What the research supports
A value proposition says what a product does for a person and why they should believe it. Behavioural research is often used to polish that message. Most of it is real. Much of it is smaller and more conditional than the blog version.
This guide covers fourteen ideas. Each one has the same four parts: what it is, what the evidence shows, how products use it, and how to use it honestly. Where the evidence is weaker than the popular claim, the guide says so plainly.
Strongest to weakest
- Strongest. Defaults, and the published minimum sizes for touch targets.
- Strong. Anchoring on numbers people compare against, and framing in risky choices. Both have been repeated and combined across many studies.
- Modest. Reviews and popularity information. The average effect is small, and it is larger at visible thresholds.
- Mixed. Loss aversion (its size is argued over), scarcity cues, the IKEA effect and banner blindness.
- Weak or contested. Decoy options, choice overload, and the idea that a nicer look makes a product easier to use.
- No lift found. Countdown timers, in the one independent test of them.
Nearly all of this work was done in labs, shops and mailings, not on software pages. No independent test was found for most of the patterns on a SaaS pricing page, such as badges, tier counts, timers or logo strips. Treat each idea as a hypothesis to test, not a rule.
An order of work
- Fix truth and access first. Remove any number, deadline, review or “was” price that is not true. Meet the target-size and contrast floors.
- Name the person’s decision and what blocks it. They may not understand the value, may not trust it, may not find the action, or may not be able to finish setup.
- Pick the idea that fits that block. Prefer the stronger one when two fit.
- Write the hypothesis, then test it on real traffic. With low traffic, say the effect cannot be detected.
- Apply one ethical test: would it still work if the person could see how and why it was built?
Value and pricing
These seven ideas shape how worth is judged: what people value, what number they compare against, and how a choice is described.
The IKEA effect
What it is. People value something they built more than the same thing built by someone else. Norton, Mochon and Ariely (2012) found that people paid more for their own folded paper and assembled boxes, and rated them close to expert work.
The evidence. It only works when the build is finished. A failed or unfinished build earns nothing. Children show the effect from about age five, and effort did not change it (Marsh, Kanngiesser and Hood 2018). That weakens the simple story that effort creates love for the result. The studies tested people building a whole object, not taking a quiz or ticking options. No source shows that a short configurator creates the effect.
How products use it. Setup quizzes and product builders are common. No outcome data were found, so it is unknown whether they work through ownership or just through personalisation.
Using it honestly. Let someone finish something small and real early, such as a first saved project, rather than a long setup that may stall. Make sure they can leave with their work.
Anchoring
What it is. A number seen first pulls later estimates towards it. Estate agents were swayed by a listing price as much as students were, although the agents denied it (Northcraft and Neale 1987).
The evidence. For numbers people compare against, it is one of the strongest findings in decision research. A review of 2,601 effect sizes found a large average effect, with wide variation between studies (Schley and Weingarten), and a large replication project repeated the classic results (Klein and others 2014). For random numbers it is contested. Ariely, Loewenstein and Prelec (2003) found large effects. Fudenberg, Levine and Maniadis (2012) found very weak ones for ordinary goods. Nobody has measured it for pricing-page layouts, such as tier order or struck-through prices.
How products use it. Struck-through prices, “save 25%” annual toggles and a high custom tier beside the middle tiers are common. When J.C. Penney dropped reference prices, comparable sales fell 25.2% (JCK 2013). Many things changed at once, so that is a hint, not proof.
Using it honestly. Quote your real full-scope price first, with a reason. If a client names a budget first, treat it as an anchor and ask about scope before you reply. A reference price must be one you have really charged.
The decoy effect
What it is. Adding a clearly worse third option can lift the share of the option it is worse than (Huber, Payne and Puto 1982). Most tiered pricing pages instead rely on the compromise effect, where the middle option gains, and on a “most popular” badge.
The evidence. It is weak outside tidy numeric choices. Across 91 attempts in 23 product classes, only 11 effects were reliable (Yang and Lynn 2014). It largely disappeared when options differed on perception or had been experienced (Frederick, Lee and Baskin 2014). A test with films found exactly zero while meeting the original authors’ conditions (Trendl, Stewart and Mullett 2021). The best-known demonstration, a magazine subscription table, comes from a popular book and a student sample, not a journal. Badges are separate. Showing the five most popular dishes lifted demand 13 to 20% in a restaurant trial, but the popularity was real and the badge design was not tested (Cai, Chen and Fang 2009).
How products use it. Three or four tiers with a badge on a middle tier is the usual pattern. True dominated options are rare.
Using it honestly. Offer one recommended package and give a reason, such as “most clients your size choose this because...”. Keep it true and check it against real sales. A small business with a few enquiries a month cannot detect a decoy effect.
Framing
What it is. The same facts, described differently, can change a choice. In the Asian disease problem, about 72% chose the certain option when it was framed as lives saved, and 22% when it was framed as deaths (Tversky and Kahneman 1981).
The evidence. Framing in risky choices is strong and has been repeated many times (Kühberger 1998, 136 studies). Framing in persuasive messages is not. Across large reviews, gain-framed and loss-framed messages differ by almost nothing (O’Keefe and Jensen). So “loss wording always beats gain wording” is not supported, and neither is the reverse.
How products use it. Prompts such as “keep your spot” and expiring credits are common. None has an independent measurement.
Using it honestly. State what the buyer gets and what staying as they are costs, in their own numbers, once. With enough traffic, test two honest versions of one sentence. Do not assume a frame wins.
Loss aversion
What it is. A loss often feels bigger than an equal gain. The popular line is that people fear losses “twice as much”.
The evidence. A review of 607 estimates from 150 articles found an average of about 1.96 (Brown and others 2024). That is an average over mostly money choices, not a constant. Later analyses put it lower, and its size is still argued over. Gal and Rucker (2018) say the evidence does not show that losses outweigh gains overall. Mrkva and others (2020, 17,720 people) found the effect at every knowledge level, but smaller for experts.
How products use it. Duolingo says it uses loss aversion in streaks. Its blog reports a 0.38% rise in daily learners from two Streak Freezes, and that learners with a seven-day streak are 3.6 times more likely to finish a course (Duolingo 2022). Both figures come from the company, the second is a correlation, and the post does not link either to loss aversion.
Using it honestly. Describe real stakes. Ask of your own habits what you would choose if you were starting today, because a tool you have used for years feels expensive to drop.
Scarcity
What it is. Limited things seem more desirable. There are three kinds: limited supply (“only 3 left”), high demand (“selling fast”) and limited time (“ends Friday”).
The evidence. Cookies in scarce supply were rated more desirable, but those were ratings from 200 people, not purchases (Worchel, Lee and Adewole 1975). A review of 131 studies found that scarcity cues generally raise purchase intention. Demand cues did best for practical goods, supply cues for experiences and time cues for high-involvement purchases (Barton, Zlatevska and Oppewal 2022). The one independent test of a 60-second countdown timer found no significant effect on whether people accepted a paid add-on (Luguri and Strahilevitz 2021). Nothing about scarcity is automatic.
How products use it. Real limits are common and lawful, such as cohort start dates, seat caps and genuine end dates. Fake versions are well documented. A study of about 11,000 shopping sites found 183 using deceptive patterns, including countdown timers that reset on reload and low-stock notices that kept coming back (Mathur and others 2019).
Using it honestly. Use only real limits and write them in text: “two places left in the October cohort”. For a freelancer the real scarcity is capacity, and saying so is honest.
The endowment effect
What it is. People ask more to give up something they hold than they would pay to get it. In the classic mug study, sellers wanted a median of about US$7 and buyers offered about US$3 (Kahneman, Knetsch and Thaler 1990).
The evidence. The gap is solid in the lab. A review found selling prices above buying prices (Yechiam, Ashby and Pachur 2017). The cause is contested. Gal and Rucker (2018) argue it is inertia, not loss aversion. No study measured it in a software trial. The best field evidence is about inertia. In a newspaper experiment, auto-renewing trials produced about 20% more subscriptions and 21% more revenue after four months. Auto-cancel did better after a year, the advantage was gone at two years, and about half of auto-renewers paid for something they did not want to use (Sahni, Miller and Strulov-Shlain). A test of three-day against seven-day trials, with 680,588 users, changed trial adoption and delayed conversion but not immediate conversion (Zhang and Duan 2025).
How products use it. Free trials and freemium plans are often said to “create ownership”. The evidence points more to inertia.
Using it honestly. Put real value in the person’s hands early, such as their own project, not a demo. Judge a trial by revenue kept at six to twelve months, not by sign-ups. Send the end-of-trial reminder yourself.
Trust and defaults
These three ideas are about whom people believe and what they do when they do nothing.
Social proof
What it is. When unsure, people treat what others do and say as evidence. It is strongest when the situation is unclear and the others are similar.
The evidence. Reviews matter, modestly. An extra half star made restaurants sell out 19 points (49%) more often, and more so where other information was scarce (Anderson and Magruder 2012). Across 96 studies the link between reviews and sales averaged 0.091, and the number of reviews counted for more than the average rating (Babić Rosario and others 2016). A single artificial up-vote raised later positive ratings by 32% in a news-site test (Muchnik, Aral and Taylor 2013). Other findings are weaker. A hotel card saying other guests reuse towels lifted reuse from 35.1% to 44.1% (Goldstein, Cialdini and Griskevicius 2008), but a later re-analysis found the pooled towel results inflated (Carlsson and others). Norms can also backfire. Telling households the average pushed low energy users up (Schultz and others 2007). In vendor-partnered data, purchase likelihood peaked at 4.2 to 4.5 stars, not 5.0 (Spiegel Research Center 2021).
How products use it. Logo strips, “join 10,000 teams” counts and “most popular” badges are standard. No independent measurement was found. Lift figures quoted by testing vendors are not planning inputs.
Using it honestly. Use proof that can be checked: named customers with permission, real counts, dated reviews, a mix of ratings. With few reviews, show the honest number. Put a specific quote that answers the visitor’s objection next to the decision. Ask every customer for a review the same way, not only happy ones.
The halo effect
What it is. One good impression leaks into judgements of unrelated traits, often without the judge noticing. Viewers of a warm lecturer rated his accent as appealing and denied that liking him had influenced them (Nisbett and Wilson 1977). In interfaces the form is the aesthetic-usability effect: attractive designs are rated easier to use.
The evidence. The halo exists. The direction of cause is disputed. Across 26 cash-machine layouts and 252 people, appeal matched perceived ease of use more than actual ease of use (Kurosu and Kashimura 1995, reported by Moran 2024). One study found beauty shaped perceived usability (Tractinsky, Katz and Ikar 2000). A later one, with 80 people, found usability shaped perceived beauty and not the reverse (Tuch and others 2012). Polish is a first impression, not a substitute for usability.
How products use it. Polished brand surfaces, big-name customers and press logos borrow the halo. No independently measured case was found.
Using it honestly. Fix first-impression defects first, such as spelling, broken layout and mismatched type. In usability tests, measure task success, not liking. When you judge a vendor or tool, score each criterion on its own before forming an overall view.
Status quo bias and defaults
What it is. People keep what they already have, and accept whatever is pre-selected (Samuelson and Zeckhauser 1988). A default works through effort, through the sense that it is recommended, and through being treated as the current state.
The evidence. This is the most credible effect in this guide. A review of 58 studies and 73,675 people found a medium-to-large effect (d = 0.68), stronger in consumer choices and when the default signals endorsement (Jachimowicz and others 2019). Automatic enrolment raised retirement-plan participation from 37% to 86% for staff with 3 to 15 months of service (Madrian and Shea 2001). Real-world nudges are smaller than published ones (1.4 against 8.7 percentage points across 126 trials), though defaults are among the more credible kinds (DellaVigna and Linos 2022). Arguing that a buyer’s current routine is costly is a different tactic, with no review behind it. It can be honest if the cost is real and you quantify it.
How products use it. Search and browser defaults, notification and permission prompts, auto-renewing subscriptions, pre-selected plans and add-ons, and privacy settings.
Using it honestly. Set the default to what you would choose for a typical person, and make “change this” easy to see. Prefer safe, reversible defaults. A default is a decision made for someone else, so be ready to explain it.
Choice and interface laws
These four are the best known “laws” in interface design. Each was measured on something simpler than a web page.
Hick’s law
What it is. In simple tasks, reaction time rises with the logarithm of the number of equally likely choices (Proctor and Schneider 2018). It is often stretched into “fewer options are better”. A second finding, choice overload, travels with it.
The evidence. Hick’s law holds for reaction time to lamps and keys, with known exceptions. It grows slowly. Cutting a 40-item menu to 35 changes log2(41), about 5.36 bits, to log2(36), about 5.17. For interfaces it is contested. Liu and others (2020) argue the law speaks against “less is better”. Choice overload is shakier. The jam study found 24 jams drew more shoppers, but only about 3% of those who stopped bought, against 30% at a table of 6 (Iyengar and Lepper 2000). That was one shop over two Saturdays. A review of 50 experiments found an average effect of virtually zero (Scheibehenne, Greifeneder and Todd 2010). A later review found overload when sets are complex, tasks are hard or preferences are unclear (Chernev, Böckenholt and Goodman 2015).
How products use it. Pricing pages show two to four tiers, and wizards, quizzes and filters narrow choices. No independent test of tier count was found.
Using it honestly. Use the law as a prompt to structure, not to delete. Group, order predictably, label in the person’s words and add search to long lists. Keep one primary action per screen. Where buyers are unsure, reduce, group and recommend with a reason. Always offer a way to see everything.
Fitts’s law
What it is. The farther away and smaller a target is, the longer it takes to point at, in proportion to the logarithm of distance over width (Fitts 1954).
The evidence. It is strong for aimed pointing. On touch screens the plain law underestimates errors on small targets, and a modified model fitted far better (Bi, Li and Zhai 2013). No independent test was found showing that a bigger button lifts sales. The solid part is the published minimum sizes: 24 by 24 CSS pixels with spacing at WCAG 2.2 level AA (SC 2.5.8), 44 by 44 at level AAA (SC 2.5.5), 44 by 44 points for Apple, and 48 by 48 dp with 8 dp between targets for Android. Spacing matters as much as size, because neighbours cause errors.
How products use it. The real use is platform guidance from Apple, Google and the W3C. These are standards, not conversion claims.
Using it honestly. Make every tappable control at least 44 by 44 pixels, and enlarge hit areas with padding that does not change the look. Put the main action where the thumb or pointer already is, at the end of the content it concludes. Keep close, dismiss and destructive controls away from it.
The Von Restorff effect
What it is. An item that differs from its neighbours is remembered better than the same item in a uniform list (Hunt 1995, on von Restorff’s original work).
The evidence. The memory effect is strong in the lab. Making a button stand out draws on a different field, visual attention, where a single item that differs on a basic feature is found quickly (Treisman and Gelade 1980; Theeuwes 1992). The much-retold test where a red button beat a green one by 21% used about 2,000 visits, and the page was already green, so red was simply the only contrasting item. The original post could not be found (Porter 2011, as retold by Poper). Emphasis also works only while it is rare.
How products use it. Pricing pages isolate the recommended plan with a tag, a border or a filled button while the others are outlined. No data show whether it lifts revenue.
Using it honestly. Give each view one primary action in one accent colour used nowhere else. On a value-proposition page, make the single key claim the odd one out in size and weight. Squint at the page to check, and run a real contrast check, because colour cannot be the only cue.
Banner blindness
What it is. People on a task skip areas that look like adverts, by position (top strips, right-hand rails) or by look. They also skip useful content placed there.
The evidence. The direction has held for two decades. An eye-tracking study with 26 people found users skipped content in ad-like zones, content that looked like an ad wherever it sat, and legitimate content next to ads (Pernice 2018). The size depends on the task. Goal-directed searchers recalled 0.06 banners on average, against 1.00 for aimless browsers (Pagendarm and Schaumburg 2001, 32 students). Blind means not attending during the task, not never looking.
How products use it. Companies mostly work around it, with native ads, “sponsored” results, and product messages placed inside the main modal, not in banner strips.
Using it honestly. Put the value proposition in the main reading flow as plain text. Keep the only copy of anything important, such as price changes, outages and terms, out of announcement bars and ad-shaped boxes. Place trust and pricing details beside the content they qualify.
Where each idea becomes a dark pattern
A technique is honest when the claim is true and the choice stays free and informed. It becomes a dark pattern when it relies on a false claim, a hidden fact, or making the person’s preferred option harder to take than yours. These patterns are common. A sweep of about 11,000 shopping sites found 1,818 instances (Mathur and others 2019). In an experiment, mild patterns lifted acceptance of a paid plan from 11.3% to 25.8%, and aggressive ones to 41.9%, but aggressive ones also caused backlash. A countdown timer alone did nothing (Luguri and Strahilevitz 2021).
This is a summary, not legal advice. Laws change and the rules differ by country, so read the current text for the places you sell in. Where a rule has not started yet, it says so.
Fake urgency and fake scarcity
This is where scarcity, framing and loss aversion turn. Examples are a timer that resets, a deadline that rolls over, invented stock or demand, and “your spot is held” when nothing is held.
- Australia (ACCC). The Australian Consumer Law already bans misleading conduct and false representations (sections 18 and 29). Emma Sleep was penalised A$15m in April 2026, partly for countdown timers that reset. Vuly and James Lane paid infringement notices over Black Friday 2025 deadlines.
- United States (FTC). The FTC Act bans deceptive practices. The FTC’s 2022 staff report on dark patterns lists fake countdown clocks and false low-stock or high-demand messages.
- European Union. The Unfair Commercial Practices Directive bans false “very limited time” claims (Annex I, point 7).
- United Kingdom. The Digital Markets, Competition and Consumers Act bans false limited-time claims. Official guidance (CMA207) names a restarting countdown clock as a false time limit and says a genuine end is unlikely to be a problem.
Fake “was” prices
This is where anchoring turns: a reference price the business never charged.
- Australia. Of Emma Sleep’s 74 products, 58 never sold at the struck-through price and 16 almost never did.
- United States. The FTC’s pricing guides require a former price to be genuine (16 CFR Part 233).
- European Union. A price cut must show the lowest price of the prior 30 days.
- United Kingdom. The government was consulting on rules for “was” prices. Check the current position.
Fake or gated proof, and fake awards
This is where social proof and the halo turn: invented reviews, bought followers, “N people viewing” counters that are not real, hidden bad reviews, and fake “as seen in” logos. No law targets good-looking design. The target is the false claim.
- Australia. Service Seeking paid A$600,000 after businesses wrote their own reviews. HealthEngine paid A$2.9m after about 17,000 negative reviews went unpublished.
- United States. The FTC rule on consumer reviews and testimonials took effect on 21 October 2024. It bans fake and AI-generated reviews, undisclosed insider reviews, review suppression and fake follower counts, with penalties of up to US$53,088 per violation.
- European Union. Annex I, points 23b and 23c, ban claiming reviews are from real users without reasonable checks.
- United Kingdom. Fake and undisclosed-incentive reviews have been banned since 6 April 2025.
Drip pricing and decoys no one can buy
Charm prices are lawful. The problem is a low headline price with mandatory fees added later. A decoy nobody can actually buy is bait advertising.
- Australia. Drip-pricing disclosure rules start on 1 July 2027. The Federal Court found eHarmony misled customers with “from” monthly prices that left out a mandatory fee.
- European Union. Bait advertising is banned (Annex I, points 5 and 6).
- United Kingdom. Drip pricing has been banned since 6 April 2025.
Trials and defaults that trap
This is where defaults, endowment and the IKEA effect turn: pre-ticked boxes, pre-selected paid add-ons, hidden auto-renewal, and cancelling that is harder than joining.
- Australia. Subscription duties on disclosure, reminders and online cancellation start on 1 July 2027.
- United States. Amazon agreed in September 2025 to pay US$1bn plus US$1.5bn in refunds over Prime sign-up and cancellation, and to offer a clear decline button. The federal click-to-cancel rule was struck down in July 2025, and a new rulemaking is under way.
- European Union. A pre-ticked box is not valid consent (the Planet49 case, 2019).
- United Kingdom. A subscription regime is due from January 2027.
Hidden choices and shaming buttons
This is where Hick’s law, Fitts’s law, the Von Restorff effect and banner blindness turn. Examples are a tiny, pale decline button beside a large “keep” button, a cheapest plan buried in clutter, and decline wording written to make the reader feel foolish for saying no. Disguised ads, built to look like part of the interface, belong here too.
- Australia. A general ban on conduct that unreasonably manipulates consumers starts on 1 July 2027. No court has read it yet. The maximum penalty is already the greater of A$100m, three times the benefit, or 30% of adjusted turnover.
- United States. The FTC Act applies, and the Amazon order requires a clear decline option.
- European Union. The Digital Services Act, Article 25, bars online platforms from deceptive or manipulative interfaces. Misleading prominence, repeated requests and hard cancellation are listed as topics for guidelines. It covers platforms only.
- United Kingdom. Rules on aggressive and misleading practices are likely to apply, though no rule names these designs.
Seven rules of practice
- Every number, deadline, stock level, rating and “was” price must be true when it is shown.
- Show the total price, including mandatory fees, with the headline price.
- Make cancelling as easy as joining, and remind people before a trial turns into a charge.
- Never pre-tick consent or paid add-ons.
- Invite reviews from every customer the same way. Disclose incentives. Do not edit, hide or gate reviews.
- Give “yes” and “no” equal weight, in plain words that do not shame.
- In health, therapeutic and finance work, check the sector rules as well.
The shortest test applies to every idea above. Would the design still work if the person could see exactly how and why it was built?