Myths: claims about copy and persuasion not to rely on
Status: canonical list for this organization; the Research log holds its passes. Each entry has a verdict and its evidence. Before citing a memorable number or a named effect in a principle, brief or page, check it here. Claims about mentioning AI and its legal floor are in the do-not-rely-on list of the note on mentioning AI. Verdicts: failed (a preregistered replication or a direct measurement found no effect, or the data were fabricated), overgeneralized (real under narrow conditions, quoted as a law), misattributed (credited to the wrong source), unsourced, misapplied (true for something else), unlawful (prohibited toward consumers under the cited law, whatever its effect). A phrase after a verdict limits its scope.
Psychology
| Claim | Verdict | Evidence |
|---|---|---|
| Hard-to-read text makes people think harder and solve problems better | failed | 17 pooled experiments found no effect (Meyer et al. 2015). Easy text reads as more credible and more true. |
| Ask early, because users run out of willpower (ego depletion) | failed | 23 labs, N 2,141, d 0.04 (Hagger et al. 2016). |
| Confident body language in the hero image changes behaviour (power posing) | failed | Failed replication; the lead author no longer believes the effect (Ranehill et al. 2015, Carney). |
| Subtle background images or words steer choices (social priming) | failed | Flag and currency priming did not replicate in Many Labs (Klein et al. 2014); Kahneman wrote that he "placed too much faith in underpowered studies" (Kahneman 2017). |
| More choice always kills conversion (the jam study as law) | overgeneralized | Meta-analytic mean effect about zero; overload appears only with complex sets, hard tasks, unclear preferences or time pressure (Scheibehenne 2010, Chernev 2015). |
| A decoy plan reliably pushes buyers to the target (the Economist example) | overgeneralized | The example is an unpublished classroom demo; 91 attempts produced only 11 reliable effects, and with verbal or pictorial descriptions the effect fell to chance (Yang and Lynn 2014; Frederick et al. 2014). |
| Signing at the top of a form makes people honest | failed | Data fabricated; paper retracted (Data Colada 98). |
| Anything resting only on Dan Ariely's research | failed | Two retractions (2021, 2026) and two expressions of concern (Ariely credibility). |
| Losses loom twice as large, so always frame as a loss | overgeneralized | Average loss aversion about 2 in risky choice, but smaller for small stakes and experienced buyers, and contested (loss aversion). Test gain against loss framing. |
| Making people feel short of money or time changes how they decide (scarcity mindset priming) | failed | 4 of the 18 originally significant studies replicated, and the original authors dispute the audit (O'Donnell et al. 2021). These studies concern lacking resources, not product scarcity cues: real, stated limits are fine; fake ones are unlawful. |
| "Be a [noun]" identity wording guarantees a lift | failed | The "be a voter" effect failed a large field test (Gerber et al. 2016). |
| Nudges reliably work | overgeneralized | After publication-bias correction information and assistance nudges show about zero; defaults remain strong in their own meta-analysis (Maier 2022, Jachimowicz 2019). |
| The Hook model and the Fogg model are validated science | overgeneralized | Useful checklists, not tested predictive theories (Fogg, Eyal). |
| People trust algorithms more than people, so lead with AI (algorithm appreciation) | overgeneralized | Appreciation appears only when AI seems more capable and personalization is unnecessary; otherwise people show aversion (Qin et al. 2025). It fades with expertise and personal stakes (do not rely on). |
| Fake countdowns, fake stock, unverified reviews or invented "was" prices are clever growth tactics | unlawful | Prohibited or regulated toward consumers in Germany: UWG annex numbers 7 (false claim that an offer is available only for a very limited time), 23b and 23c (unverified and fake reviews); section 5 UWG on false statements about availability, price advantages and price, which presumes a reduction misleading when the earlier price was charged only briefly; section 11 PAngV for price reductions on goods (legal context, not legal advice; https://www.gesetze-im-internet.de/uwg_2004/__5.html, https://www.gesetze-im-internet.de/uwg_2004/anhang.html and https://www.gesetze-im-internet.de/pangv_2022/__11.html, read 29 October 2026). |
Web reading and testing
| Claim | Verdict | Evidence |
|---|---|---|
| People have an eight-second attention span, shorter than a goldfish | unsourced | Commonly traced to a 2015 Microsoft Canada marketing report that cited a secondary stats site; no study measures attention span as one number, and the claim has no primary scientific source. |
| Visuals are processed 60,000 times faster than text | unsourced | No primary source located. |
| Communication is 7 percent words, 38 percent tone, 55 percent body language | misapplied | Mehrabian: the numbers apply only to communicating feelings and attitudes; "Unless a communicator is talking about their feelings or attitudes, these equations are not applicable" (http://www.kaaj.com/psych/smorder.html). |
| The three-click rule | failed | Joshua Porter, UIE, 2003: no correlation between click count and success or satisfaction (44 users, 8,000+ clicks) (https://articles.centercentre.com/three_click_rule/). |
| Users do not scroll | overgeneralized | 43 percent of viewing time is below the fold, but attention drops sharply at the fold (Fessenden 2018). |
| Users read exactly 20 percent of words | overgeneralized | A 2008 model estimate from 2005 logs of a small panel; direction, not a constant (Nielsen 2008). |
| Good web writing improves usability by 124 percent | overgeneralized | Real result, one site, 51 users, against a deliberately promotional control (Morkes and Nielsen 1997). |
| Grade 5 to 7 copy doubles conversion | overgeneralized | Correlation across pages with very different offers (Unbounce 2024). |
| Changing a button colour or CTA wording lifted conversion by X percent | overgeneralized | In roughly 2,600 categorized tests out of 6,700, the average effect was 0.0 percent for colour and minus 0.3 percent for CTA wording (Qubit 2017). |
| Experts and B2B buyers want sophisticated, technical language | failed | Preference for plain versions rose with education, to 86 percent among lawyers (Trudeau 2012); needless complexity lowers judged intelligence (Oppenheimer 2006). |
| Inline validation cuts errors by 22 percent (Baymard) | misattributed | Wroblewski and Etre 2009, 22 participants (Wroblewski 2009). |
| 46 percent of people judge credibility by design | overgeneralized | 46.1 percent of comments, not of people, in a study of 2,684 participants; it does not show cause (Fogg et al., Consumer WebWatch, October 2002: https://advocacy.consumerreports.org/research/how-do-people-evaluate-a-web-sites-credibility/, read 29 October 2026). |
| You have 50 milliseconds to make a first impression | misapplied | Lindgaard et al., Behaviour & Information Technology, 2006: about visual appeal judgments, not comprehension of copy. |
| Five times as many people read the headline as the body | misapplied | Ogilvy, 1963, about print ads (Ogilvy). The direction holds on the web; the number is not web data. CXL's 2018 eye tracking of a 300-word article found 97 percent read the title and 60 percent skimmed to the end (web reading note). |
| Most A/B tests win | failed | Real win rates are 8 to 33 percent even at Microsoft, Google, Booking and Airbnb, and many "wins" are false positives (Kohavi et al. 2022). |
| Optimize for AI answers with llms.txt or special markup | unsourced for search | Google says no special files or markup are needed for AI Overviews or AI Mode; GEO tactics rest on one lab paper (Google 2025, GEO). A file written for an agent that a user points at the site is a different job; see the agent-reader principles. |
| E-E-A-T is a ranking factor | misapplied | Google: E-E-A-T is not itself a ranking factor; trust matters most (Google helpful content). |
Copywriting canon
| Claim | Verdict | Evidence |
|---|---|---|
| Long copy always outsells short copy | overgeneralized | Length follows awareness and the size of the decision; Ogilvy stressed it most for expensive purchases (disagreements). |
| A new category name makes you the category king | overgeneralized | Survivorship bias; only 7 percent of the Nasdaq IPOs Dunford examined created a new category (Play Bigger critique). |
| Curiosity headlines outperform clear ones | failed for product pages | Caples: "For every curiosity headline that succeeds in getting results, a dozen will fail"; curiosity belongs in ads and content for readers not yet looking (Caples). |