We traced 17 famous UX statistics to their sources. Five have none.
The $1-to-$100 Forrester stat has no Forrester report behind it. The Stanford 75% is really 46.1% of comments. A filterable ledger of every verdict, with primary sources and citations that hold up.
ClapBack Research · July 10, 2026
We traced 17 of the most-quoted UX statistics back to their primary sources. Five have no source at all, five say something meaningfully different from what gets repeated, and the survivors come with conditions their fans never mention.
If you have ever put “every $1 in UX returns $100” in a slide, this page is going to hurt a little. It hurt us too. The whole premise of ClapBack is that UX claims should come with receipts, so we applied the standard to our own industry’s favorite numbers: for each statistic we followed the citation chain to the earliest findable document, read it, and compared what it says to what gets quoted.
How UX statistics go bad
Almost every broken stat in our sample failed the same way: a real, narrow finding got its conditions filed off. A 2010 survey about slow pages becomes a law about “bad experiences.” A content-analysis of 15 women rejecting health websites becomes “94% of first impressions are design-related.” A marketing rule of thumb on a 2001 IBM web page becomes “research” and then, two decades later, mutates into a Forrester attribution nobody can produce.
The roast
UX has a replication crisis it outsourced to listicles. The same tired numbers circle the industry like luggage nobody claims.
The receipt
The mechanism is citogenesis: listicles cite listicles, which cite a blog post, which cites nothing. In our tracing, the most common terminal node was a marketing page or a contributed post, not a study. Where a real paper existed, the popular version usually misstated its scope, its metric, or its year.
The three worst offenders
“Every $1 invested in UX returns $100”
The most-quoted number in UX has no study behind it. The earliest instance we could find is a 2015 Forbes Technology Council post written by the CEO of a UX agency, asserting the figure with no link and Forrester spelled “Forester.” No Forrester report containing it has ever surfaced. The likely ancestor: IBM’s 2001 “Cost Justifying Ease of Use” page, which we recovered from the Wayback Machine. It offers “every dollar invested in ease of use returns $10 to $100” and labels it, in its own words, a rule of thumb.
“75% judge credibility by design (Stanford)”
Stanford’s credibility research is real and worth citing. In Fogg et al.’s study of 2,684 participants, “design look” was the single most-mentioned factor in credibility judgments. The actual number: 46.1% of participants’ comments. The 75% version appears in no Stanford publication we could locate; it lives exclusively in agency blogs that cite each other. Somewhere between 2003 and now, 46.1% of comments became 75% of users, and nobody checked.
“Fixing a bug after release costs 100x more”
The chart everyone screenshots attributes this to the “IBM Systems Sciences Institute,” which was an internal IBM training program, not a research body, and which published no such study. Laurent Bossavit chased every branch of the citation tree for his book on software engineering folklore, and The Register covered the debunk in 2021. Later-is-costlier is directionally supported by Boehm’s 1981 project data. The tidy 100x constant is folklore.
5 / 17
of the most-quoted UX statistics have no traceable source at all. Not an old source, not a weak source: none. Verdicts and citation chains for every stat are in the ledger below.
The ledger: every stat, every verdict
Filter by verdict. Open any row for what the primary source actually says, the link to it, and a copy-ready citation for the versions that survive scrutiny. Where we write “do not cite,” we mean it: there is nothing at the end of that chain.
The earliest findable instance is a 2015 Forbes Council post by a UX-agency CEO, with Forrester misspelled and no report linked. No Forrester publication contains the figure. Most plausible origin: a mutation of IBM's 2001 'rule of thumb' marketing line.
A real 2009 Forrester blog post by Mike Gualtieri compares well-designed vs poorly designed sites: up to 200% higher visit-to-order and over 400% higher visit-to-lead conversion. Two different funnel metrics, best-vs-worst gaps, unpublished underlying data, and nothing about redesign lift or a UI/UX split.
Safe to cite
In 2009, Forrester analyst Mike Gualtieri wrote that well-designed sites can have up to 200% higher visit-to-order and over 400% higher visit-to-lead conversion rates than poorly designed ones; the underlying data was never published.
The number is real but it comes from a 2010 white paper by web-performance vendor Gomez (Equation Research survey, n=1,500), where 'bad experience' means slow or failing pages at peak traffic. Self-reported intent, 16 years old, sold as a general UX statistic ever since.
Safe to cite
In a 2010 survey of 1,500 consumers commissioned by web-performance vendor Gomez, 88% said they would be less likely to return to a site after a bad experience, where bad experience meant slow or failing pages.
Real, peer-reviewed, and correctly attributed. The judgment formed in 50ms is specifically visual appeal of a static screenshot, not usability or trust. The paper's point is that the snap aesthetic judgment can bias everything after it.
Safe to cite
Lindgaard et al. (2006, Behaviour & Information Technology) found that people form a stable judgment of a web page's visual appeal within 50 milliseconds, and that first impression can color later judgments.
The real study is Sillence et al. (CHI 2004, Northumbria University, not Stanford): 15 women evaluating health websites. 94% of comments about sites they rejected concerned design. Trust in the sites they kept was driven by content. A niche mistrust finding became a universal law through repetition.
Safe to cite
In Sillence et al.'s 2004 study of 15 women evaluating health websites, 94% of comments about rejected sites concerned design, while trust in kept sites was driven by content.
Traces to an unsourced line in a UX-agency listicle, copy-pasted for a decade. No study defines 'online businesses', 'fail', or measures usability as the cause. Real business-failure data (US BLS) shows about 20% of new businesses fail in year one, for many reasons.
Correctly attributed. Nielsen derived it from Weinreich et al.'s logs of 25 users' real browsing: time-on-page allows reading at most 28% of an average page's words at 250wpm, roughly 20% in practice. An upper-bound model, not eyetracking, and Nielsen says so himself.
Safe to cite
Analyzing logs of 25 users' real-world browsing, Nielsen (2008) estimated that visitors have time to read at most 28% of the words on an average page, about 20% in practice.
The model is real: N(1-(1-L)^n) with L averaging 0.31 across the analyzed projects gives 85% at n=5. But L ranged from roughly 0.12 to 0.58; at L=0.15 five users find about 56%. It applies to iterative qualitative testing with homogeneous users, and Spool & Schroeder (2001) found 5 users catching about 35% on complex sites.
Safe to cite
Nielsen & Landauer's 1993 model implies about 5 users surface 85% of the problems a given qualitative test can find, assuming an average per-user detection rate of 31%; with harder-to-detect problems, five users find far less.
No study measures a per-field constant. The roots are single-site anecdotes: Imaginary Landscape cut its contact form from 11 fields to 4 and conversions rose 120% (circa 2008, one company, no controls), plus HubSpot's correlational scan of 40,000 landing pages. Modern A/B tests show cutting fields sometimes does nothing or hurts lead quality.
Safe to cite
In one oft-cited case study (circa 2008), web firm Imaginary Landscape cut its contact form from 11 fields to 4 and conversions rose 120%; controlled tests since show the effect is not universal.
Genuine Baymard survey finding, but from the 2021-2022 wave. Baymard re-runs the survey; the current published figure is 19%, behind extra costs at 39%. Self-reported, multi-select, and excludes people who were just browsing.
Safe to cite
In Baymard's latest survey of US online shoppers, 19% cited forced account creation as a reason for abandoning a checkout; earlier waves put it at 24 to 26%.
Baymard's running meta-average across cart-abandonment studies. The 70.19% snapshot fossilized into thousands of posts; the live figure is 70.22% across 50 studies as of 2026. Two decimals on a moving aggregate is false precision, but roughly 7 in 10 carts holds.
Safe to cite
Baymard's running meta-average across 50 studies puts cart abandonment at 70.22% as of 2026, roughly 7 in 10 carts, barely moved since 2020.
Correct methodology, correct attribution, stale year. 83.6% was 2023; 79.1% was 2025; the 2026 report says 83.9%, the most common detected accessibility failure for the eighth straight year. Automated detection on homepages only, so it undercounts.
Safe to cite
In the 2026 WebAIM Million report, 83.9% of the top one million home pages had text failing WCAG 2 AA contrast, the most common accessibility failure for the eighth straight year.
Stanford's real credibility research (Fogg et al., 2,684 participants) found 'design look' mentioned in 46.1% of participants' comments about site credibility, the single most-cited factor. No Stanford publication contains a 75% figure; it appears to be 46.1% inflated through retellings.
Safe to cite
In Stanford's web credibility research (Fogg et al., 2002-2003, 2,684 participants), design look was the most-cited factor in credibility judgments, mentioned in 46.1% of participants' comments.
IBM really published the line, on a 2001 'Cost Justifying Ease of Use' marketing page (recovered via the Wayback Machine), explicitly as a rule of thumb. Its own citations are project-specific usability ROI case studies and a defect-cost ratio about bug fixing. Marketing copy, not a study. Probably the ancestor of the $1-to-$100 Forrester myth.
Safe to cite
A 2001 IBM ease-of-use marketing page offered the rule of thumb that a dollar invested in ease of use returns $10 to $100; an advocacy claim, not a study.
The canonical zombie citation of software engineering. The 'IBM Systems Sciences Institute' was an internal training program, not a research institute, and no underlying study has ever been produced (Bossavit; Hillel Wayne; The Register, 2021). Later-is-costlier has directional support in Boehm's 1981 data; the tidy 100x constant does not.
Safe to cite
Defects generally cost more to fix the later they are found (Boehm, 1981, project-specific data), but the famous 100x chart attributed to the IBM Systems Sciences Institute traces to no findable study.
No named dataset exists. Almost certainly Quettra's 2015 mobile-app retention data (average Android app loses 77% of daily active users within 3 days of install) rebranded as a SaaS fact. Mobile DAU decay and SaaS account churn are different animals.
Safe to cite
Quettra's 2015 analysis found the average Android app loses 77% of daily active users within 3 days of install; the '75% of SaaS users churn in week one' version of this claim has no traceable source.
Correctly attributed and honestly framed by its author: an operator's estimate from his own products and clients, popularized through Intercom's onboarding book. No dataset behind it, and McKenzie never claimed one. Quote it as a practitioner's rule of thumb, never as 'studies show'.
Safe to cite
Patrick McKenzie's rule of thumb from running and advising SaaS businesses: 40 to 60% of free-trial signups use the product once and never come back. A credible operator's anecdote, not a study.
How to cite UX statistics without embarrassing yourself
Four habits cover almost every failure in the ledger. They are also exactly what we require of our own product’s findings, so we can confirm they survive contact with skeptical engineers.
Name the source and the year, inline. “Baymard’s 2025 survey” can be checked in ten seconds. “Studies show” is how a 2010 page-speed survey spends 16 years impersonating a UX law.
Quote the metric the study measured, not the one you wish it measured. Visual appeal in 50ms is not “opinions form in 50ms.” Comments about rejected health sites are not “first impressions.”
Check whether the number is alive. Baymard and WebAIM update their figures; quoting the 2023 snapshot in 2026 is a soft lie with a hard source. The current numbers: 70.22% cart abandonment, 83.9% low-contrast homepages.
Anecdotes are fine when labeled. Patrick McKenzie’s 40-to-60% one-and-done trial users is honest, useful, and openly an operator’s estimate. It only becomes a problem when it gets promoted to “research.”
If you need the numbers for a signup-flow argument specifically, we assembled the surviving ones (with their conditions attached) in why users abandon your signup flow. And when someone sends you a deck with the $100 ROI stat in it, send them here. We’ll keep the ledger current.
Is the every $1 in UX returns $100 statistic real?
No. The earliest findable instance is a 2015 Forbes Council post by a UX-agency CEO, with no Forrester report behind it. It is most likely a mutation of an unsourced rule of thumb from a 2001 IBM marketing page.
Do 5 users really find 85% of usability problems?
Only under conditions. Nielsen and Landauer's 1993 model gives 85% when the average user exposes 31% of problems; with harder-to-detect problems five users find far less, and Spool and Schroeder measured about 35% on complex sites. It is guidance for iterative qualitative testing, not a law.
What percentage of users judge credibility by design?
Stanford's actual finding (Fogg et al., 2003, n=2,684) is that design look was mentioned in 46.1% of participants' comments about site credibility, the most-cited single factor. The widely quoted 75% appears in no Stanford publication.
What is the current average cart abandonment rate?
Baymard's running meta-average across 50 studies is 70.22% as of 2026. The often-quoted 70.19% is an older snapshot of the same living number.
How many homepages fail contrast requirements?
83.9% of the top one million home pages had text failing WCAG 2 AA contrast in the 2026 WebAIM Million report, the most common accessibility failure for the eighth straight year.
Reading about UX debt is the slow way.
Paste your URL and an AI agent will use your product like a real user, find what breaks, and hand you the tickets. From $49, no install, no retainer.