutes, 13.6% between two and three hours, and 6.6% more than three hours.
| Measure | Category | Share |
|---|---|---|
| Frequency of use | Several times per day | 61.3% |
| About once per day | 21.5% | |
| A few times per week | 12.4% | |
| Weekly or less | 4.8% | |
| Daily time on platform | Under 30 minutes | 14.7% |
| 30–60 minutes | 33.2% | |
| 1–2 hours | 31.9% | |
| 2–3 hours | 13.6% | |
| Over 3 hours | 6.6% | |
| Follows brand accounts | Yes, many | 38.7% |
| Yes, a few | 47.1% | |
| No | 14.2% | |
| Prior purchase via Instagram | Yes | 68.4% |
| No | 31.6% | |
| Checks follower count | Always or often | 41.6% |
| Sometimes | 37.3% | |
| Rarely or never | 21.1% |
The final row of Table 4.2 deserves emphasis, because it sets up the central tension of the findings. Only 41.6% of respondents said that they always or often check a brand's follower count before deciding whether to trust it. Yet, as Section 4.2.1 demonstrates, follower count predicted their choices with near-total consistency. What people report attending to and what actually drives their judgement are, in this domain, substantially different things.
Eight paired comparisons were administered, two within each niche, each pitting a high-follower profile against a low-follower profile from the same category. The result is the clearest single finding of the study: in every one of the eight pairings, the profile selected by the majority of respondents was the high-follower profile. Not one low-follower profile secured a majority in any comparison, in any niche.
| Pair | Niche | High-follower profile | Low-follower profile | Chose high | Chose low | χ² | p |
|---|---|---|---|---|---|---|---|
| P1 | Beauty | BH1 (4.8 M) | BL1 (18.4 K) | 88.9% | 11.1% | 228.4 | < .001 |
| P2 | Beauty | BH2 (2.6 M) | BL2 (7.9 K) | 85.7% | 14.3% | 191.6 | < .001 |
| P3 | Clothing | CH1 (6.1 M) | CL1 (22.7 K) | 86.2% | 13.8% | 197.3 | < .001 |
| P4 | Clothing | CH2 (3.3 M) | CL2 (11.2 K) | 83.0% | 17.0% | 163.8 | < .001 |
| P5 | Food | FH1 (3.9 M) | FL1 (15.6 K) | 81.4% | 18.6% | 148.2 | < .001 |
| P6 | Food | FH2 (1.8 M) | FL2 (6.4 K) | 78.2% | 21.8% | 125.7 | < .001 |
| P7 | Informational | IH1 (8.2 M) | IL1 (26.1 K) | 92.6% | 7.4% | 281.9 | < .001 |
| P8 | Informational | IH2 (4.4 M) | IL2 (9.8 K) | 89.8% | 10.2% | 239.1 | < .001 |
| Total | All | — | — | 85.7% | 14.3% | — | < .001 |
Note. Chi-square goodness-of-fit tests against an expected 50/50 distribution, df = 1 in each case.
Aggregated to niche level, 87.3% of beauty selections, 84.6% of clothing selections, 79.8% of food selections and 91.2% of informational selections went to the high-follower profile, for an overall figure of 85.7% against 14.3%. Chi-square goodness-of-fit tests against an expected even split were significant at p < .001 for every individual pairing.
Two features of the distribution merit comment. First, the informational niche produced the most extreme preference, with 91.2% of selections favouring the large account. This is consistent with the theoretical expectation that scale is read as institutional authority where the judgement concerns credibility of information rather than quality of a product. Second, the food niche produced the weakest preference at 79.8% — still an overwhelming majority, but a full eleven-and-a-half percentage points below the informational figure. The interview data, reported in Section 4.3, attribute this to a competing cultural schema in which small food businesses are read as artisanal rather than as marginal.
Every respondent rated all sixteen profiles on the five-item trust scale. Aggregated to tier level, high-follower profiles achieved a mean trust score of 5.55 on the seven-point scale against 4.08 for low-follower profiles — a difference of 1.47 scale points. The high-tier mean sits well above the scale midpoint of 4.0; the low-tier mean sits barely above it, indicating that low-follower profiles were not actively distrusted so much as regarded with indifference.
The pattern held in every niche without exception. In beauty, high-follower profiles scored 5.62 against 4.11 for low-follower profiles, a gap of 1.51. In clothing, 5.48 against 4.03, a gap of 1.45. In food, 5.31 against 4.29, a gap of 1.02 — the narrowest of the four. In the informational niche, 5.79 against 3.87, a gap of 1.92 — the widest, and the only case in which the low-tier mean fell below the scale midpoint. Low-follower news and media accounts were, in other words, the only stimulus category that respondents actively distrusted rather than merely discounted.
The same structure appears in the intention data, at a slightly lower absolute level. Across all profiles, high-follower accounts elicited a mean intention score of 5.21 against 3.76 for low-follower accounts, a difference of 1.45 scale points. Notably, the low-tier mean falls below the scale midpoint in three of the four niches, indicating net disinclination rather than mere indifference.
Niche-level figures were: beauty 5.14 against 3.72; clothing 5.29 against 3.81; food 4.98 against 3.94; informational 5.41 against 3.58. Food again shows the smallest gap at 1.04 points, and the informational niche the largest at 1.83 points, reproducing the ordering observed in the trust data. The consistency of this ordering across two independently measured constructs is itself evidence that the niche effect is systematic rather than incidental.
| Niche | Tier | Trust M | Trust SD | Intention M | Intention SD | Trust Δ | Intention Δ |
|---|---|---|---|---|---|---|---|
| Beauty | High | 5.62 | 0.94 | 5.14 | 1.08 | 1.51 | 1.42 |
| Beauty | Low | 4.11 | 1.12 | 3.72 | 1.21 | ||
| Clothing | High | 5.48 | 0.98 | 5.29 | 1.05 | 1.45 | 1.48 |
| Clothing | Low | 4.03 | 1.17 | 3.81 | 1.19 | ||
| Food | High | 5.31 | 1.03 | 4.98 | 1.14 | 1.02 | 1.04 |
| Food | Low | 4.29 | 1.09 | 3.94 | 1.16 | ||
| Informational | High | 5.79 | 0.89 | 5.41 | 1.02 | 1.92 | 1.83 |
| Informational | Low | 3.87 | 1.21 | 3.58 | 1.28 | ||
| Overall | High | 5.55 | 0.97 | 5.21 | 1.07 | 1.47 | 1.45 |
| Overall | Low | 4.08 | 1.15 | 3.76 | 1.21 |
Paired-samples t-tests were conducted on the within-respondent difference between high-tier and low-tier mean scores. Both were significant at p < .001 with large effect sizes by Cohen's (1988) conventions.
| Comparison | Mean difference | 95% CI | t | p | Cohen's d |
|---|---|---|---|---|---|
| Perceived brand trust | 1.47 | [1.36, 1.58] | 24.87 | < .001 | 1.38 |
| Purchase / engagement intention | 1.45 | [1.32, 1.58] | 21.43 | < .001 | 1.26 |
| Trust — beauty only | 1.51 | [1.36, 1.66] | 19.82 | < .001 | 1.24 |
| Trust — clothing only | 1.45 | [1.29, 1.61] | 17.94 | < .001 | 1.16 |
| Trust — food only | 1.02 | [0.87, 1.17] | 13.41 | < .001 | 0.87 |
| Trust — informational only | 1.92 | [1.75, 2.09] | 22.16 | < .001 | 1.47 |
A two-way mixed ANOVA with follower tier and niche as within-subjects factors confirmed a large main effect of tier on trust, F(1) = 618.42, p < .001, partial η² = .402, together with a smaller but significant main effect of niche, F(3) = 21.68, p < .001, partial η² = .045, and a significant tier × niche interaction, F(3) = 8.42, p < .001, partial η² = .018. Mauchly's test indicated a violation of sphericity for the niche factor, and Greenhouse–Geisser corrected degrees of freedom are reported. The interaction term provides formal support for H5: the magnitude of the follower-count advantage is not constant across commercial domains.
Bonferroni-corrected pairwise comparisons of the tier effect across niches showed that the informational niche differed significantly from the food niche (p < .001) and from clothing (p = .003), while beauty and clothing did not differ from one another (p = .412). The ordering is therefore best summarised as informational > beauty ≈ clothing > food.
Treating follower count as a continuous variable through a base-ten logarithmic transformation — appropriate given the extreme positive skew of raw counts — Pearson correlations were computed across all respondent–profile observations.
Logged follower count correlated with perceived brand trust at r = .612 (p < .001) and with purchase or engagement intention at r = .547 (p < .001). Perceived trust and intention correlated at r = .694 (p < .001). Instagram use intensity showed only weak associations with the outcome variables, at r = .164 with trust and r = .203 with intention, and was essentially uncorrelated with the follower-count variable, as the experimental design requires.
| Variable | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| 1. Follower count (log₁₀) | — | |||
| 2. Perceived brand trust | .612*** | — | ||
| 3. Purchase / engagement intention | .547*** | .694*** | — | |
| 4. Instagram use intensity | .118* | .164** | .203** | — |
Note. * p < .05, ** p < .01, *** p < .001.
The relative magnitude of the first two coefficients is theoretically informative. Follower count is more strongly associated with trust (.612) than with intention (.547), while trust is more strongly associated with intention (.694) than follower count is with either. This pattern of coefficients is precisely what a mediated structure predicts, and it motivates the formal test reported in Section 4.2.7.
Purchase or engagement intention was regressed on three blocks of predictors: demographic and usage controls, then follower tier, then perceived brand trust. Assumption checks were satisfactory: the maximum variance inflation factor was 1.68, well below the threshold of concern; the Durbin–Watson statistic was 1.94; residual plots showed no systematic heteroscedasticity; and standardised residuals were approximately normally distributed.
| Predictor block | β | SE | t | p | R² | ΔR² |
|---|---|---|---|---|---|---|
| Step 1 — controls | .081 | .081*** | ||||
| Age | −.128 | .031 | −4.13 | < .001 | ||
| Gender (female = 1) | .042 | .029 | 1.45 | .148 | ||
| Instagram use intensity | .187 | .030 | 6.23 | < .001 | ||
| Prior purchase via Instagram | .164 | .030 | 5.47 | < .001 | ||
| Step 2 — follower tier added | .307 | .226*** | ||||
| Follower tier (high = 1) | .489 | .027 | 18.11 | < .001 | ||
| Step 3 — trust added | .512 | .205*** | ||||
| Follower tier (high = 1) | .212 | .026 | 8.15 | < .001 | ||
| Perceived brand trust | .548 | .026 | 21.08 | < .001 |
Note. β = standardised regression coefficient. *** p < .001 for F change.
Three results follow. First, the control block alone explains 8.1% of variance, with age negatively and usage intensity positively associated with intention. Second, adding follower tier raises explained variance to 30.7% — an increment of 22.6 percentage points, which is to say that follower tier alone accounts for nearly three times as much variance in intention as all demographic and usage controls combined. Third, adding perceived trust raises explained variance to 51.2% while reducing the standardised coefficient on follower tier from .489 to .212. The tier coefficient remains significant, which rules out complete mediation, but its magnitude falls by more than half.
A bootstrapped mediation analysis was conducted following Hayes (2018), Model 4, with 5,000 bootstrap resamples and bias-corrected accelerated confidence intervals. Follower tier served as the independent variable, perceived brand trust as the mediator and purchase or engagement intention as the outcome.
| Path | Description | β | 95% BCa CI | p |
|---|---|---|---|---|
| a | Follower tier → trust | .612 | [.571, .653] | < .001 |
| b | Trust → intention | .548 | [.503, .593] | < .001 |
| c | Total effect: tier → intention | .547 | [.504, .590] | < .001 |
| c′ | Direct effect: tier → intention | .212 | [.161, .263] | < .001 |
| ab | Indirect effect via trust | .335 | [.281, .392] | < .001 |
The indirect effect is significant, with a bias-corrected confidence interval that does not include zero. The direct effect also remains significant, establishing partial rather than full mediation. The proportion of the total effect carried by the indirect path is .335 ÷ .547 = 61.2%.
This decomposition is the analytical core of the report. Roughly three-fifths of the influence of follower count on purchase intention operates through the elevation of perceived trustworthiness — consumers see a large audience, infer that the brand is reliable, and become more willing to buy. The remaining two-fifths operates independently of trust. Section 4.3 identifies what that residual path consists of: bandwagon motivation, identity signalling, and the simple convenience of choosing the option that requires no further deliberation.
The magnitude of the follower-tier effect varied systematically across respondent subgroups. The clearest gradient was by age: the trust gap between tiers declined monotonically from 1.71 scale points among respondents aged 18–24 to 1.02 points among those aged 55 and above, with the intention gap following an almost identical trajectory from 1.66 to 0.94.
A moderation analysis with age band entered as a continuous moderator confirmed a significant interaction with follower tier in predicting trust, β = −.147, p < .001. Younger respondents, who are the heaviest users of the platform, are also the most responsive to its most prominent numerical cue — a finding that runs contrary to the intuition that digital natives are more sophisticated readers of platform metrics, and closer to the position that habituation to an interface increases rather than decreases reliance on its default cues.
| Subgroup | Category | Trust Δ | Intention Δ | Chose high (%) |
|---|---|---|---|---|
| Age | 18–24 | 1.71 | 1.66 | 89.4% |
| 25–34 | 1.58 | 1.52 | 87.1% | |
| 35–44 | 1.39 | 1.36 | 84.6% | |
| 45–54 | 1.14 | 1.09 | 80.2% | |
| 55 and above | 1.02 | 0.94 | 77.8% | |
| Gender | Female | 1.52 | 1.49 | 86.3% |
| Male | 1.41 | 1.40 | 84.9% | |
| Non-binary | 1.44 | 1.42 | 85.1% | |
| Use intensity | Several times per day | 1.58 | 1.54 | 87.6% |
| Once daily or less | 1.24 | 1.19 | 81.3% | |
| Prior IG purchase | Yes | 1.53 | 1.57 | 87.2% |
| No | 1.34 | 1.21 | 82.4% | |
| Follower scepticism | High (agree metric is unreliable) | 1.31 | 1.28 | 82.1% |
| Low (disagree) | 1.57 | 1.55 | 87.9% |
Gender produced no meaningful variation; a one-way ANOVA on the trust gap across gender categories was non-significant, F(2) = 1.87, p = .155. Use intensity did matter: respondents who opened the platform several times daily showed a trust gap of 1.58 against 1.24 for lighter users. Prior purchase experience through Instagram was associated with a larger intention gap, 1.57 against 1.21, suggesting that consumers who have already transacted on the platform are more, not less, reliant on its native credibility cues.
The final quantitative analysis addresses H6, and produces the study's most theoretically awkward result. The scepticism scale carried three items, and they were endorsed at different rates. 58.7% of respondents agreed or strongly agreed that follower counts can be bought, and 49.2% agreed that many brands on Instagram have followers that are not real. A smaller share — 38.2% — agreed with the evaluative item that a high follower count does not mean a brand is trustworthy, with 27.4% neutral and 34.4% disagreeing. In other words, a clear majority knows the metric can be purchased; rather fewer are willing to say it therefore carries no weight.
If this belief functioned as consumers report it functioning, high-scepticism respondents should have selected high-follower profiles at a rate close to chance. They did not. Among high-scepticism respondents, 82.1% still selected the high-follower profile, against 87.9% among low-scepticism respondents. The difference is statistically significant, χ²(1) = 6.14, p = .013, but the effect size is negligible: Cramér's V = .11. Scepticism reduced the trust gap from 1.57 to 1.31 scale points and the intention gap from 1.55 to 1.28.
H6 is therefore supported in direction but not in magnitude. Explicit disbelief in the validity of the metric attenuates its influence by roughly a sixth. It does not come close to neutralising it. Four respondents in five who say the number is meaningless nonetheless choose the brand that has more of it.
Interview transcripts were analysed through the six-phase reflexive thematic analysis procedure of Braun and Clarke (2006): familiarisation, generation of initial codes, searching for themes, reviewing themes, defining and naming themes, and reporting. Initial open coding generated 147 distinct codes, which were collapsed through two rounds of axial grouping into 19 sub-themes and finally into six themes. A second analyst independently coded a subset of transcripts; agreement at theme level was high, and the two disagreements concerned boundary cases between Themes 3 and 4, which were resolved by tightening the definition of Theme 4 to require explicit reference to a ratio or comparison rather than to engagement in general.
The prevalence of each theme is reported as the share of interviewees whose transcripts contained at least one instance coded to that theme. Prevalence is reported for descriptive purposes only and is not intended as a quantitative claim.
| Theme | Sub-themes | Prevalence |
|---|---|---|
| 1. Followers as a cognitive shortcut | Speed of judgement; absence of alternatives; 'first thing you see' | 94.7% |
| 2. Scale as risk reduction | Recourse if something goes wrong; assumed legitimacy; assumed logistics | 89.5% |
| 3. Scepticism about bought followers | Awareness of follower markets; suspicion of round numbers; resigned acceptance | 78.9% |
| 4. Engagement ratio as verification | Comments-to-followers checks; suspicion of low comment counts; effort cost | 68.4% |
| 5. Niche-conditional relevance | Small is artisanal (food); small is amateur (news); scale as authority | 63.2% |
| 6. Bandwagon and identity signalling | Wanting to be part of something; shareability; fear of appearing naive | 57.9% |
Theme 1: Followers as a cognitive shortcut (94.7%). Almost every participant described the follower count as a rapid substitute for evaluation they had neither the time nor the means to perform. The recurring framing was one of triage rather than judgement: participants were not claiming that follower count is a good measure, but that it is the measure that is available in the two or three seconds they allocate to an unfamiliar profile. One participant (P07) described scanning the number before the profile picture had finished loading. Another (P14) observed that checking anything else would require leaving the profile, and that leaving the profile is effectively the same as abandoning the brand. This theme provides the mechanism for the peripheral-route prediction of Section 2.2.3.
Theme 2: Scale as risk reduction (89.5%). Participants consistently translated audience size into an inference about operational and legal accountability. The reasoning was explicit and structurally similar across transcripts: a brand with millions of followers has too much to lose to defraud an individual customer; it presumably has a functioning returns process; it presumably ships when it says it will; and if it does not, there is a public forum in which complaint is visible. P03 framed this as the difference between a business and a person, arguing that a small account might simply stop replying whereas a large one cannot afford to. This theme maps directly onto the trust-mediation pathway quantified in Section 4.2.7, and specifically onto the ability and integrity components of the Mayer et al. (1995) trust decomposition.
Theme 3: Scepticism about bought followers (78.9%). Nearly four participants in five raised, unprompted, the possibility that a follower count had been purchased. Several described the mechanics accurately and one had investigated pricing out of curiosity. What is striking is the disposition that accompanied this knowledge. Rather than producing systematic discounting, awareness produced a kind of resigned accommodation: participants acknowledged that the number might be inflated, then continued to use it. P11, when shown their own survey responses, described their choices as embarrassing but reasonable, on the grounds that a possibly-inflated large number is still better evidence than a definitely-small one. This is the qualitative correlate of the scepticism paradox reported in Section 4.2.9.
Theme 4: Engagement ratio as verification (68.4%). Roughly two-thirds of participants described a secondary check — comparing visible likes or comments against the follower total — as their method for detecting inauthentic accounts. Crucially, almost all of them described this as something they do occasionally, for high-stakes purchases, or 'if something feels off', rather than as routine practice. The verification behaviour that participants report as their safeguard is thus reserved for precisely the situations in which they are already suspicious, and is not applied in the ordinary case. The content analysis in Section 4.4 shows that had they applied it, the ranking would have inverted.
Theme 5: Niche-conditional relevance (63.2%). Participants articulated clearly different schemas for different categories. In food, a small account was frequently read as a positive signal — independent, local, hand-made, not yet discovered. In beauty, small was read as risky, with participants citing concerns about ingredient safety, regulatory compliance and dermatological reaction. In clothing, small was read as ambiguous: potentially a boutique, potentially a drop-shipping operation. In the informational field, small was read almost uniformly as a deficiency, with participants associating audience scale with editorial resource, fact-checking capacity and institutional standing. P02 put it directly: a small restaurant is a discovery, but a small news account is someone's opinion. This theme explains the niche ordering observed in Sections 4.2.1 to 4.2.3.
Theme 6: Bandwagon and identity signalling (57.9%). The least universal but most theoretically interesting theme concerns motives that do not pass through trust at all. Participants described wanting to be associated with brands that others recognise, anticipating how a purchase would read if shared, and — in several cases — an aversion to appearing to have been taken in by something obscure. P19 described the low-follower options as feeling like a risk of looking foolish rather than a risk of losing money. This theme corresponds to the residual direct path in the mediation model, and suggests that the two-fifths of the effect not carried by trust is substantially social rather than epistemic.
The sixteen stimulus profiles were coded independently on eleven attributes. The purpose of this strand was diagnostic: if high-follower profiles differ systematically from low-follower profiles on observable quality-relevant characteristics, then the survey effect might be attributable to those characteristics rather than to follower count as such. The mean follower count in the high tier was 4.39 million against 14,760 in the low tier, a ratio of approximately 297 to 1.
| Coded attribute | High tier | Low tier | Difference |
|---|---|---|---|
| Verified badge present | 100% | 0% | +100 pp |
| Complete bio (category, description, contact) | 100% | 62.5% | +37.5 pp |
| External link in bio | 100% | 75.0% | +25.0 pp |
| Story highlights present | 100% | 50.0% | +50.0 pp |
| Mean number of highlights | 8.6 | 2.4 | +6.2 |
| Mean posts per week | 5.4 | 2.7 | +2.7 |
| Mean caption length (words) | 42 | 68 | −26 |
| Branded hashtag in use | 100% | 37.5% | +62.5 pp |
| User-generated content reposted | 62.5% | 37.5% | +25.0 pp |
| Brand replies to comments | 37.5% | 87.5% | −50.0 pp |
| Photographic production quality (1–5) | 4.7 | 3.1 | +1.6 |
| Mean engagement rate | 1.08% | 4.36% | −3.28 pp |
Note. pp = percentage points. Engagement rate computed as (likes + comments) ÷ followers, averaged across the twelve most recent non-promotional grid posts.
High-follower profiles were more professionalised on eight of the eleven coded attributes. All were verified, all had complete bios, all used story highlights and branded hashtags, all posted more frequently, and all scored higher on photographic production quality. This confirms that follower count does not vary in isolation in naturalistic material — a point of considerable importance for interpretation, discussed in Section 5.1.
On three attributes, however, the low tier performed better. Low-follower brands wrote substantially longer captions, averaging 68 words against 42. They replied to comments far more often, with 87.5% doing so against 37.5% in the high tier. And, most consequentially, they achieved mean engagement rates roughly four times higher: 4.36% against 1.08%. Across the sixteen profiles, logged follower count and engagement rate correlated at r = −.874 (p < .001).
The comparison between Sections 4.3 and 4.4 produces the sharpest finding in the study. In interviews, 68.4% of participants named the engagement-to-follower ratio as their method for distinguishing genuine accounts from inflated ones. In the content analysis, that ratio favours the low-follower tier by a factor of four. Had respondents actually applied the verification heuristic they described, the forced-choice results would have been reversed. Instead, 85.7% chose the tier with the lower engagement rate.
Three interpretations are possible and are not mutually exclusive. The first is that engagement rate is simply not visible enough: it requires the consumer to notice a like count, notice a follower count, and perform a mental division, none of which the interface facilitates. The second is that respondents lacked the reference distribution needed to interpret the ratio; without knowing that 1% is low and 4% is high, the arithmetic is uninformative. The third, supported by Theme 4, is that verification is triggered by suspicion rather than performed by default, and nothing about a large, professionally presented account triggers suspicion.
A secondary comparative observation concerns comment responsiveness. Low-follower brands replied to comments at more than twice the rate of high-follower brands — an attribute that maps directly onto the benevolence dimension of trust in the Mayer et al. (1995) framework. Respondents did not reward it, most likely because it is not observable from the profile grid without opening individual posts.
The content analysis establishes both a limit and a strengthening of the study's central claim. The limit is that follower tier is confounded in naturalistic material with professionalisation, and the survey therefore cannot fully isolate the number from the presentation quality that typically accompanies it. Respondents may have been responding in part to production values rather than to the count itself.
The strengthening is more substantial. The single most diagnostic quality indicator available on the platform — engagement rate — pointed decisively in the opposite direction to respondents' choices, and it was the indicator that respondents themselves nominated as their safeguard. A judgement process that follows the visible, easily faked, unreliable number while ignoring the computable, harder-to-fake, more reliable ratio is not a judgement process that is tracking quality. It is one that is tracking salience.
Taken together, the three strands support a coherent account. Follower count operates as a highly salient peripheral cue in a low-elaboration environment. It activates a social-proof inference which elevates perceived brand trust, chiefly through inferences about accountability and operational reliability rather than about product quality. Elevated trust translates into stronger purchase and engagement intention, and carries approximately three-fifths of the total effect. A residual direct path, comprising bandwagon motivation and identity signalling, carries the remaining two-fifths.
The strength of the mechanism is conditioned by the interpretive schema the consumer brings to the category. Where smallness has a positive cultural coding, as in food, the effect weakens; where scale is equated with institutional authority, as in news and media, it intensifies. It is conditioned further by age and by platform-usage intensity, in both cases in the direction of greater susceptibility among the most habituated users.
The mechanism is not, however, meaningfully conditioned by explicit disbelief. This is the finding that most resists comfortable interpretation. Consumers who articulate an accurate critique of the metric behave almost identically to those who do not. Whatever process is generating the observed preference, it is not one that stated belief has much purchase on — which is consistent with heuristic processing operating below the level at which declarative knowledge intervenes.
| Hypothesis | Key evidence | Outcome |
|---|---|---|
| H1 Follower count → trust | M = 5.55 vs 4.08; t = 24.87, p < .001, d = 1.38; r = .612 | Supported |
| H2 Follower count → intention | M = 5.21 vs 3.76; t = 21.43, p < .001, d = 1.26; r = .547 | Supported |
| H3 Trust → intention | r = .694, p < .001; β = .548 in Step 3 | Supported |
| H4 Trust mediates | Indirect ab = .335, CI [.281, .392]; c′ remains significant | Partially supported |
| H5 Niche moderates | Tier × niche F(3) = 8.42, p < .001, ηp² = .018 | Supported |
| H6 Scepticism attenuates | 82.1% vs 87.9% choice; V = .11; Δ falls 1.57 → 1.31 | Partially supported |
The central empirical result of this study is easily stated: across four commercial and informational niches, and across eight independent paired comparisons, respondents preferred the high-follower brand in every case, trusted it substantially more, and expressed markedly stronger intention to buy from or engage with it. The magnitude is not marginal. A gap of 1.47 scale points on a seven-point trust measure, with a Cohen's d of 1.38, is a large effect by any conventional standard, and it was obtained under conditions — anonymised brands, standardised presentation, randomised order — designed to strip away the correlates that would ordinarily inflate it.
The mediation decomposition specifies the mechanism. Approximately 61% of the effect of follower count on intention runs through perceived trust. The interview data give this pathway concrete content: participants did not report inferring that a large brand makes better products. They reported inferring that a large brand is accountable — that it can be found, complained about, held to a returns policy, and damaged reputationally if it behaves badly. In the vocabulary of Mayer, Davis and Schoorman (1995), the inference concerns integrity and ability rather than benevolence. This is a sensible inference to draw from scale, and it is worth noting that it is not obviously irrational. A brand with a large public audience genuinely does face higher reputational costs for misconduct. The problem is not that the inference is invalid; it is that the premise — that the audience is real — is unverified.
The residual 39%, running directly from follower count to intention without passing through trust, is illuminated by Theme 6. Participants described wanting to be associated with recognised brands, anticipating how a choice would be perceived if shared, and fearing the appearance of naivety. These are social rather than epistemic motives. They do not concern whether the brand is good; they concern what choosing it says about the chooser. Dual-process models accommodate this readily, since peripheral cues need not operate through belief at all.
The niche moderation is theoretically orderly. The effect was largest in the informational field, where the judgement concerns credibility of information and where scale is culturally coded as institutional resource — editorial staff, fact-checking, correspondents. It was smallest in food, the only niche in which a competing positive schema attaches to smallness: the independent, the local, the artisanal. Beauty and clothing occupied an indistinguishable middle position, consistent with their intermediate levels of perceived purchase risk. This ordering maps neatly onto Cialdini's (2009) condition that social proof exerts maximum force under maximum uncertainty, provided one recognises that uncertainty here is not only about product quality but about which interpretive schema applies.
The confound identified in Section 4.4 requires honest treatment. High-follower profiles were more professionalised on eight of eleven coded attributes, and this covariation is a genuine feature of naturalistic material rather than an artefact. It is therefore not possible to claim that the number alone, stripped of everything that accompanies it, produced the observed effect. What can be claimed is stronger in a different way: the one available indicator that is genuinely diagnostic of audience authenticity — engagement rate — favoured the low tier by a factor of four, and was ignored. Respondents were not following the best available quality signal. They were following the most prominent one.
The findings both confirm and extend the existing literature in specific ways.
De Veirman, Cauberghe and Hudders (2017) found that higher follower counts increased perceived likeability of individual influencers, moderated by product divergence. The present study replicates the directional finding in an adjacent context — brand accounts rather than personal accounts — and demonstrates that the effect survives the removal of the parasocial mechanism. This matters theoretically. If follower count worked chiefly through the perception of an appealing person, it should not transfer to a commercial entity with no personal presence. That it does transfer, and at large effect size, supports a pure social-proof or signalling interpretation rather than a relational one.
Djafarova and Rushworth (2017) reported qualitatively that follower numbers featured prominently in credibility judgements. The interview strand here reproduces that finding and adds a complication they did not report: participants who name follower count as a criterion also, in the same interview, name its unreliability. The coexistence of the two is the phenomenon requiring explanation.
Lou and Yuan (2019) found that influencer credibility rather than follower count directly predicted trust in branded content. The present findings are compatible with this if credibility is understood as the mediator rather than as a competitor to follower count — which is precisely what the mediation model estimates. Where Lou and Yuan's design placed the two variables in competition, the design used here places them in sequence, and the sequential specification fits the data well.
Casaló, Flavián and Ibáñez-Sánchez (2020) concluded that content originality and uniqueness outweighed follower count as antecedents of opinion leadership once content characteristics were controlled. This appears to conflict with the present findings, but the conflict may be one of measurement window rather than of substance. Their design permitted evaluation of content; the forced-choice task used here permits only evaluation of a profile header, which is the situation an unfamiliar visitor actually faces. Content quality may well dominate over a longer exposure. Follower count dominates in the first three seconds, and in a scrolling feed the first three seconds are frequently all there is.
The micro-influencer literature, which holds that smaller accounts outperform on engagement and relationship depth, is directly corroborated by the content analysis: low-tier profiles achieved engagement rates roughly four times higher and replied to comments at more than twice the rate. What the present study adds is that this genuine advantage does not translate into first-impression trust, because it is not visible at the moment the judgement is made. The micro-influencer thesis is a claim about performance among an audience already acquired. It is not a claim about acquisition, and the evidence here suggests it should not be read as one.
Finally, the scepticism paradox connects to a broader literature on the limits of media literacy as a corrective. The result reported in Section 4.2.9 — an attenuation of roughly one-sixth from explicit, accurate disbelief — is consistent with dual-process accounts in which declarative knowledge fails to intervene in heuristic processing under low motivation, and with Tversky and Kahneman's (1974) observation that knowing about a bias does not reliably prevent its operation.
The study identifies a class of signal that classical signalling theory does not accommodate well. Spence's (1973) framework requires that a signal be differentially costly across types in order to be informative. Follower count fails this requirement, since the cost of acquisition is largely uncoupled from underlying quality. Standard theory would therefore predict that rational consumers discount it to irrelevance. They do not. The observed behaviour is better described as heuristic response to a salient interface element than as inference from a credible signal, which suggests that signalling theory in digital contexts requires an explicit account of signal salience alongside signal cost — a variable that mattered little when signals were things like factory investment or warranty length, and matters a great deal when signals are numbers rendered in bold type at the top of a screen.
The finding that mediation is partial rather than complete also has theoretical consequences. Models that treat trust as the sole conduit between platform cues and behavioural intention are under-specified. A substantial minority of the effect bypasses trust entirely and appears to be driven by social-identity considerations. Future modelling in this area should specify at least two parallel paths.
If a metric that is cheap to falsify exerts a large effect on consumer trust, the platform that displays it prominently bears some responsibility for the resulting distortion. Three interventions follow from the findings. First, displaying engagement rate alongside follower count would surface the one indicator that actually discriminates between the tiers — and would do so at the moment of judgement rather than requiring the consumer to compute it. Second, audience-authenticity indicators, analogous to the audited circulation figures that print media adopted a century ago, would restore the cost asymmetry that makes a signal informative. Third, reducing the visual prominence of raw counts relative to other profile information would lower the salience that Section 5.3.1 identifies as the operative variable.
From a consumer-protection standpoint, the scepticism paradox is the most policy-relevant finding. It indicates that awareness campaigns and media-literacy education, which operate on declarative belief, should not be expected to correct this particular distortion. Interventions must change what the interface shows, not what the consumer knows.
This study set out to determine whether Instagram follower count is statistically related to perceived brand trust and purchase intention, and to identify the pathway through which any such relationship operates. Three strands of evidence were collected and integrated.
Four conclusions are drawn.
First, high follower count affects perceived brand trust. The relationship is positive, large, statistically robust and present in every niche examined. Consumers evaluating an unfamiliar Instagram brand treat audience size as evidence of trustworthiness, and the difference between a profile with millions of followers and one with thousands is worth approximately one and a half points on a seven-point trust scale — the difference between an account that is actively trusted and one that is regarded with indifference.
Second, high follower count affects purchase intention, and it does so principally through trust. The total effect on intention is large, and roughly three-fifths of it is transmitted by elevated trust. The remaining two-fifths operates directly, driven by bandwagon and identity-signalling motives that do not require the consumer to believe anything about the brand's reliability at all. Both paths point in the same direction, and their combination is what produces the near-unanimous forced-choice outcome.
Third, the effect is systematically conditioned but never reversed. Commercial niche, respondent age, usage intensity and prior transactional experience all modulate the magnitude. None of them modulate the direction. Even in the food niche, where a competing artisanal schema is available and actively articulated by participants, nearly four selections in five still went to the large account. There is no subgroup, and no category, in which small audiences enjoyed an advantage.
Fourth, knowing better does not help much. Explicit, accurate scepticism about the validity of follower counts reduced the effect by about a sixth and left the overwhelming preference intact. This is the finding with the widest implications, because it indicates that the distortion is not a knowledge deficit and cannot be corrected by informing people. The metric works not because consumers believe it is meaningful but because it is the most prominent thing on the screen at the moment a judgement is required.
Taken together, these results establish that the number displayed at the top of an Instagram brand profile is doing substantial commercial work — allocating trust, shaping intention, and determining which brands are considered at all — on the basis of a property that is only loosely connected to anything a consumer would, on reflection, care about.
Aaker, D. A. (1991). Managing brand equity: Capitalizing on the value of a brand name. New York: Free Press.
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211.
Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgments. In H. Guetzkow (Ed.), Groups, leadership and men (pp. 177–190). Pittsburgh: Carnegie Press.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51(6), 1173–1182.
Berger, J., & Milkman, K. L. (2012). What makes online content viral? Journal of Marketing Research, 49(2), 192–205.
Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77–101.
Casaló, L. V., Flavián, C., & Ibáñez-Sánchez, S. (2020). Influencers on Instagram: Antecedents and consequences of opinion leadership. Journal of Business Research, 117, 510–519.
Chaiken, S. (1980). Heuristic versus systematic information processing and the use of source versus message cues in persuasion. Journal of Personality and Social Psychology, 39(5), 752–766.
Chaudhuri, A., & Holbrook, M. B. (2001). The chain of effects from brand trust and brand affect to brand performance: The role of brand loyalty. Journal of Marketing, 65(2), 81–93.
Cialdini, R. B. (2009). Influence: Science and practice (5th ed.). Boston: Pearson.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.
Creswell, J. W., & Plano Clark, V. L. (2018). Designing and conducting mixed methods research (3rd ed.). Thousand Oaks, CA: Sage.
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340.
De Veirman, M., Cauberghe, V., & Hudders, L. (2017). Marketing through Instagram influencers: The impact of number of followers and product divergence on brand attitude. International Journal of Advertising, 36(5), 798–828.
Deutsch, M., & Gerard, H. B. (1955). A study of normative and informational social influences upon individual judgment. Journal of Abnormal and Social Psychology, 51(3), 629–636.
Djafarova, E., & Rushworth, C. (2017). Exploring the credibility of online celebrities' Instagram profiles in influencing the purchase decisions of young female users. Computers in Human Behavior, 68, 1–7.
Dodds, W. B., Monroe, K. B., & Grewal, D. (1991). Effects of price, brand, and store information on buyers' product evaluations. Journal of Marketing Research, 28(3), 307–319.
Erdem, T., & Swait, J. (1998). Brand equity as a signaling phenomenon. Journal of Consumer Psychology, 7(2), 131–157.
Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention and behavior: An introduction to theory and research. Reading, MA: Addison-Wesley.
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50.
Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2019). Multivariate data analysis (8th ed.). Andover: Cengage Learning.
Hayes, A. F. (2018). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach (2nd ed.). New York: Guilford Press.
Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication and persuasion: Psychological studies of opinion change. New Haven, CT: Yale University Press.
Jin, S. V., Muqaddam, A., & Ryu, E. (2019). Instafamous and social media influencer marketing. Marketing Intelligence & Planning, 37(5), 567–579.
Kaplan, A. M., & Haenlein, M. (2010). Users of the world, unite! The challenges and opportunities of social media. Business Horizons, 53(1), 59–68.
Keller, K. L. (1993). Conceptualizing, measuring, and managing customer-based brand equity. Journal of Marketing, 57(1), 1–22.
Ki, C.-W., & Kim, Y.-K. (2019). The mechanism by which social media influencers persuade consumers: The role of consumers' desire to mimic. Psychology & Marketing, 36(10), 905–922.
Kirmani, A., & Rao, A. R. (2000). No pain, no gain: A critical review of the literature on signaling unobservable product quality. Journal of Marketing, 64(2), 66–79.
Krippendorff, K. (2018). Content analysis: An introduction to its methodology (4th ed.). Thousand Oaks, CA: Sage.
Lou, C., & Yuan, S. (2019). Influencer marketing: How message value and credibility affect consumer trust of branded content on social media. Journal of Interactive Advertising, 19(1), 58–73.
Mayer, R. C., Davis, J. H., & Schoorman, F. D. (1995). An integrative model of organizational trust. Academy of Management Review, 20(3), 709–734.
Metzger, M. J., & Flanagin, A. J. (2013). Credibility and trust of information in online environments: The use of cognitive heuristics. Journal of Pragmatics, 59, 210–220.
Morgan, R. M., & Hunt, S. D. (1994). The commitment–trust theory of relationship marketing. Journal of Marketing, 58(3), 20–38.
Nunes, J. C., Ordanini, A., & Giambastiani, G. (2021). The concept of authenticity: What it means to consumers. Journal of Marketing, 85(4), 1–20.
Nunnally, J. C. (1978). Psychometric theory (2nd ed.). New York: McGraw-Hill.
Ohanian, R. (1990). Construction and validation of a scale to measure celebrity endorsers' perceived expertise, trustworthiness, and attractiveness. Journal of Advertising, 19(3), 39–52.
Petty, R. E., & Cacioppo, J. T. (1986). Communication and persuasion: Central and peripheral routes to attitude change. New York: Springer-Verlag.
Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879–903.
Spence, M. (1973). Job market signaling. Quarterly Journal of Economics, 87(3), 355–374.
Sundar, S. S. (2008). The MAIN model: A heuristic approach to understanding technology effects on credibility. In M. J. Metzger & A. J. Flanagin (Eds.), Digital media, youth, and credibility (pp. 73–100). Cambridge, MA: MIT Press.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.
Section A — Screening and demographics
Section B — Forced-choice paired comparisons
Respondents were shown eight pairs of profiles, two per niche, each pair comprising one high-follower and one low-follower account from the same category. Pair order and left–right position were randomised.
Instruction (transactional niches): 'If you needed a product in this category today and had to pick one of these two accounts to buy from, which would you choose?'
Instruction (informational niche): 'If you wanted to follow one of these two accounts for news, which would you choose?'
Response options: [Profile A] / [Profile B]. No neutral option was provided.
Section C — Profile evaluation scales
Each of the sixteen profiles was presented individually, in randomised order, followed by the items below on a seven-point scale (1 = strongly disagree, 7 = strongly agree).
Perceived brand trust (α = .912)
Purchase / engagement intention (α = .884)
Section D — Individual difference measures
Follower-count scepticism (α = .812)
Instagram use intensity (α = .796)
Section E — Suspicion probe and debriefing
Semi-structured format. Prompts were used flexibly; the order was adapted to the flow of each conversation. Interviews lasted between 38 and 62 minutes.
Domain 1 — General evaluation habits
Domain 2 — Reactions to specific stimulus profiles
Domain 3 — Explicit reasoning about follower counts
Domain 4 — Reflection on own survey choices
Closing
Table C.1. Content-analysis variables and operational definitions.
| Variable | Type | Operational definition | κ |
|---|---|---|---|
| Verified badge | Binary | Blue verification badge present adjacent to account name | 1.00 |
| Bio completeness | Binary | Bio contains category label, descriptive line and contact route | .93 |
| External link | Binary | At least one clickable link present in the bio | 1.00 |
| Story highlights | Binary | One or more saved highlight covers displayed | .96 |
| Highlight count | Count | Number of distinct highlight covers | .91 |
| Posting frequency | Count | Grid posts published in the 28 days preceding coding, ÷ 4 | .89 |
| Caption length | Count | Mean word count across the 12 most recent grid captions | .86 |
| Branded hashtag | Binary | A hashtag unique to the brand appears in ≥ 3 of 12 recent captions | .88 |
| User-generated content | Binary | ≥ 2 of 12 recent posts credit an external creator or customer | .84 |
| Comment responsiveness | Binary | Brand account replies in ≥ 3 of 12 recent posts' comment threads | .82 |
| Production quality | 1–5 scale | Composite rating of lighting, composition, colour consistency, retouching | .74 |
| Engagement rate | Continuous | Mean of (likes + comments) ÷ followers across 12 recent posts, as% | .97 |
Note. κ = Cohen's kappa between two independent coders. Overall κ across all variables = .87.
Table C.2. Scale item statistics for the two focal constructs (all profiles pooled).
| Item | M | SD | Item–total r | α if deleted |
|---|---|---|---|---|
| Trust C1 — seems trustworthy | 4.86 | 1.43 | .812 | .887 |
| Trust C2 — confident buying | 4.71 | 1.51 | .794 | .891 |
| Trust C3 — delivers what it promises | 4.79 | 1.46 | .836 | .882 |
| Trust C4 — professional and well run | 5.02 | 1.38 | .768 | .897 |
| Trust C5 — would not be let down (rev.) | 4.64 | 1.55 | .741 | .903 |
| Intention C6 — would consider buying | 4.53 | 1.62 | .781 | .841 |
| Intention C7 — likelihood of purchase high | 4.38 | 1.68 | .806 | .819 |
| Intention C8 — would recommend | 4.44 | 1.64 | .763 | .858 |
Table C.3. Distribution of responses to follower-count scepticism items.
| Item | Disagree (1–3) | Neutral (4) | Agree (5–7) |
|---|---|---|---|
| D1. Follower counts can be bought | 22.9% | 18.4% | 58.7% |
| D2. High count does not mean trustworthy | 34.4% | 27.4% | 38.2% |
| D3. Many brands have followers that are not real | 26.1% | 24.7% | 49.2% |
Note. Item D2 was used to classify respondents into high- and low-scepticism groups in Section 4.2.9. The gap between D1 (58.7% aware that counts can be bought) and D2 (38.2% willing to conclude that counts are therefore uninformative) is itself notable: a substantial group knows the metric can be faked yet declines to treat that knowledge as disqualifying.
Table C.4. Forced-choice consistency across the eight paired comparisons.
| Number of high-follower choices made | Share of respondents |
|---|---|
| 8 of 8 (fully consistent) | 58.3% |
| 7 of 8 | 19.6% |
| 6 of 8 | 11.2% |
| 5 of 8 | 5.7% |
| 4 of 8 (no preference) | 2.9% |
| 3 or fewer of 8 | 2.3% |
Note. Almost three respondents in five selected the high-follower profile in every single comparison; fewer than one in twenty selected the low-follower profile more often than not.
This research is free to cite and quote. Attribution to Kicksta Research with a link to this page is all we ask.
APA
Kicksta Research Team (2026). The Statistical Relationship Between Instagram Follower Count, Perceived Brand Trust and Purchase Intention: A Mixed-Methods Study Across Four Commercial Niches. Kicksta. https://kicksta.co/research/instagram-follower-count-brand-trust
BibTeX
@techreport{kicksta2026followers,
title = {The Statistical Relationship Between Instagram Follower Count,
Perceived Brand Trust and Purchase Intention},
author = {{Kicksta Research Team}},
year = {2026},
institution = {Kicksta},
url = {https://kicksta.co/research/instagram-follower-count-brand-trust}
}This study was designed and run by the Kicksta Research Department, a team established to study consumer psychology and digital consumer behaviour. It is staffed by trained researchers alongside postgraduate and undergraduate contributors.
Tako Zarandia served as Head of Research, overseeing the research process and coordinating its execution. She holds a Master’s degree in Strategic Communication and Public Relations and a Bachelor’s degree in Business Administration with a specialisation in Marketing. Her background spans human resources, marketing and business development, and she has previously contributed to clinical research projects with the Research Center and Clinic Deaderma, work requiring structured methodology, data analysis and adherence to research standards.
Kicksta sells an Instagram growth service and is therefore an interested party in this subject. The methodology above is published in full so the work can be checked rather than taken on trust. This study is not independent research and is not presented as such.
This study’s own recommendation to brands is not to buy followers — bought audiences collapse exactly the engagement signal that distinguishes a real one. Kicksta grows Instagram audiences organically, by putting your profile in front of real accounts in your niche.