attck-research

Evidence review: How Users Build Trust in Brands

Article: How Users Build Trust in Brands
Reviewed: November 2026 · Method version 2 · How these reviews work

How this review works

Each factual claim in the article was checked against the source it cites, then challenged by reviewers whose job was to find the strongest evidence against it: replication attempts, meta-analyses, later papers that cite the original work, and research from other fields that study the same mechanism. A separate reviewer weighed both sides, and a script re-checked every quoted source. Where a claim was overstated or misattributed, the article was corrected. The research below supports the article as it now reads.

Background research

Introduction

Consumer WebWatch, a project of Consumers Union, wrote five areas into its guidelines for improving web credibility: say who runs the site, label the advertising, offer customer service, publish corrections, post a privacy policy.

In 2002, B.J. Fogg’s Stanford lab worked with Consumer WebWatch and asked 2,684 people to judge the credibility of real websites and say why.

The site’s design look came up in 46.1% of the comments, more than anything else.

Confirmed in the report. Later experiments (Lindgaard et al. 2011) found snap trustworthiness judgments of homepages are largely driven by visual appeal.

The authors wrote: “We found a mismatch, as in other areas of life, between what people say is important and what they actually do.”

Quote confirmed verbatim. The ‘say’ side comes from Consumer WebWatch’s separate 2002 poll, which the report contrasts with this study.

What People Are Actually Trusting

Mayer, Davis and Schoorman (1995) define trust as “the willingness of a party to be vulnerable to the actions of another party based on the expectation that the other will perform a particular action important to the trustor, irrespective of the ability to monitor or control that other party.”

The publisher page is paywalled; the definition was confirmed as quoted in an open-access review.

The Mayer paper names three qualities that recur in the literature as grounds for trust: ability, benevolence and integrity.

Confirmed second-hand through the Colquitt et al. (2007) meta-analysis abstract.

It pays. In data on 107 brands, Arjun Chaudhuri and Morris Holbrook found that brand trust and brand affect together were tied to two kinds of loyalty: repeat purchase, which went with higher market share, and attitudinal loyalty, which went with a higher relative price.

The study uses survey data from one point in time and separates purchase loyalty (linked to market share) from attitudinal loyalty (linked to relative price). A later longitudinal experiment found trust’s effect on loyalty grew over time.

Finding 1

Sillence and colleagues followed 15 women facing a risky health decision as they searched the web for advice over four consecutive weeks.

Their spoken comments supported a staged model: design appeal predicted rejection, while selection (trusting a site’s advice) was predicted by the credibility of its information and how personal the content felt.

Qualitative content analysis of 15 participants; ‘predicted’ is descriptive, not statistical. A 2023 study with about 600 participants found aesthetics also raised trust, not only drove rejection.

That reconciles Fogg’s numbers with common sense. Looks work as a filter. Poor design can lose trust in seconds, and good design can win a first hearing, but trust itself is earned by the content and by how the company behaves over time.

Sillence and colleagues’ own six-month follow-up of the same study describes a staged process: sites were rejected on poor design, and well-designed sites then had to earn trust with credible, personal content. Three studies with more than 1,100 web users found that aesthetics had the largest effect on first impressions while content was what drove people to revisit or recommend a site, and a two-week field experiment found the halo from good looks fades with use. Looks do raise initial trust on their own, so the article now says good design wins a first hearing rather than that it cannot earn trust.

Finding 2

Kim and colleagues found trust was repaired more successfully when the accused party apologized for a failure of competence but denied culpability for a failure of integrity.

Confirmed in the PubMed abstract and consistent with a 2017 meta-analysis (36 samples, N = 4,731, d = 0.44, stronger for competence violations). A 2025 meta-analysis of human-robot studies found repair effects small overall, though apology was among the most effective strategies.

For a brand, the reading is narrow but useful. When an error is yours, say so plainly. Denial can look better in the moment, especially against a charge of dishonesty, but it becomes the worse choice once evidence of fault appears.

Kim et al. found denial beat apology for integrity charges when no further evidence came out, and apology beat denial once evidence of guilt appeared.

The Outage GitLab Ran in Public

On January 31, 2017, a GitLab engineer trying to rebuild a lagging database replica just before midnight UTC ran a deletion on the primary production database instead; about 300 GB of data was gone before the command was stopped.

Changes made between 17:20 UTC and midnight were lost for good: by the company’s estimate roughly 5,000 projects, 5,000 comments and 700 new user accounts.

Confirmed in the postmortem. Git repositories and wikis were not affected by the data loss.

In Kim’s terms, a competence failure met with an apology is the combination that repaired trust better than denial.

Kim et al. compared apology with denial only. A 2017 meta-analysis found apology repairs competence violations better than integrity violations.

Pull quote

“When trust is lost, it can almost always be traced back to a breakdown in one of these three drivers”: authenticity, logic and empathy.

The quoted sentence appears in HBR’s summary of the article, not in the authors’ body text, which says broken trust ‘usually’ traces to one driver.

Conclusion

Trust is the slowest of the three to build, because it rests on behavior a user can only see over time, and it is usually quicker to lose than to win.

The idea that trust is easier to destroy than to build traces to Paul Slovic’s 1993 ‘asymmetry principle’. Later experiments found the asymmetry holds for firsthand experience but not always for information people only read about.

What changed in the article

These edits were made on November 12, 2026 after the review.

What People Are Actually Trusting: unverified

Claim reviewed: A polished interface speaks mostly to ability; benevolence and integrity are earned through behavior.

Now reads: A polished interface makes a strong first impression, and in snap judgments people read good design as a sign of trustworthiness in general. But benevolence and integrity are hard to prove with visuals alone. They are earned through behavior.

Why: This is the author’s own inference and has no source. No study has tested which of the three trust factors visual polish affects. Experiments do show that good looks raise overall trustworthiness ratings, not only judgments of competence, so ‘mostly to the first’ is a guess. Now cites Lindgaard, Dudek, Sen, Sumegi & Noonan, An exploration of relations between visual appeal, trustworthiness and perceived usability of homepages, ACM TOCHI (2011).

What People Are Actually Trusting: supported with caveats

Claim reviewed: In data on 107 brands, Chaudhuri and Holbrook found that brand trust and brand affect together drove loyalty, and loyalty led in turn to higher market share and a higher relative price.

Now reads: It pays. In data on 107 brands, Arjun Chaudhuri and Morris Holbrook found that brand trust and brand affect together were tied to two kinds of loyalty: repeat purchase, which went with higher market share, and attitudinal loyalty, which went with a higher relative price.

Why: The paper does link brand trust and affect to loyalty and loyalty to market share and price, using 107 brands. But it splits loyalty in two: purchase loyalty goes with market share, and attitudinal loyalty with relative price. The data come from one point in time, so ‘drove’ and ‘led’ describe a modeled path, not demonstrated cause and effect.

Finding 1: supported with caveats

Claim reviewed: Looks work as a filter: they can lose trust quickly but cannot earn it on their own (reconciling Fogg’s numbers with Sillence).

Now reads: That reconciles Fogg’s numbers with common sense. Looks work as a filter. Poor design can lose trust in seconds, and good design can win a first hearing, but trust itself is earned by the content and by how the company behaves over time.

Why: The authors of the cited study said the same thing in their six-month follow-up of the same women: poor design got good sites rejected, and well-designed sites then had to earn trust with credible, personal content. Larger studies agree that looks dominate first impressions while content drives whether people come back or recommend a site, and the halo from good looks fades with use. But looks alone do raise initial trust, and even real money in trust games, so ‘cannot earn it on their own’ is too strong; the fix is to say looks win a first hearing and content and behavior earn the trust. Now cites Thielsch, Blotenberg & Jaron, User Evaluation of Websites: From First Impression to Recommendation, Interacting with Computers (2014).

Finding 2: supported with caveats

Claim reviewed: When later evidence showed guilt, having apologized worked better than having denied.

Now reads: For a brand, the reading is narrow but useful. When an error is yours, say so plainly. Denial can look better in the moment, especially against a charge of dishonesty, but it becomes the worse choice once evidence of fault appears.

Why: The finding is in the abstract: when later evidence showed guilt, an earlier apology beat an earlier denial. The article’s gloss that ‘denial only helps when you are genuinely innocent’ is slightly too simple, because in the same study denial also beat apology for integrity charges before any later evidence appeared.

The Outage GitLab Ran in Public: supported with caveats

Claim reviewed: In Kim’s terms, a competence failure met with an apology is the combination most likely to repair trust; GitLab’s outage was a failure of ability and its response worked on integrity and benevolence.

Now reads: In Kim’s terms, a competence failure met with an apology is the combination that repaired trust better than denial.

Why: Kim’s study compared only apology against denial, so it shows an apology beats denial for a competence failure, not that it is the response ‘most likely’ to repair trust among all options. A meta-analysis backs apology for competence failures, but a broader one found repair effects modest overall. The article already says GitLab’s outcome was not measured, which is the right hedge.

Pull quote: supported with caveats

Claim reviewed: Frances Frei and Anne Morriss (Begin with Trust): “Trust has three drivers: authenticity, logic, and empathy. When trust is lost, it can almost always be traced back to a breakdown in one of them.”

Now reads: “When trust is lost, it can almost always be traced back to a breakdown in one of these three drivers”: authenticity, logic and empathy.

Why: The quoted idea is on the HBR page, but in the article’s summary paragraph, which introduces it with ‘the authors argue’, and the wording differs: HBR says ‘three core drivers of trust’ and ‘one of these three drivers’. The body text is softer, saying broken trust ‘usually’ traces to one driver. Pass 2 did not find the sentence; it is there in the page summary.

Conclusion: supported with caveats

Claim reviewed: Trust is the slowest to build, because it rests on behavior seen over time, and the quickest to lose.

Now reads: Trust is the slowest of the three to build, because it rests on behavior a user can only see over time, and it is usually quicker to lose than to win.

Why: Trust being quicker to lose than to gain is a well-known idea from risk research, and experiments support it when people experience outcomes firsthand. But it is not universal: when people only read about good and bad outcomes, trust moved equally in both directions, and a 2026 review describes the asymmetry as dependent on context. The sentence has no source. Now cites Slovic, Perceived Risk, Trust, and Democracy, Risk Analysis (1993).