oddlynew
Sign inStart for free
LibrarySource: Psychology of persuasion and its replication recordv7 · Updated Report

Source: Psychology of persuasion and its replication record

Status: source note, first research pass on 29 September 2026 by a research agent. Every citation was checked against a publisher, journal, author page or reputable secondary source in that pass; exceptions are flagged in place; section 3 carries its own "Last checked" date. Each effect carries a rating: robust, mixed, or weak/failed. Legal notes are orientation, not legal advice.

In short

  • Rules it backs: use the robust effects honestly: anchoring, defaults, picture superiority, left-digit pricing (principle 19); treat decoys, scarcity cues and choice-overload claims as hypotheses (20); never use persuasion you would not show the reader (21). Principles holds where this note differs.

  • Is this effect real? 2. Replication scorecard; failed and misapplied claims are in Myths.

  • What the law forbids (fake urgency, unverified reviews, reference prices, hard cancellation): 3. EU and German legal context, orientation, not legal advice.

Evidence ratings

  • Robust: replicated at scale or supported by a meta-analysis that survives bias checks.

  • Mixed: real but smaller, context-dependent, or contested.

  • Weak / failed: failed preregistered replication, the data were fabricated, or there is essentially no empirical test.

  • A range such as "weak to mixed" sits between two classes; for testing, treat it as the lower one. Mandatory marks a legal requirement, not an evidence class.

0. Background: the replication crisis

  • Open Science Collaboration (2015), "Estimating the reproducibility of psychological science", Science 349. URL: https://www.science.org/doi/10.1126/science.aac4716

    • The team repeated 100 studies from three top journals. 36% of the repeats were significant, against 97% of the originals.

    • Replication effect sizes were about half the original size.

    • Lesson: treat any single dramatic study as a hypothesis to A/B test, not a law.

  • Klein et al. (2014), "Investigating variation in replicability: A Many Labs replication project", Social Psychology 45(3). URL: https://econtent.hogrefe.com/doi/10.1027/1864-9335/a000178

    • 13 effects were tested in 36 samples (N = 6,344). 10 replicated, anchoring strongly.

    • Flag priming and currency priming did not replicate.

    • Pattern: big, obvious effects hold up; small, subtle priming effects do not.

1. Resources

1.1 Robert Cialdini, Influence (1984; Influence, New and Expanded, Harper Business 2021) and Pre-Suasion (Simon & Schuster 2016)

URLs:

Summary: The 2021 edition adds a seventh principle, Unity (shared identity), to the original six: reciprocation, commitment and consistency, social proof, liking, authority and scarcity. Pre-Suasion argues that what people attend to just before a message changes how they receive it.

Principles for copy:

  1. Social proof works best when it is specific and close to the reader. "Teams like yours (40-person agencies) use X" beats "thousands of users". In Goldstein, Cialdini & Griskevicius 2008 (JCR, https://academic.oup.com/jcr/article-abstract/35/3/472/1856257), a sign saying most guests in this room reused towels got 49% reuse. The standard environmental appeal got 37%, and identity-based norm messages got 41 to 44%.

  2. Never advertise a bad norm. "Most users never finish setup" teaches readers that not finishing is normal. Schultz et al. 2007 (Psychological Science, https://journals.sagepub.com/doi/10.1111/j.1467-9280.2007.01917.x) found that a descriptive norm pushed low users toward the average (a boomerang effect). Adding an approving cue (injunctive norm, e.g. a smiley) removed the boomerang.

  3. Reciprocity: give something useful before you ask. Examples: free tools, templates, a genuinely useful free tier.

  4. Authority and credibility: name real customers, real numbers and real credentials.

  5. Unity: speak as a member of the reader's group ("built by engineers for engineers") only when it is true.

  6. Scarcity only when it is real. Fake scarcity is illegal in Germany (see section 3).

Evidence: Mixed.

  • Norm effects are real but smaller and more context-bound than the book suggests.

  • Bohner & Schlüter 2014 (PLOS ONE, https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0104086) repeated the towel study in two German hotels. Norm messages beat no message, but did not beat the standard appeal.

  • A Bayesian synthesis still supports the norm effect: Scheibehenne, Jamil & Wagenmakers 2016, Psychological Science, https://journals.sagepub.com/doi/abs/10.1177/0956797616644081. A multilevel Bayesian reanalysis of the same seven studies found the evidence moderate to inconclusive once the studies may differ in effect: Carlsson, Schimmack, Williams & Bürkner 2017, Psychological Science, https://doi.org/10.1177/0956797616684682, read 29 October 2026.

  • Much of Pre-Suasion rests on attention and association (priming) effects. That is the category that has fared worst in replication.

1.2 Daniel Kahneman, Thinking, Fast and Slow (Farrar, Straus and Giroux, 2011)

Summary: the book popularised the System 1 (fast, intuitive) and System 2 (slow, deliberate) model, and gave copywriters their standard concepts: framing, anchoring, loss aversion and WYSIATI ("what you see is all there is").

Principles for copy:

  1. WYSIATI: readers judge only from what is on the page. Put the one decisive fact (price, what it replaces, the proof) where it will be seen. If it is not on the page, it does not exist for System 1.

  2. Frame attributes positively when the numbers are equivalent. "99.9% uptime" beats "down 0.1% of the time". Levin, Schneider & Gaeth 1998 (OBHDP, https://www.sciencedirect.com/science/article/abs/pii/S0749597898928047) separates attribute, risky-choice and goal framing.

  3. Anchors move estimates. The first number a reader sees shapes what follows. Tversky & Kahneman 1974 (Science, https://www.science.org/doi/10.1126/science.185.4157.1124): a rigged wheel landing on 10 or 65 shifted estimates of African UN members from about 25% to about 45%.

  4. Show the cost of the status quo in concrete terms (hours lost, errors shipped). Do not assume a loss frame beats a gain frame; test both.

  5. Design for System 1 at the moment of decision. Keep the path short and readable, and cut the steps that call for effortful thinking.

Evidence:

1.3 Processing fluency: Oppenheimer (2006), Alter & Oppenheimer (2009) and related work

Principles for copy:

  1. Use short, common words. Complexity reads as less intelligent, not more.

  2. High contrast and legible type. Readability feeds perceived truth and perceived ease.

  3. Describe onboarding steps in plain, easy-to-read text. Hard-to-read instructions make the task itself feel harder.

  4. Repeat the core claim consistently across hero, features and CTA (call to action). Repetition raises perceived truth, so use it only for true claims.

  5. Do not make text harder to read on purpose "to make people think". That idea failed replication (Myths: Psychology).

Evidence:

  • Truth and fluency effects: robust (Dechêne et al. 2010 meta-analysis).

  • The long-words finding: mixed (Oppenheimer 2006, small student samples; not replicated at scale as far as I found).

  • Song & Schwarz: mixed (a small single report, not independently replicated at scale as far as I found).

  • The disfluency benefit: failed. Meyer et al. 2015 (JEP: General, https://pubmed.ncbi.nlm.nih.gov/25844628/) pooled 17 experiments and found no effect.

1.4 BJ Fogg, Tiny Habits (Houghton Mifflin Harcourt, Dec 31 2019) and the Fogg Behavior Model (B=MAP)

Summary: B=MAP says a behaviour happens when Motivation, Ability and a Prompt converge at the same moment. The easier the behaviour, the less motivation it needs.

Principles for copy:

  1. Raise ability before motivation. Cut steps and fields, and state the time cost ("2-minute setup").

  2. Every prompt needs a clear, single next action, and it should arrive when motivation is present, for example right after a success moment.

  3. Shrink the first action to something trivially easy: "Import one repo", not "Migrate your stack".

  4. Celebrate small wins in product copy to reinforce the behaviour.

Evidence: Weak to mixed. It is a useful conceptual checklist. I found little independent experimental validation of the model as a predictive theory. The underlying idea (reduce friction) is well supported elsewhere, for example by the defaults literature.

1.5 Nir Eyal, Hooked: How to Build Habit-Forming Products (Portfolio/Penguin, 2014), and its critiques

Summary: the Hook model (trigger, action, variable reward, investment) is standard in SaaS growth. Eyal includes a "Manipulation Matrix" ethics test: would you use the product yourself, and does it materially improve users' lives? His later book, Indistractable (2019), addresses the backlash.

Principles for copy:

  1. Link external triggers (emails, notifications) to an internal trigger the user already feels, such as anxiety about a broken build.

  2. Make the action the simplest possible one.

  3. Let users invest (data, configuration) so the product improves with use, and say so in the copy.

  4. Apply the Manipulation Matrix test before shipping any habit loop.

Evidence: Weak. This is a practitioner synthesis, not tested as a model. Critics argue it teaches variable-reward mechanics that make stopping harder, and that "hooked" is not the same as valued. See Axbom, https://axbom.com/nir-eyal-habit-danger/.

1.6 Stephen Wendel, Designing for Behavior Change (O'Reilly, 2013; 2nd ed. 2020)

Summary: the CREATE action funnel (Cue, Reaction, Evaluation, Ability, Timing, Experience) and the DECIDE process build in measurement.

Principles for copy:

  1. Audit each CTA through CREATE: is there a cue, is the gut reaction positive, is the cost and benefit clear, is it doable, is it urgent now, and does past experience help or hurt?

  2. Evaluation: state the benefit and the cost side by side, and make the ask small.

  3. Timing: ask when the user has just felt the need.

  4. Measure the behaviour itself, not clicks. Treat every "principle" as a hypothesis to test.

Evidence: Mixed. The framework draws on robust findings, and its emphasis on experimentation is its strength. The funnel itself has not been validated as a model.

1.7 Rory Sutherland, Alchemy: The Surprising Power of Ideas That Don't Make Sense (2019)

Summary: introduces psycho-logic, the view that people run on evolved heuristics rather than economic logic. It also argues that costly signals (effort, expense, sacrifice) carry meaning that cheap information cannot.

Principles for copy:

  1. Visible effort signals quality and commitment. Examples: a real founder note, detailed docs, a guarantee that would cost you money if you broke it.

  2. Reframe rather than re-engineer. Changing the perceived value (the wait, the risk, the effort) can beat improving the metric.

  3. Reduce perceived risk: free trials, easy cancellation, "no credit card".

  4. Test the "illogical" idea as well as the logical one.

Evidence: Weak. It is witty, advertising-industry anecdote. Signalling theory itself is well established in economics and biology, but the book's specific claims are rarely tested.

1.8 Richard Thaler and Cass Sunstein, Nudge (Yale University Press, 2008; Nudge: The Final Edition, 2021)

Summary: introduced choice architecture and defaults. The 2021 edition adds sludge: friction that works against the user, such as cancelling being harder than signing up.

Principles for copy and product:

  1. Defaults are the strongest lever you have. Pre-select the option that is genuinely best for most users, such as the annual plan with a clear saving or a sensible notification level.

  2. Remove sludge from paths that serve the user (sign-up, cancel, export). For consumer contracts in Germany, the cancel and withdrawal paths are also legal requirements (section 3).

  3. Make the recommended option explicit. Defaults work partly through endorsement ("this is what we recommend").

  4. Do not treat "nudges" in general as proven. Test them.

Evidence:

1.9 Decoy effect and Dan Ariely's Economist example

Source: Huber, Payne & Puto 1982, "Adding asymmetrically dominated alternatives", JCR. URL: https://academic.oup.com/jcr/article-abstract/9/1/90/1839380

Ariely, Predictably Irrational (2008), https://en.wikipedia.org/wiki/Predictably_Irrational

  • The Economist offers were web-only $59, print-only $125 and print+web $125.

  • With the decoy, 16% of MIT students chose web-only and 84% chose print+web.

  • Without the decoy, 68% chose web-only and 32% chose print+web. (Figures as reported: https://en.wikipedia.org/wiki/Decoy_effect)

  • This was a classroom demonstration, not a peer-reviewed field test.

Ariely credibility issues

Replication of the decoy effect

  • Frederick, Lee & Baskin 2014, "The limits of attraction" (JMR 51(4), https://journals.sagepub.com/doi/abs/10.1509/jmr.12.0061, read 29 October 2026): across 38 studies, the effect appeared in 4 of 5 tests with purely numeric attributes, in 2 of 5 mixed tests, and in none of 27 where an attribute could be seen, tasted or touched.

  • Yang & Lynn 2014, "More evidence challenging the robustness and usefulness of the attraction effect" (JMR 51(4), https://journals.sagepub.com/doi/10.1509/jmr.14.0020, read 29 October 2026): 91 attempts across 23 product classes produced only 11 reliable effects; with verbal or pictorial descriptions, effects fell to chance level.

  • The effect held mainly with simple numeric attributes. With realistic stimuli, an inferior option could just as easily reduce the target's share.

  • Huber, Payne & Puto replied to both papers with "Let's be honest about the attraction effect" (https://journals.sagepub.com/doi/10.1509/jmr.14.0208).

Principles for copy:

  1. A dominated option can make the target plan look better, but only when attributes compare cleanly (same features, clear numbers).

  2. Do not ship fake plans nobody should buy. Test real tiers instead.

  3. Make the comparison dimension explicit, for example price per seat.

Evidence: Mixed for the decoy effect. Weak for the Economist anecdote, and do not cite Ariely as authority.

1.10 Chip and Dan Heath, Made to Stick (Random House, 2007)

Summary: the SUCCESs checklist (Simple, Unexpected, Concrete, Credible, Emotional, Stories) and the popularised curse of knowledge.

The curse of knowledge:

Principles for copy:

  1. Find the core, one sentence a newcomer can repeat. Test it on someone outside the team; you are the tapper.

  2. Concrete beats abstract: "catches the N+1 query before merge" instead of "improves performance".

  3. Open with the unexpected (break a schema), then close the gap.

  4. Make credibility testable ("try it on your repo in 60 seconds"), not just authoritative.

  5. Tell a customer story with a specific person, problem and outcome.

Evidence: Mixed. The curse of knowledge is well supported. The checklist is a synthesis that has not been tested as a package.

1.11 Jonah Berger, Contagious (Simon & Schuster, 2013) and Magic Words (Harper Business, 2023)

Summary:

Key studies:

Principles for copy:

  1. Use concrete, specific nouns and verbs in support replies, docs and UI.

  2. Give users something that makes them look smart when they share it (social currency): benchmarks, reports.

  3. Tie the product to a frequent trigger in the user's week ("every time CI fails...").

  4. Be confident where warranted. Drop needless hedges.

  5. Identity nouns ("be a contributor") are worth testing but unproven (see evidence).

Evidence:

  • Concreteness: robust to mixed (large field data).

  • Virality and arousal: mixed (one influential dataset plus experiments).

  • Noun/identity wording: weak to mixed. The famous "be a voter" effect failed a large field test (Gerber et al. 2016, PNAS, https://www.pnas.org/doi/10.1073/pnas.1513727113).

1.12 Concreteness, specificity, precise numbers, and the picture superiority effect

Principles for copy:

  1. Show the product. Real screenshots and diagrams are remembered better than feature prose.

  2. Precise figures read as measured: "cut build time 37%", not "about a third", but only if it is true.

  3. Replace adjectives with evidence: numbers, named customers, before and after.

Evidence:

  • Picture superiority: robust (for memory).

  • Precision effects: mixed (fewer replications; with experts, too much precision backfires: Loschelder et al. 2016).

1.13 Pricing-page psychology

Principles for pricing pages:

  1. Show three tiers with the recommended one in the middle and visibly marked. The center-stage and compromise effects work in the same direction.

  2. Show the highest tier or annual total first as a legitimate anchor. Never use fake "was" prices. In Germany a price reduction on goods offered to consumers states the lowest price of the previous 30 days (§ 11 PAngV); for a service such as a subscription, a reduction from a price you never really charged misleads about the price (§ 5 UWG; section 3).

  3. Use a 9-ending only when it changes the left digit (€49 vs €50). For premium, emotional or brand-led products, round numbers may fit better.

  4. Keep plans few and comparable when buyers are uncertain. Offer more options where buyers are experts with clear preferences.

  5. Put features in the same rows in the same order so comparison stays fluent.

Evidence:

  • Left-digit effect: robust.

  • Charm endings on sales: mixed to robust (field evidence, moderated by context).

  • Center-stage and compromise effects: mixed (few independent large replications found).

  • Choice overload: mixed. The average is close to zero; the effect depends on moderators.

1.14 Loss aversion (debated)

Principles for copy:

  1. Loss framing is not automatically stronger. Test "Stop losing 6 hours a week" against "Get 6 hours back".

  2. Loss aversion is weaker for small stakes and experienced buyers (Harinck et al. 2007; Mrkva et al. 2020), which describes many B2B buyers.

  3. Endowment can help: a trial with the user's own data can create something to lose (no source yet; test it). Make it easy to keep, and never hold it hostage.

Evidence: Mixed. It holds on average in risky choice, but it is not universal and its size varies.

1.15 Scarcity

  • Barton, Zlatevska & Oppewal 2022 (Journal of Retailing 98(4), https://www.sciencedirect.com/science/article/pii/S0022435922000434). 416 effect sizes from 131 studies: scarcity cues raise purchase intentions.

    • Demand-based scarcity works best for utilitarian products.

    • Supply-based scarcity works best for experiences.

    • Time-based scarcity works best for high-involvement products.

  • Luguri & Strahilevitz 2021 (section 1.16): in their second experiment, "must act now" messages did not make consumers more likely to buy a costly service.

  • O'Donnell et al. 2021 (PNAS 118(44), https://www.pnas.org/doi/10.1073/pnas.2103313118; correction of November 2022: https://www.pnas.org/doi/10.1073/pnas.2217321119, read 29 October 2026) set out to replicate 20 studies of how lacking resources such as money or time changes thinking; these are not studies of product scarcity cues. 4 of the 18 originally significant studies replicated; 80% of effects were smaller and 30% went the opposite way. The original authors dispute the audit (https://www.pnas.org/doi/10.1073/pnas.2206054120).

Principles:

  1. Real limits (a genuine launch cohort cap, a real deadline) can be stated plainly.

  2. Never use fake countdowns, fake stock or fake "3 people viewing". These are illegal in Germany and erode trust.

Evidence:

  • Product scarcity cues on intentions: mixed (mostly intention measures).

  • Scarcity mindset priming: weak.

1.16 Dark patterns and ethics

Principles:

  1. Dark patterns work, which is exactly why they are regulated. The measured conversion gain is real, and so is the legal and trust cost.

  2. Keep symmetry: cancel is as easy as sign-up; declining is as visible as accepting.

  3. No confirmshaming ("No thanks, I like wasting time"), no pre-ticked paid add-ons, no hidden costs.

Evidence: Robust that dark patterns shift behaviour. Robust that they carry legal risk in the EU.

2. Replication scorecard

EffectStatusKey source
Anchoring (numeric estimates)RobustKlein et al. 2014; Tversky & Kahneman 1974
Incidental anchors in valuationWeaker (half to a third the size)Maniadis, Tufano & List 2014
Framing / prospect theoryRobustRuggeri et al. 2020
Default effectsRobust (heterogeneous)Jachimowicz et al. 2019; Johnson & Goldstein 2003
Nudges in generalMixed (about zero after bias correction, except structure: undecided)Maier et al. 2022
Loss aversionMixed (λ about 2 on average; moderated and contested)Gal & Rucker 2018; Mrkva et al. 2020; Brown et al. 2024; Yechiam & Zeif 2025
Fluency and truth effectRobustDechêne et al. 2010; Alter & Oppenheimer 2009
Disfluency improves thinkingFailedMeyer et al. 2015
Ego depletionFailed (d = 0.04, 95% CI from -0.07 to 0.15; 23 labs, N = 2,141)Hagger et al. 2016, https://journals.sagepub.com/doi/10.1177/1745691616652873
Power posing (hormones, behaviour)Failed; lead author: "I do not believe that 'power pose' effects are real"Ranehill et al. 2015, https://journals.sagepub.com/doi/abs/10.1177/0956797614553946; Carney statement, https://faculty.haas.berkeley.edu/dana_carney/pdf_My%20position%20on%20power%20poses.pdf
Social or behavioural primingFailed or much weakerKlein et al. 2014; Kahneman 2017 comment; the web-page priming study (Mandel & Johnson 2002, https://academic.oup.com/jcr/article-abstract/29/2/235/1839982) has, as far as I found, no large preregistered replication
Choice overload (jam study)Mixed (mean effect about zero; moderators)Scheibehenne et al. 2010; Chernev et al. 2015
Decoy / attraction effectMixed (fragile outside numeric displays)Frederick et al. 2014
Scarcity mindset primingWeak (4 of 18 replicated)O'Donnell et al. 2021
Product scarcity cuesMixed (positive meta-analysis, intention-based)Barton et al. 2022
Descriptive social normsMixed (real, smaller, context-bound)Goldstein et al. 2008; Bohner & Schlüter 2014; Scheibehenne et al. 2016
Honesty pledge at top of formFailed (fabricated, retracted)Data Colada 98; PNAS retraction
Picture superiorityRobustShepard 1967
Left-digit pricingRobustThomas & Morwitz 2005; Anderson & Simester 2003

3. EU and German legal context (brief; not legal advice)

Last checked: 27 October 2026, against the UWG annex and § 5 UWG as amended by BGBl. 2026 I Nr. 43 (in force 27 September 2026) and the European Parliament's legislative train for the Digital Fairness Act. Everything below held on that date. The monthly Loop: Recheck the legal blocks keeps this line current.

  1. UWG blacklist (Anhang zu § 3 Abs. 3 UWG, always unlawful toward consumers): https://www.gesetze-im-internet.de/uwg_2004/anhang.html (read 29 October 2026)

    • Nr. 7: falsely claiming that goods or services are available only for a very limited time. This covers fake countdowns and fake "offer ends tonight".

    • Nr. 5: bait offers without adequate stock or capacity.

    • Nr. 23b: claiming reviews come from real buyers or users without reasonable and proportionate verification.

    • Nr. 23c: submitting, commissioning or misrepresenting fake reviews or endorsements, including on social media.

    • Nr. 23b and 23c apply since 28 May 2022 (Omnibus Directive (EU) 2019/2161). § 5b Abs. 3 UWG also requires you to say whether and how you verify reviews (https://www.omsels.info/die-verbote-oder-was-darf-ich-nicht/%C2%A7%C2%A7-5a-5b-uwg-informationspflichten/%C2%A7-5a-abs-1-uwg-wesentliche-informationspflichten/verbraucherbewertungen-%C2%A7-5b-abs-3-uwg).

    • The blacklist protects consumers. B2B copy is still subject to the general ban on misleading practices.

    • Since 27 September 2026 (Drittes Gesetz zur Änderung des UWG, BGBl. 2026 I Nr. 43, transposing Directive (EU) 2024/825), the blacklist also bans: a sustainability label not based on a certification scheme or set by a public authority (Nr. 2a); a generic environmental claim without proof of recognised excellent environmental performance (Nr. 4a); an environmental claim about the whole product or business when it holds for one aspect only (Nr. 4b); a claim of neutral, reduced or positive climate impact based on offsetting (Nr. 4c); presenting legal requirements that apply to every product in the category as a feature of the offer (Nr. 10a); and misleading claims about software updates, durability and reparability of goods, including goods with digital elements (Nr. 23d, points a to g). Nr. 5, 7, 23b and 23c are unchanged. § 5 Abs. 3 Nr. 4 UWG adds that a claim about future environmental performance needs clear, public and verifiable commitments in a detailed plan that an independent expert checks regularly (https://www.recht.bund.de/bgbl/1/2026/43/regelungstext.pdf; https://www.gesetze-im-internet.de/uwg_2004/__5.html, read 27 October 2026).

  2. § 11 PAngV (since 28 May 2022): an advertised price reduction for goods offered to consumers, including strike-through prices, must state the lowest total price charged in the previous 30 days (https://www.gesetze-im-internet.de/pangv_2022/__11.html, read 29 October 2026). § 11 names goods only. For services, SaaS plans included, § 5 UWG still applies: a misleading statement about a price advantage is unlawful (§ 5 Abs. 2 Nr. 2), and advertising a reduction from a price charged only for an unreasonably short time is presumed misleading (§ 5 Abs. 5) (https://www.gesetze-im-internet.de/uwg_2004/__5.html, read 29 October 2026). Both limit "anchor" pricing with fake reference prices.

  3. § 312k BGB cancellation button (since 1 July 2022): online-concluded consumer subscriptions need a button labelled "Verträge hier kündigen" or an equally unambiguous label, leading straight to a confirmation page with a button labelled "jetzt kündigen" or an equally unambiguous label. Both buttons and the confirmation page stay permanently available and easy to reach. Without them, consumers may cancel at any time without notice (https://www.gesetze-im-internet.de/bgb/__312k.html, read 29 October 2026).

  4. Withdrawal button (Directive (EU) 2023/2673, new Art. 11a Consumer Rights Directive; in Germany § 356a BGB, since 19 June 2026): consumer distance contracts with a withdrawal right that are concluded through an online interface need a withdrawal function labelled "Vertrag widerrufen" (EU text: "withdraw from contract here") or an equally unambiguous label, prominent and available throughout the withdrawal period. It leads to a confirmation function labelled "Widerruf bestätigen" (EU text: "confirm withdrawal") or an equally unambiguous label (https://eur-lex.europa.eu/eli/dir/2023/2673/oj/eng; https://www.gesetze-im-internet.de/bgb/__356a.html, read 29 October 2026).

  5. DSA Art. 25: online platforms "shall not design, organise or operate their online interfaces in a way that deceives or manipulates" users (Regulation (EU) 2022/2065, Art. 25 and Recital 67: https://eur-lex.europa.eu/eli/reg/2022/2065/oj/eng, read 29 October 2026). Recital 67 gives examples: repeated prompts, cancelling harder than signing up. It does not apply where the UCPD or GDPR already cover the practice. A typical SaaS is not an "online platform" but is covered by UWG and UCPD.

  6. Digital Fairness Act: the Commission's work programme lists it for Q4 2026. It targets dark patterns, addictive design, influencer marketing and unfair personalisation. As of 27 October 2026 it had not been tabled; legal press expected the proposal on 11 November 2026, and it is unlikely to apply before the end of 2028 (https://www.europarl.europa.eu/legislative-train/theme-protecting-our-democracy-upholding-our-values/file-digital-fairness-act; https://cms.law/en/deu/legal-updates/inside-the-eu-s-planned-digital-fairness-act-digital-design-subscription-traps-and-influencer-marketing).

4. 20 psychology-backed rules for website and product copy

  1. Write in short, common words. Complexity lowers perceived intelligence and credibility. Oppenheimer 2006; Alter & Oppenheimer 2009. Robust (fluency) / mixed (long words).

  2. Make it easy to read: high contrast, legible type, scannable structure. Fluency raises perceived truth, and readers scan more than they read: one model puts what users have time to read at about 20% of a page's words, at most 28% (NN/g 2008, https://www.nngroup.com/articles/how-little-do-users-read/; a model estimate, not a constant). Reber & Schwarz 1999. Robust.

  3. Put the decisive fact where the eye lands. Readers judge only what is on the page. Kahneman 2011 (WYSIATI). Mixed (sound principle, not a single tested effect).

  4. Replace adjectives with concrete specifics: numbers, named outcomes, real screenshots. Packard & Berger 2021; Heath & Heath 2007. Robust to mixed.

  5. Show, do not only tell. Pictures and diagrams are remembered better than text. Shepard 1967. Robust (memory).

  6. Frame equivalent facts positively ("99.9% uptime"). Levin et al. 1998; Ruggeri et al. 2020. Robust.

  7. Test loss framing against gain framing; do not assume loss wins. Gal & Rucker 2018; Mrkva et al. 2020; Brown et al. 2024. Mixed.

  8. Set a legitimate anchor first (full annual price, top tier, cost of the status quo). Anchors move estimates, but fake reference prices are unlawful. Tversky & Kahneman 1974; Klein et al. 2014; § 5 UWG; § 11 PAngV for goods. Robust.

  9. Pre-select the option that is genuinely best for most users and label it recommended. Johnson & Goldstein 2003; Jachimowicz et al. 2019. Robust.

  10. Put the recommended tier in the middle of three. Valenzuela & Raghubir 2009; Simonson 1989. Mixed.

  11. Use 9-endings only when they change the left digit; use round prices for premium or emotional positioning. Thomas & Morwitz 2005; Anderson & Simester 2003; Wadhwa & Zhang 2015. Robust (left digit) / mixed (round vs precise).

  12. Keep options few and comparable when buyers are uncertain; allow depth for experts. Chernev et al. 2015; Scheibehenne et al. 2010. Mixed.

  13. Use specific, close social proof ("teams like yours"), and never state a bad norm. Goldstein et al. 2008; Schultz et al. 2007. Mixed (real, smaller than claimed).

  14. Make proof checkable: real names, logos with permission, verified reviews with a note on how they are verified. Heath & Heath 2007 (testable credentials); UWG Anhang Nr. 23b/23c, § 5b Abs. 3. Mixed (psychology) / mandatory (law).

  15. Cut friction before adding motivation. State the time cost, shrink the first action, and give one clear CTA per screen. Fogg B=MAP; Wendel CREATE; defaults literature. Mixed.

  16. Describe steps in plain, easy-to-process language; hard-to-read instructions make the task feel harder. Song & Schwarz 2008. Mixed.

  17. Beat the curse of knowledge by testing copy on outsiders. Your sense of clarity is not evidence. Newton 1990 via Heath & Heath; Camerer, Loewenstein & Weber 1989. Robust (phenomenon).

  18. Repeat one core claim consistently across the page and funnel, and only true claims. Dechêne et al. 2010. Robust.

  19. Use real scarcity only, stated plainly. Fake urgency is blacklisted. Barton et al. 2022; UWG Anhang Nr. 7. Mixed (psychology) / mandatory (law).

  20. Keep choice symmetric and avoid dark patterns. Cancel as easy as sign-up, no confirmshaming, no pre-ticked paid add-ons. They convert, which is why they are banned. Luguri & Strahilevitz 2021; Mathur et al. 2019; § 312k BGB; DSA Art. 25 (online platforms); Directive 2023/2673. Robust (effect and legal risk).

Meta-rule: test anything rated mixed or weak before relying on it: an A/B test where traffic allows, otherwise the methods in the low-traffic testing playbook. Published effects shrink by about half on replication (OSC 2015).

5. Do not rely on

The claims not to rely on, with verdicts, are in Myths: Psychology; this note holds the evidence its rows cite.

Verification notes

These figures come from secondary sources, not the primary works:

  • Newton tapper study: her dissertation is not online. The 120, 3 and 50% figures come from Heath & Heath via Wikipedia and SSIR.

  • The Economist decoy figures: taken from Wikipedia's account of Ariely's Predictably Irrational (2008); the book itself was not checked.

Where the research pass could not find independent replication (Song & Schwarz 2008, Mandel & Johnson 2002, validation of the Fogg model), the entry says "as far as I found" or similar; treat those as open, not settled.

Oddlynew Copy and Messaging PrinciplesSign in

Referenced by

and more