Paper · Websites

Landing page techniques with the strongest evidence remove effort and doubt, and small sites cannot confirm their gains by A/B test

We graded 82 landing page techniques from 192 sources. The strongest remove effort and doubt: a fast first screen, short labelled forms, credible design, a recommended option, a refund promise and terms beside the button. Small sites should test with people before A/B tests.

In Motion Automation research team · 31 October 2026

From above at dusk, one person sits still at a large monitor in a small design studio while colleagues and city traffic streak past.

Abstract

A landing page asks a visitor for one action, such as leaving an email or paying, and its conversion rate is the share of visitors who take it. Advice on raising that rate, called conversion rate optimization (CRO), is plentiful and unevenly supported: several famous findings shrank when repeated, single-site gains overstate typical effects, and some tactics work by deceiving visitors, which US regulators now act against. We read 192 sources in full and graded 82 techniques: A for a controlled experiment or large study, B for consistent practitioner data, C for expert opinion or a published standard. The grade-A techniques remove effort or doubt: a fast first screen (a landing page 31% faster to show its main content sold 8% more in a randomized test), concise and scannable text, short forms with visible labels and checks as each field is left, credible design, a preselected recommended option, a refund promise, and terms beside the button. Anchoring (the pull of the first number seen) and social proof (showing what others chose) have modest effects; decoy plans and the claim that more options deter buyers did not hold up on average. In one experiment, hiding a paid plan’s terms doubled acceptance, from 14.8% to 30.1%; that tactic and seventeen other deceptive ones are excluded. At 100 to 1,000 visitors a week, an A/B test, which splits visitors at random between two versions, detects within three months only lifts of about 18% to 114%, so a site with that traffic should first judge its pages by counting visitors at each step and by usability tests with five people.

Introduction

A visitor who arrives at a web page from a link, an advertisement or a search result decides quickly whether to stay. In an eye-tracking study of 120 people, 57% of viewing time went to the part of a page visible before scrolling, called above the fold, and 74% to the first two screens1. A landing page is a page built for such arrivals that asks for one action: an email address, a click on a priced button, a payment. That action is the conversion, and the conversion rate is the number of visitors who take it divided by the number who arrive2. A funnel is the sequence of steps from first view to the final action, each with its own rate. Conversion rate optimization (CRO) is the practice of changing a page to raise that rate and checking each change with data. Across 41,000 landing pages in one benchmark, the median page converted 6.6% of visitors, and the median finance and insurance page 8.3%3.

Two kinds of knowledge feed CRO. The first comes from user experience (UX) research, the study of how easily and pleasantly people use an interface; its usability measures are task success, time and errors. Usability studies show that people scan rather than read4, that they judge a site’s credibility (whether they believe it) largely by its look5, and that every unneeded form field or unclear error costs time and completions67. The second comes from behavioral economics, the study of systematic patterns in how people choose. Anchoring is the pull of a first number on later estimates8; the default effect is the tendency to keep a preselected option9; social proof is the tendency to follow what others like oneself do; and choice overload is the claim that more options can lower choosing10. The decoy effect is the claim that adding a clearly worse option shifts choices toward the option it resembles. Risk reversal moves the cost of a bad purchase from buyer to seller, through refunds, guarantees and free trials. CRO practice turns these findings into page techniques, and it tests them with a controlled experiment called an A/B test: visitors are split at random between two versions of a page, so that a difference in conversion can be credited to the change.

That practice rests on uneven ground. Some famous findings did not survive repetition: the average choice-overload effect across 50 experiments is close to zero11, and a well-known decoy example failed in two replications with 515 and 2,003 people12. Large companies report that most ideas they test do not win; about a third did at Microsoft and about one in ten at Booking.com13, so a published single win overstates what a page should expect. A 2026 replication of four popular page patterns in randomized tests with millions of users each found their earlier reported effects “highly exaggerated”14. Some tactics also work by deceiving visitors. In a randomized experiment with 3,777 US adults, hiding the terms of a paid plan doubled acceptance, from 14.8% to 30.1% (p. 75)15, and the US Federal Trade Commission (FTC) has listed such tactics as “dark patterns”16 and sued companies over them1718.

A small seller cannot learn this ground by testing. An A/B test needs thousands of visitors per version to detect the modest effects mature programs find1319, and a new site may draw a few hundred a week. It must therefore choose techniques by the strength of outside evidence, avoid the deceptive ones, and measure in ways its traffic can support.

This report asks three questions. Which landing page techniques have the strongest evidence, page area by page area? Which tactics should a page never use, and on what legal or ethical source? How should a site with a few hundred visitors a week measure whether its pages work?

The techniques with grade A share one mechanism: they remove effort or doubt at the moment a visitor decides. Speed, a clear first screen, scannable text, short labelled forms, credible design, a recommended default, a refund promise and visible terms all rest on controlled experiments or large studies. Persuasion effects from the laboratory are smaller and less reliable on real pages, and their deceptive versions are excluded. At a few hundred visitors a week, an A/B test can confirm only very large lifts, so a small site should first judge its pages by counting visitors at each step and by usability tests with five people. The report contributes:

  1. A catalog of 82 techniques in eleven page areas, each with what to do, why it works, an evidence grade and a source with page or section (Section 3); the fifteen strongest are in Table 1.
  2. A record of famous findings that shrank or failed, and of what practitioners rate highest (Section 4).
  3. A list of excluded deceptive tactics with sources and their US legal status on 31 October 2026 (Section 5).
  4. An order of measurement for a site with a few hundred visitors a week (Section 6).

Sources, grades and measures

All sources were fetched and read in full on 31 October 2026: research papers and meta-analyses, usability research, benchmark reports from vendors with stated samples, published case studies, US statutes, rules, court opinions and agency guidance, and posts by CRO practitioners on X and LinkedIn from 2025 and 2026. Sources known only from a search result or another article’s citation are not cited. For each source, working notes record the exact quotes used, with page or section numbers.

Each technique received one grade for the strongest evidence behind it:

  • A: a controlled experiment (a randomized A/B test or laboratory experiment) or a large study (a meta-analysis, or a usability, eye-tracking or observational study with a stated method and sample in the thousands).
  • B: consistent practitioner data: benchmark reports with a stated sample, several published cases that agree, qualitative usability research from an established lab.
  • C: expert opinion, a published standard or rule without effect data, or a single case.

A grade describes the evidence, not the size of the effect. Where a source reports a gain, the report gives the number with its context (site type, sample, design) and does not treat one site’s gain as a rule.

How conversion is measured

A conversion rate needs one action, counted the same way in every period, and a stated denominator. Nielsen advises counting it per page and over windows long enough that random swings at low volume do not mislead2. Form analytics split the denominator further: view-to-completion counts everyone who saw a form, starter-to-completion only those who typed in it. Across 739 forms and about 93 million form views, 34.6% of viewers and 51.7% of starters completed20. Benchmarks differ by these choices. Unbounce’s 6.6% median excludes pages with no conversions and counts clicks as well as form submissions3, so a new page should compare itself with its own earlier weeks before it compares itself with a benchmark.

Page speed is measured by Google’s Core Web Vitals, three field measures taken from real visits: Largest Contentful Paint (LCP), the time until the largest text block or image in view is shown; Interaction to Next Paint (INP), the delay between a tap or click and the screen’s response; and Cumulative Layout Shift (CLS), a score for how much visible content moves unexpectedly. A page passes when 75% of visits meet each threshold: LCP at or under 2.5 seconds, INP at or under 200 milliseconds, CLS at or under 0.121222324.

An A/B test reports whether the difference between versions is statistically significant: unlikely to have arisen by chance if the versions were equal, conventionally at a 5% level. Its power is the chance it detects a real effect of a given size, usually set at 80%. The minimum detectable effect is the smallest difference a test of a given size can detect at that power. For a baseline conversion rate p and an absolute difference δ, a two-version test needs about 16 p(1 − p)/δ² visitors per version at 80% power and a two-sided 5% level (Section 3.2)13; Miller19 gives the same rule. Checking results repeatedly and stopping at the first significant reading inflates false positives: with ten looks, a nominal 5% test needs a reported significance near 1% to keep a true 5% rate19.

Table 1. The fifteen techniques with the strongest evidence and the widest use across page types. ID: row in the catalog of Section 3. Grade: A, controlled experiment or large study; B, consistent practitioner data.
IDTechniqueEvidence in one lineGrade
S1Show the main content within 2.5 s on mobileVodafone’s faster landing page: 8% more sales in a randomized testA
S5Put the offer, proof and main button in the first screen57% of viewing time above the fold (120 people, eye tracking)A
H2Write half the words, scannable, conclusion firstConcise text 58%, scannable 47% better on usability (controlled test)A
F1Ask only for what the next step needsCheckouts need 8 fields, average 11.3; 1-field pages 18.2% vs 4.2% at 9+B
F2Check each field when the visitor leaves itSuccess +22%, completion time −42% in a lab testA
F4Visible label above every field, never a placeholder aloneGuideline set cut completion time and eye fixations in a controlled studyA
C1Name the next step on the button6 of 8 users misread “Get started”B
T1Professional, error-free designDesign look: 46.1% of 2,440 credibility comments, 54.6% on finance sitesA
T4Real reviews and testimonials, early and specific5 reviews vs none: purchase likelihood +270% (one retailer)B
T8Show the method, formula and cited ruleExplanations raised willingness to pay for financial adviceB
K1Calculator result on the page; email as an extraUsers “feel tricked” when results need contact detailsB
P3Preselect or highlight one recommended optionDefault effect d = 0.68 across 58 datasetsA
P8Offer a refund promise at the point of hesitationFree returns raised purchase 31.6% (randomized, 13,481 customers)A
P12Price, renewal and cancel terms next to the buttonHiding terms doubled acceptance, 14.8% to 30.1%A
Q6At low traffic, test with five people before A/B testsFive users find about 85% of usability problemsA

Technique catalog by page area

The catalog runs from the first screen down the page and on to measurement. Each area opens with what its evidence shows, then lists its techniques (Tables 2 to 12). Figure 1 shows how the grades fall across areas: pricing, testing, forms and trust hold most of the techniques backed by experiments, while calls to action, calculators, lead magnets and mobile have none at grade A.

Figure 1. Evidence grade of each technique in the catalog, by page area. Each technique is counted once at its grade.

Speed and the first screen

Two randomized tests on landing pages with no visible difference between versions tie speed to sales. Vodafone’s version with a 31% better LCP (5.7 against 8.3 seconds) sold 8% more, with 15% more leads per visit; each version drew about 34,000 visits a day25. Rakuten 24’s faster version raised conversion by 33.13% and revenue per visitor by 53.37% over a month26. Neither reports a significance test for the business measures. A regression over 37 brands found that a site 0.1 seconds faster on mobile went with 8.4% more conversions in retail and 10.1% in travel; in lead generation, the sector closest to a sign-up page, the overall rate fell 1.9% while the step from the first form page to submission rose 21.6% (pp. 16 to 26)27 (Figure 2). That study was commissioned by Google, measured speed in the lab and published only significant coefficients. The widely repeated claim that 53% of mobile visits end after three seconds rests on one line about a 2016 opt-in sample of 3,700 sites28.

Figure 2. Change in mobile conversion, and in the funnel step next to the action, associated with a 0.1-second faster mobile site, by sector. Brands per sector in brackets.
Table 2. Speed and the first screen.
IDTechnique: what to doWhy it worksGradeSource
S1Show the main content fast: LCP at or under 2.5 s for 75% of mobile visits. Render the first screen on the server, size and preload the hero image, defer third-party scripts.Faster versions sold more in two randomized landing page tests (+8% sales; +33.13% conversion).Aweb.dev25; Lee et al.26; Walton21
S2Make the page that holds the form or payment the fastest, not only the home page.In 37 brands, the largest speed-linked gains sat at the step beside the action: product page to basket +9.1%, form step 1 to submission +21.6% per 0.1 s.BDeloitte Digital and Fifty-Five (pp. 16, 25)27
S3Stop layout shift: give every image a width and height, reserve space for late content and for results, keep CLS at or under 0.1.Content that moves pushes buttons away from the finger; the tested Rakuten version cut CLS by 92.72%.CMihajlija and Walton24; Lee et al.26
S4Respond to every tap within 0.1 s, show a working state after 1 s and progress after 10 s; keep INP at or under 200 ms.0.1 s feels instant, 1 s keeps the flow of thought, 10 s is the limit of attention; redBus sold 7% more after cutting INP (before and after).CNielsen29; Wagner and Pollard23; Kumar and Rajpal30
S5Put the promise, one line of proof and the main button in the first screen; detail below.57% of viewing time falls above the fold, 74% in the first two screens (120 people, eye tracking).AFessenden (Study Results)1
S6Let the next section show at the bottom of the first screen, so the page does not look finished.People stop scrolling at a false floor.CFessenden (Implications)1
S7Measure speed on real visits (field data at the 75th percentile), not only in lab tools.Lab tools miss INP and most layout shift.CWalton21; Rajpal and Gopalakrishnan31

Headline and offer

The strongest evidence on page text is old and holds. Seventy-nine percent of test users scanned new pages rather than reading word by word4. In a controlled test with 51 people, a concise version of a page scored 58% better on a usability composite, a scannable version 47% and an objective version 27%, and all three together 124%32; the measure was usability, not conversion. Eye tracking of 232 people showed attention concentrated on the left edge and the first lines, the F-pattern, which the authors read as a sign of unformatted text rather than a layout to copy3334. In Unbounce’s data, pages written at a fifth- to seventh-grade reading level had a median conversion rate of 11.1%, against 5.3% for professional-level text, an observational difference in which finance was the one industry where harder text did better3. The advice practitioners repeat most, a first screen that says what the product is, who it is for and what happens on a click, has no experiment behind it but appears in 17 of the 51 posts and teardowns kept for this report (Section 4.2).

Table 3. Headline and offer.
IDTechnique: what to doWhy it worksGradeSource
H1Say in the first screen what the product is, who it is for and what happens on a click.Visitors judge in seconds whether a page can help them; the most repeated practitioner advice.CDry35; Copy Hackers36
H2Write half the words: lead with the conclusion, use meaningful subheads and bullets, one idea per paragraph.Concise text scored 58% and scannable layout 47% better on usability in a controlled test; 79% of users scan.AMorkes and Nielsen (Study 3)32; Nielsen4
H3Start headings, bullets and paragraphs with the two words that carry the meaning.Eye tracking of 232 people: fixations concentrate on the left edge and first lines.ANielsen33; Pernice (Antidotes)34
H4State facts and numbers; drop boasts.Objective text raised satisfaction significantly; hype costs credibility.BMorkes and Nielsen (Study 3)32; Fogg (guideline 9)37
H5Write in plain words, at about a fifth- to seventh-grade reading level.Median conversion 11.1% at that level against 5.3% at professional level (observational; finance an exception).BUnbounce3
H6Lead with the one thing only this product does; mention features every rival has briefly.In 323 business-software leader homepages, 58% said “AI-powered”; sameness gives no reason to choose.CLaja38; Laja39
H7Name the audience and situation in plain words (“for small accounting firms”).Visitors look first for whether the page is for someone like them.CWolf40

Calls to action

A call to action (CTA) is the button or link that asks for the conversion. The best evidence concerns its words. In a Nielsen Norman Group study, six of eight users clicked a “Get started” button expecting information and landed in a sign-up flow41. Removing site navigation from landing pages raised conversion by 16% and 28% on HubSpot’s mid-funnel pages and by 0 to 4% on free-content pages42; a widely cited 100% gain from one retailer is a single vendor case43. Practitioners disagree on one button against two: a page fed by one source fits one action, and a page with mixed visitors may carry a second, low-commitment action44. A reported 90% gain from changing “your” to “my” on a button comes from one page with no sample or significance given45, and is not used here.

Table 4. Calls to action.
IDTechnique: what to doWhy it worksGradeSource
C1Name the next step on the button: “See my result”, “Request a quote”, never “Get started” or “Learn more”.A button is a promise; generic labels drew users into flows they did not expect.BHarley and Flaherty (Conclusion)41; Hill46
C2One primary action per page; add a second, low-commitment action only where visitors arrive with mixed intent.Fewer competing choices; mixed traffic has mixed intent.CHill44; Dry35
C3Remove site navigation and footer links from offer pages (pricing, preorder, download).Fewer exits: +16% and +28% on mid-funnel pages, 0 to 4% on free-content pages.BUrban42; Chopra43
C4Put one risk-reducing line under the button: what it costs, what is charged today, how to cancel.Visitors hesitate at the commitment point; the line answers the question there.CHill46; Dry35
C5Make the button look like a button and stand out from the page by contrast.Visitors who have decided must find the action at once.CAagaard45
C6Repeat the main button after the proof on long pages.Late deciders act where they finish reading.CDry35

Trust and proof

The largest credibility study asked 2,684 people to rate the credibility of live websites and coded 2,440 of their comments. Design look was named in 46.1% of comments overall and in 54.6% for finance sites, where visitors also commented more often on the company’s motive (21.0% against 15.5%) and reputation (21.8% against 14.1%) (pp. 6, 23, 61)5. For reviews, Northwestern’s Spiegel Research Center found that a product with five reviews was 270% more likely to be bought than one with none, that the lift was larger for higher-priced items (380% against 190%), and that ratings between 4.0 and 4.7 sold better than a perfect 5.0 (pp. 6 to 16)47; the data are observational and come from retailers. In a CXL eye-tracking and survey study, testimonials with photos and logos of well-known customers were remembered better48. Social proof as a statement about what most people do has a real but small effect: five hotel-towel replications each failed alone, and pooled they show about six percentage points4950. For money tools, a randomized experiment with 449 people found that short explanations of how a financial adviser algorithm reached its advice raised willingness to pay for it51.

Table 5. Trust and proof.
IDTechnique: what to doWhy it worksGradeSource
T1Keep design professional and error-free: consistent type, no typos, broken links or layout faults.Design look was the most named credibility cue (46.1%; 54.6% for finance).AFogg et al. (pp. 6, 23)5
T2Show the real organization: a named founder with credentials, a postal address, a contact, and how the business earns money.Finance-site visitors commented more on motive and reputation.AFogg et al. (p. 61)5; Fogg (guidelines 2 to 5)37
T3Show price, what is included and refund terms on the page.Users rule out sites that hide basic information.BHarley52; Loranger53
T4Collect the first few real reviews or testimonials; show name, photo and role; keep imperfect ratings; link outside profiles.First reviews carry most of the lift; 4.0 to 4.7 beats 5.0; photo testimonials are remembered better.BCollinger et al. (pp. 6 to 16)47; Labay48
T5Hide social proof while numbers are small (a waitlist of twelve).A small count signals that few people want it.CCardello54
T6State true norms (“most customers who tried it…”) and expect a small effect.Pooled towel studies: about six points; single replications failed.AScheibehenne et al.49; Bohner and Schlüter50
T7Near payment, enclose the card fields, show a padlock and at most two recognisable seals; claim only what can be backed.19% of US shoppers abandoned a checkout because they did not trust the site with their card.BBaymard Institute55; Holst56; Holst57
T8Beside each number, show the formula, the inputs, the rule cited (with its section) and a last-reviewed date.Verifiable sources raise credibility; explanations raised willingness to pay for financial advice.BFogg (guidelines 1, 3)37; Kohler (item 11)58; Ben David et al.51

Forms and email capture

Form research is the most consistent area of the catalog. Wroblewski’s laboratory test with 22 people found that checking each field as it was left, against checking only on submit, raised success by 22%, cut errors by 22% and cut completion time by 42%; checking while people were still typing did worst6. A controlled study of 20 form guidelines (visible labels above fields, format rules stated in advance, errors placed next to the field) cut completion time and eye fixations759. Baymard finds that most checkouts need 8 fields but show 11.3 on average60, and that 18% of US shoppers abandoned a purchase because an account was required55. The raw count of fields predicts less than it seems (Figure 3): pages with one field converted 18.2% and pages with nine or more 4.2%61, but one-field forms tend to carry low-commitment offers, and a form analytics vendor’s own pages disagree on whether field count matters2062. Two famous cases are weaker than their reputation: a 120% gain from cutting a contact form from 11 to 4 fields compared 10 with 26 submissions six months apart63, and a $12 million gain from removing one field at Expedia is an executive’s remark at a conference64.

Figure 3. Left: landing page conversion by number of form fields. Right: share of people who started a form and submitted it, by the form’s purpose and device.
Table 6. Forms and email capture.
IDTechnique: what to doWhy it worksGradeSource
F1Ask only for what the next step needs: an email for a waitlist; collect the rest later.Each field costs reading and decision time; checkouts need 8 fields and show 11.3.BScott60; NP Digital61
F2Check each field when the visitor leaves it; once an error shows, re-check as they type and clear it when fixed.Success +22%, errors −22%, time −42% against checking on submit (lab, 22 people).AWroblewski6; Scott65
F3Never show an error on a field the visitor has only just entered.Checking while people typed did worst; early errors read as accusations.AWroblewski6; Scott65
F4Put a visible label above every field; never use placeholder text as the only label.Labels stay visible for checking and fixing; part of the guideline set that cut time and fixations.ASeckler et al.7; Sherwin66
F5On forms longer than a few fields, mark required fields and mark optional ones “(optional)”.When only optional fields were marked, 32% of test users hit a validation error.BScott67; Budiu68
F6State format rules (for example, how to enter a percentage) before the visitor types.The largest single effect in the guideline study’s interviews.ASeckler et al.7; Bargas-Avila and Seckler59
F7Place each error next to its field, say how to fix it, and use colour, icon and text together.Visitors see the message where they fix it; colour-blind visitors still see it.BKrause69; Seckler et al.7
F8Never require an account before payment or a result; offer it afterwards.18% of US shoppers abandoned because an account was required.BBaymard Institute55
F9Keep short forms on one step; group six or more fields into steps with a Continue button.Layout made no difference to success in a lab test; grouped steps were fastest.CWroblewski70; Zuko Analytics20
F10If a popup asks for email, show it after the visitor has read, and offer something tied to the page.Average popup 3.09%, top tenth 9.28%; only 8% of the top tenth appeared in the first 4 seconds.BSumo71
F11Say beside the field what the email will be used for, and use it only for that.Collecting leads under one stated purpose and using them for another is a listed deceptive pattern.CFederal Trade Commission (pp. 18 to 19)16

Pricing pages

Pricing pages draw on the behavioral findings, and those findings vary in strength. Defaults hold up best: a meta-analysis of 58 datasets with 73,675 people found an average effect of d = 0.68, about 27 percentage points on yes-or-no choices, stronger when the default reads as a recommendation72. Anchoring on willingness to pay has a moderate average effect (r = 0.267, which the authors call an upper bound), larger for meaningful anchors than random ones73; the famous social security number auction74 was weak in a direct replication75. Choice overload averages close to zero (D = 0.02 across 50 experiments)11 and appears only when options are complex, the task is hard and buyers are unsure of what they want76. The decoy effect held in 4 of 5 studies where every attribute was a number and in none of 27 where an attribute had to be experienced or seen12. For risk reversal, a randomized field test with 13,481 shoppers at a Japanese fashion retailer found that offering free returns late in the purchase raised the purchase rate by 31.6%, most for first-time buyers and higher-priced items77. For trials, sites that required a card converted 39.0% of trial users to paid against 12.0% without one, across 803 sites, though card-required trials start far less often7879. Business buyers in usability studies left for competitors when a price was hidden53.

Table 7. Pricing pages and offers.
IDTechnique: what to doWhy it worksGradeSource
P1Publish the price on the page.Business buyers left for competitors when prices were hidden; hidden prices read as evasive.CLoranger53; Laja80
P2Offer three to five plans, as columns, with the same attributes in every row and headers that stay visible.Overload appears with complex options and unsure buyers; consistent rows make differences scannable.AChernev et al.76; Scheibehenne et al.11; Moran and Dykes81
P3Highlight one recommended plan, preselect its billing period, and say why it fits most buyers.Defaults move choices (d = 0.68), most when they read as advice.AJachimowicz et al.72; Johnson and Goldstein9
P4Show a meaningful reference price next to the offer: the later price beside a founding price, or the cost of the current alternative.Estimates drift toward the first number seen; meaningful anchors work better than random ones.ATversky and Kahneman8; Li et al. (pp. 1, 6)73
P5Place the plan most buyers should consider in the first two columns.The first two plan positions drew the most attention in eye tracking.BLabay82; Labay83
P6Offer monthly and annual billing; prompt the annual switch early.Annual plans kept more revenue from existing accounts after cancellations and upgrades (88% against 76%, observational); upgrades peak in month two.BChartMogul84
P7End prices in 9 ($99, not $100).Shoppers treat a 99 ending as 15 to 25 cents lower on $5 grocery items; untested for business software.CStrulov-Shlain85
P8Offer a refund promise at the point of hesitation, with the window stated plainly.Free returns raised purchase 31.6% in a randomized field test, most for first-time buyers.ALuo et al.77; RevenueCat86
P9Choose the trial on purpose: card required for fewer, likelier buyers; no card for more starts.39.0% against 12.0% trial-to-paid across 803 sites.BRecurly78; First Page Sage79
P10Do not rely on a decoy plan; if used, every attribute must be a number and the decoy plainly worse.The famous decoy failed with 515 and 2,003 people; decoys fail when attributes are experienced.AFrederick et al. (p. 497)12; Huber et al.87; Ariely88
P11Run a founding offer only with a real cap and a real close date, and show the later price beside it.Backers cluster in the first and last weeks of a dated campaign; a false cap is deceptive.CKuppuswamy and Bayus89; Unlock SaaS90
P12Put price, renewal, lock period and how to cancel next to the button, before card entry; ask for separate consent to recurring charges.Hiding terms doubled acceptance in an experiment; US law requires disclosure and consent before billing.ALuguri and Strahilevitz (p. 74)15; Federal Trade Commission (p. 11)16; Federal Trade Commission91

Interactive calculators

A calculator page offers a free answer and asks for something in return. Nielsen Norman Group’s research on calculators and quizzes found that users “want to try calculators without having to commit to an organization”, and that they are “particularly annoyed when they must submit contact information to get their results at the end of a calculator… people often feel tricked and annoyed” (Users Don’t Want to Share Contact Information)92. Every participant offered a guest option on one tool took it (item 4)58. Users also run calculators backward, from a target to the inputs that reach it, and experienced users judge outputs harshly92. Marketers in a vendor-sponsored survey rated calculators most effective in the middle of the funnel, when buyers compare options93; no study with a stated sample measured calculator conversion.

Table 8. Interactive calculators.
IDTechnique: what to doWhy it worksGradeSource
K1Show the result on the page, updating as inputs change; offer the full report by email as an extra.Users feel tricked when results need contact details; all participants offered a guest option took it.BKohler92; Kohler (items 4, 5)58
K2Ask only for essential inputs, say why each is asked, give an example beside it.Users enter data readily when they see how it changes the result.BKohler (items 6 to 8)58
K3Use honest default values, or none, and label any example values as examples.Defaults shape what users think is reasonable.BKohler (item 10)58
K4Explain the output in context (what the number means for the visitor’s decision) and show the formula behind it.Transparent calculation raises trust and understanding.BKohler (item 11)58; Ben David et al.51
K5Support backward use: let visitors set a target and see which inputs reach it.Users run calculators forward and backward.BKohler92
K6Give each calculator its own page, titled with the words people search for, with the tool on the page rather than in a popup.Users open calculators straight from search results, even from unfamiliar organizations.BKohler (item 1)58
K7Treat a calculator as a middle-of-funnel offer: ask for an email after the visitor has seen value.Marketers rate calculators most effective at the comparison stage.CContent Marketing Institute and ion interactive93
K8Let visitors use the working tool at the top of the page instead of describing it.One software company’s test of the product as its homepage doubled conversions (one site).CDry94

Lead magnets and follow-up email

A lead magnet is something free and useful a visitor receives in exchange for an email address, such as a guide or a sample report; email capture is the form that asks for the address. Evidence here is benchmark data. Welcome emails, sent automatically right after sign-up, draw the highest engagement of regular sends, though benchmarks disagree on the level: 83.63% opened in GetResponse’s data and 35.53% in Omnisend’s9596. In NN/g’s newsletter studies, users valued a low-effort sign-up over explanations, and most ignored unwanted email rather than unsubscribing97. Double opt-in, which confirms an address with a second click, trades some sign-ups for a cleaner list; no controlled comparison was found, and 6.28% of financial-services senders in one benchmark used it9598.

Table 9. Lead magnets and follow-up email.
IDTechnique: what to doWhy it worksGradeSource
L1Send the welcome email at once, with the promised file or result in it.Welcome emails draw the highest engagement of regular sends.BGetResponse95; Omnisend96
L2Ask for the email alone; keep sign-up to one field and one click.Users value low-effort sign-up over explanations.CFlaherty97; Scott60
L3Show the cover and one inside page of the magnet beside the form, and say what it covers.Visitors trade an address only for something they can picture; it also fixes what the address is used for (F11).CHarley52; Federal Trade Commission (pp. 18 to 19)16
L4Use single opt-in by default; switch to double opt-in if bad addresses or spam complaints rise.Fewer steps against verified addresses; no controlled comparison found.CMailchimp98; GetResponse95
L5Send follow-ups tied to what the person asked for, not a general newsletter.Relevant, timely emails drew the most positive responses.CFlaherty97
L6In every marketing email: a postal address, a visible opt-out honored within 10 business days, a truthful sender and subject line; keep result emails transactional.Required by the CAN-SPAM Act, with no exception for business recipients.CFederal Trade Commission99

Mobile

Most mobile evidence concerns the hand and the keyboard. The accessibility standard for the web, the Web Content Accessibility Guidelines (WCAG), sets some mobile rules too (Section 3.10). Fewer people finish forms on phones: 47.5% of mobile starters against 54.5% on desktop across 739 forms20. Baymard’s benchmark of mobile sites found that most still opened the wrong keyboard for some fields, and in testing, auto-correct silently changed email addresses100.

Table 10. Mobile.
IDTechnique: what to doWhy it worksGradeSource
M1Make tap targets at least 1 cm by 1 cm, spaced apart; never below 24 by 24 CSS pixels.Small or crowded targets cause slow and wrong taps.BHarley (Size Matters)101; W3C Web Accessibility Initiative (2.5.8)102
M2Open the right keyboard: type="email", type="tel", inputmode="decimal" for money and rates; turn off auto-correct and auto-capitalize on email and names.Numeric keys are larger targets; auto-correct corrupted addresses in testing.BHolst100
M3Design forms for the phone first and test them on a phone.Mobile starters finish 7 points less often than desktop starters.BZuko Analytics20
M4Keep sticky bars and cookie banners from covering the field or button in focus.Required by WCAG 2.2 success criterion 2.4.11.CW3C Web Accessibility Initiative (2.4.11)102

Accessibility

Accessibility is the degree to which people with disabilities can use a page. WCAG is published by the World Wide Web Consortium; version 2.2 added criteria for target size, focus that sticky elements must not hide, sign-in without memory tests, not asking for the same information twice, and help kept in the same place102. WebAIM’s scan of the top million home pages in 2026 found a detected failure on 95.9% of them: low-contrast text on 83.9%, images without alternative text on 53.1%, form inputs without labels on 51% and empty links on 46.3%103. Those failures are also conversion failures, since an unlabeled field or an unreadable button stops every visitor who cannot use it.

Table 11. Accessibility.
IDTechnique: what to doWhy it worksGradeSource
X1Tie a visible label to every input in code.Unlabeled inputs appear on 51% of top home pages and stop screen-reader users.AWebAIM (Form Labeling)103
X2Meet WCAG AA contrast for text, including fine print; give meaningful images alternative text; give every link and button a name.These failures, with missing labels and page language, make up 96% of detected errors.AWebAIM (WCAG Conformance)103
X3Never ask for the same information twice in one flow; carry the email from the calculator into the waitlist.Required by WCAG 2.2 criterion 3.3.7.CW3C Web Accessibility Initiative (3.3.7)102
X4Let people paste and use password managers, or sign in by email link; avoid puzzle tests.Required by criterion 3.3.8; pages with a CAPTCHA (a puzzle meant to tell people from automated programs) averaged 7.7 more errors.CW3C Web Accessibility Initiative (3.3.8)102; WebAIM103
X5Keep help and contact links in the same place on every page.Required by criterion 3.2.6.CW3C Web Accessibility Initiative (3.2.6)102

The only WCAG version named in a US rule is 2.1 at level AA, for state and local government websites, with compliance dates moved to 2027 and 2028104. The Department of Justice holds that the Americans with Disabilities Act covers the online offerings of businesses open to the public, without a technical standard for them105.

Measurement and testing

The experimentation literature gives small sites two warnings. Most ideas fail: about a third of tested ideas improved their target measure at Microsoft, 15% at Bing, about 10% at Booking.com and 8% in Airbnb search (Table 2)13. Thomke reports that Booking.com runs about 25,000 tests a year and that for every test that succeeds nearly ten do not106. At a 10% success rate and 80% power, 22% of significant wins are false; at 20% power, 52.9% are (Table 3)13. Of 650 tests contributed to GoodUI, 169 won, 286 were positive but not significant, 152 negative but not significant and 43 lost107. Big surprises are usually errors, a rule called Twyman’s law (pp. 3 to 5)108. The second warning is size (Figure 4): at the traffic a new site can expect, a test can confirm only very large effects. Five-person usability tests are the counterweight: under Nielsen’s model, five users find about 85% of the usability problems in a design109. Other small tests need no traffic split either: a five-second test shows a page for five seconds and asks what it offers, a first-click test asks where a person would click first, and a message test asks target buyers to rate the wording110111.

Figure 4. Smallest relative lift a two-version A/B test can detect, by weekly traffic and test length, for a 2% and an 8% baseline conversion rate.
Table 12. Measurement and testing.
IDTechnique: what to doWhy it worksGradeSource
Q1Define one conversion event per page and count it the same way every week; record each funnel step and form starts as well as completions.At low volume, rates swing; steps show where people stop.CNielsen2; Zuko Analytics20
Q2Size every A/B test before it starts (16 p(1 − p)/δ² visitors per version), fix the length and do not stop at the first significant reading.Repeated looks inflate false positives; underpowered tests produce false wins.AKohavi et al. (Section 3.2)13; Miller19
Q3Treat a win as provisional; at low success rates, ask for stronger evidence before acting.About 1 in 3 ideas won at Microsoft and about 1 in 10 at Booking.com; at 80% power, 22% of wins can be false.AKohavi et al. (Tables 2, 3)13; Thomke106
Q4Recheck any surprisingly large lift for errors before acting on it.Large effects are usually bugs or data errors (Twyman’s law).AKohavi et al. (pp. 3 to 5)108
Q5Treat published single-site lifts as upper bounds, not forecasts.High-powered replications found popular pattern effects highly exaggerated.AKohavi14; Linowski107
Q6At low traffic, test in rounds of five people from the target group, and add five-second, first-click and message tests before any A/B test.Five users find about 85% of usability problems; A/B tests need thousands per version.ANielsen109; Chan110; Speero112; Wynter111
Q7Count only visitors who reached the changed step when analysing a test.Removing unexposed visitors cuts noise; in one example it roughly halved the sample needed.AKohavi et al. (Section 3.2)113
Q8Drop any variant that wins through a false or hidden claim.The FTC treats a tested false claim as deception.CFederal Trade Commission (p. 3)16

Findings that shrank or failed, and what practitioners rate highest

Famous findings weaker than their reputation

Table 13 lists the findings most often repeated in CRO advice that did not hold up as stated. Each still appears in practitioner posts read for this report.

Table 13. Famous findings and what the fuller record shows.
Finding as usually toldFuller recordSource
Six jams sell more than 24Average choice-overload effect across 50 experiments is about zero; it appears only under specific conditionsIyengar and Lepper10; Scheibehenne et al.11; Chernev et al.76
A decoy subscription doubles the premium choiceFailed in replications with 515 and 2,003 people; decoys work with numeric attributes onlyAriely88; Frederick et al. (p. 497)12
Random numbers anchor willingness to payWeak in a direct replication; meta-analytic r = 0.267, an upper boundAriely et al.74; Fudenberg et al.75; Li et al.73
“Most guests reuse towels” raises reuseEach of five replications failed alone; pooled effect about six pointsBohner and Schlüter50; Scheibehenne et al.49
Faster pages always convert moreLead-generation sites: overall mobile conversion −1.9% per 0.1 s, form step +21.6%; desktop results contradictory and unpublishedDeloitte Digital and Fifty-Five (pp. 24 to 26, 47)27
53% of mobile visits end after 3 sOne line about a 2016 opt-in sample; the bounce curve comes from an unpublished modelAn28
11 to 4 fields raised conversion 120%One site, 10 against 26 submissions, six months apartMoloney63
One field cost Expedia $12 millionAn executive’s conference remark; no test designHeath64
“My” beats “your” on buttons by 90%One landing page, no sample or significanceAagaard45
Removing navigation doubles conversionOne vendor case; HubSpot’s own tests: 0 to 28% by page typeChopra43; Urban42
Published pattern lifts transferReplications with millions of users found them highly exaggeratedKohavi14

What practitioners rate highest

Posts on X and LinkedIn from 2025 and 2026 by CRO practitioners, together with published landing page teardowns, give a view of current practice that the grades above do not. Of 51 posts and teardowns kept after removing promotions and posts without a claim, the most recommended tactics were a clear first screen (17), specific social proof (12), answering objections and risk (11), stating what differs from rivals (10), using customers’ own words (9), saying who the page is for (7), showing the product working (6), specific button text (6), showing the price (4), repeating the button (4) and a short capture form (4). These counts measure how often a tactic is recommended, not how well it works. Nearly all are changes to the message rather than to layout, and they agree with the grade-A techniques on clarity and scannable text. The one tactic in these posts that the evidence contradicts is the reliance on borrowed lifts: the only experimental result among them is a replication that found popular layout patterns far weaker than claimed14. One popular post recommended a moving testimonial carousel “so users think there’s way more”; that suggestion is deceptive and excluded.

Deceptive tactics excluded

A deceptive tactic, called a dark pattern in the FTC’s 2022 staff report, is a design choice that leads people to act against their own interest by tricking or pressuring them16. Such tactics raise acceptance. In a randomized experiment, hidden terms doubled acceptance of a paid plan, confirmshaming (a decline link worded to shame) raised it from 14.8% to 19.6%, and making refusal harder raised it to 23.6%; false scarcity had no effect (Figure 5). Mild versions raised sign-ups without upsetting people, so a seller cannot rely on visitors’ anger to expose them (p. 66)15. A crawl of about 11,000 shopping sites found 1,818 instances on 1,254 sites, including 157 deceptive countdown timers on 140 sites114. Table 14 lists the tactics this catalog excludes, with each one’s source and status.

Figure 5. Acceptance of a paid plan by interface in a randomized experiment. Grey: control; red: deceptive or pressuring variant.
Table 14. Deceptive tactics excluded from the catalog. Status: rule means a binding US rule or statute names it; enforced, the FTC has sued or settled over it (complaints are allegations); listed, named by FTC staff as a dark pattern and deceptive under Section 5 of the FTC Act when the claim is false. ROSCA: Restore Online Shoppers’ Confidence Act.
TacticWhat it isSourceStatus
Fake countdown timerA clock that resets, or ends while the offer staysFederal Trade Commission (pp. 4, 22)16; Mathur et al.114Listed
False deadline“Ends soon” with no real endFederal Trade Commission (p. 22)16Listed
False scarcityLow stock or few seats claimed when untrueFederal Trade Commission (pp. 4, 21)16Listed
Fake activity messagesInvented “3 people are viewing” or “just signed up”Federal Trade Commission (pp. 4, 21)16; Mathur et al.114Listed
Fake or AI-written reviews and testimonialsFrom people who do not exist or never used the productFederal Trade Commission (§ 465.2)115Rule; civil penalties
Untypical results as testimonyA standout result shown without what customers generally getFederal Trade Commission (§ 255.2(b))116FTC guides
Undisclosed insider praiseStaff, owners or relatives reviewing without saying soFederal Trade Commission (§ 465.5)115Rule
Paid or filtered reviewsRewards for positive reviews; asking only happy customers; hiding bad onesFederal Trade Commission (§§ 465.4, 465.7)115; Federal Trade Commission117Rule
Fake social indicatorsBought followers, views or likesFederal Trade Commission (§ 465.8)115Rule
Hidden subscriptionA one-time or free offer that quietly renewsFederal Trade Commission (pp. 11 to 12)16; Federal Trade Commission91Rule (ROSCA); enforced
Hard to cancelOnline sign-up but cancel by phone, mazes, forced reasonsFederal Trade Commission (pp. 12 to 15)16; Federal Trade Commission17; Federal Trade Commission18Rule (ROSCA); enforced
Hidden termsKey terms in fine print, behind links or on hoverFederal Trade Commission (pp. 7, 14)16; Luguri and Strahilevitz (p. 74)15Enforced
Drip pricingMandatory fees shown only at checkoutFederal Trade Commission (pp. 8 to 9)16; Federal Trade Commission118Rule for tickets and lodging; Section 5 elsewhere
Pre-checked paid extrasAdding paid items by defaultFederal Trade Commission (pp. 22, 25)16Enforced
Confirmshaming“No, I don’t want to save money”Federal Trade Commission (p. 25)16; Federal Trade Commission17Listed; barred in the Amazon order
False hierarchy and trick questionsBright button for the seller’s choice, grey link for the other; double negativesFederal Trade Commission (pp. 18, 25)16Listed
Disguised ads and fake independent rankingsAn ad posing as news or a neutral review siteFederal Trade Commission (pp. 5 to 6, 23 to 24)16; Federal Trade Commission (§ 465.6)115Rule; enforced
Deceptive email headers or subjectsA sender or subject line that misleadsFederal Trade Commission99Statute (CAN-SPAM)

This subsection states the status of the rules a subscription landing page meets; it is a summary of the texts read, not legal advice. The FTC’s amended Negative Option Rule of 2024, known as click-to-cancel, was vacated by the Eighth Circuit on 8 July 2025 on procedural grounds119. The FTC restored the earlier rule text, asked for public comment on a new rule from 13 March to 13 April 202691, and had proposed no new rule by 31 October 2026120. Enforcement continues under the Restore Online Shoppers’ Confidence Act (ROSCA), which requires clear disclosure of terms before billing information is taken, express informed consent and a simple way to stop recurring charges (pp. 11, 14)16. Amazon settled in September 2025 for a $1 billion penalty and $1.5 billion in refunds17. On 29 October 2026 the FTC and eight states sued Toast, whose customers are restaurants, alleging hidden fees and hard cancellation; the complaint applies ROSCA to business buyers, which Toast disputes18. California’s automatic renewal law, amended from 1 July 2025, requires terms next to the consent button, online cancellation and records of consent, and covers purchases for personal, family or household purposes121; whether it reaches sales to businesses is a reading of that scope, not a ruling. The reviews rule has been in force since 21 October 2024115. The FTC’s mail order rule covers merchandise and its guide states that it “does not cover services”122; its guidance on taking orders for products not yet made (state that the product is planned and may not ship, tell buyers within a reasonable time if it will not, refund promptly) gives a usable standard for preorder pages and for fake-door pages, which offer a product before it exists to measure demand. The CAN-SPAM Act applies to commercial email with no exception for business recipients, with penalties of up to $53,088 per violating email99. The fees rule of May 2025 covers only live-event tickets and short-term lodging118.

Measuring a page at low traffic

Suppose a new site draws tens to hundreds of visitors a week, from direct links and, once search engines list its pages, from search. Figure 4 shows what that traffic allows. With 500 visitors a week for twelve weeks, a two-version test on an 8% sign-up rate detects only a relative lift of about 25%; on a 2% payment rate, about 51%. At 100 visitors a week, a 2% rate needs a lift above 100% even after three months. Effects that large are rare in mature testing programs13, so an A/B test at this traffic will mostly end without an answer. Raw counts carry wide uncertainty too: 3 payments from 100 visitors is a rate between 1.0% and 8.5% at 95% confidence (Wilson interval, computed for this report), and 10 sign-ups from 200 visitors a rate between 2.7% and 9.0%.

At this traffic, an owner learns more by reading counts against targets set before launch, and by finding problems with people rather than with traffic splits. Take a small service business, such as a plumbing firm or an accounting practice, whose site asks visitors to request a quote. Its conversion is the quote request, its funnel runs from page view through form start to the request sent, and Table 15 lists the order in which the business would measure them.

Table 15. Order of measurement for a small site’s pages. IDs refer to Tables 2 to 12.
WhenWhat to measureHowRead as
Before launchUsability problems on the main pagesFive people from the target group each try the pages while thinking aloud; fix, then a second round of five (Q6)Problems found and fixed, not a rate
Before launchWhether the first screen is understoodFive-second test: show the home page first screen, ask what it offers, for whom, and what the button does (H1)Share who answer all three correctly
Before launchSpeed and accessibilityField LCP, INP, CLS; WebAIM-type automated check; keyboard-only pass (S1, X1 to X5)All pass thresholds
From launch, weeklyFunnel per page: view, button click, form start, form sent, checkout start, payment, refund asked, by source tagOne named event per step (Q1); counts and starter-to-completion ratesCounts against targets set before launch; trend by week
When a page passes 1,000 visitors a weekFirst A/B testOne change with a lift large enough to detect (Figure 4), sized in advance, fixed length (Q2, Q7)Significant at a fixed end; recheck large wins (Q4)

Changes made between readings should be logged with their date, so that a change in a count can be checked against what changed on the page; a before-and-after comparison is reported as such and not as a measured lift112123.

Discussion

The pattern across the catalog has a plain mechanism. A visitor at a landing page weighs the value of acting against its cost in effort and risk. Techniques with strong evidence lower that cost without changing the offer: a page that shows its content sooner, text that can be scanned, a form that asks less and reports errors sooner, a visible price and refund, design that signals a competent business. Persuasion findings from the laboratory try instead to raise the felt value of the same offer, and their effects are smaller, depend on conditions, and in several famous cases did not repeat. The deceptive tactics raise acceptance by hiding cost, which is why they work in experiments and why regulators act against them.

The evidence has limits for sites that sell to businesses. Most experiments and benchmarks come from retail, travel and media sites, where purchases are smaller and quicker than a firm’s choice of a monthly tool; Fogg’s finding that finance-site visitors weigh motive and reputation more, and Unbounce’s finding that harder text did better in finance, both suggest that money pages differ. Several grade-A results come from one site each (Vodafone, Rakuten 24, the Japanese retailer), and none of the speed tests reports a significance test for its business measures. Benchmark medians depend on their exclusions, and practitioner counts measure popularity. The legal summary reflects texts read on one date, during an open FTC rulemaking and a pending case on whether ROSCA reaches business buyers.

The catalog suggests tests that could be run at higher traffic. Whether showing a calculator result before asking for email raises or lowers email capture could be tested directly once a calculator page draws a few thousand visitors a month. Whether a visible formula and cited rule raise trust among professional buyers, as the experiment with a financial adviser suggests for consumers, could be tested with message tests on such buyers before any traffic split. And whether the refund promise that raised purchases for fashion shoppers transfers to a monthly business tool would need a randomized offer at checkout.

Data and materials

Every source was read in full on 31 October 2026. Figure 1 counts the grades in Tables 2 to 12, Figures 2, 3 and 5 plot numbers reported in the sources named in their captions, and Figure 4 and the Wilson intervals in Section 6 are computed from the formulas cited there.

Sources

  1. Therese Fessenden, “Scrolling and attention,” Nielsen Norman Group, 2018. nngroup.com
  2. Jakob Nielsen, “Conversion rates,” Nielsen Norman Group, 2013. nngroup.com
  3. Unbounce, “Conversion benchmark report 2024,” 2024. unbounce.com
  4. Jakob Nielsen, “How users read on the web,” Nielsen Norman Group, 1997. nngroup.com
  5. B. J. Fogg, Cathy Soohoo, David Danielson, Leslie Marable, Julianne Stanford and Ellen R. Tauber, “How do people evaluate a web site’s credibility? Results from a large study,” Stanford Persuasive Technology Lab and Consumer WebWatch, 2002. credibility.stanford.edu
  6. Luke Wroblewski, “Inline validation in web forms,” A List Apart, 2009. alistapart.com
  7. Mirjam Seckler, Silvia Heinz, Javier A. Bargas-Avila, Klaus Opwis and Alexandre N. Tuch, “Empirical evaluation of 20 web form optimization guidelines,” CHI ’13 Extended Abstracts on Human Factors in Computing Systems, 2013. static.googleusercontent.com
  8. Amos Tversky and Daniel Kahneman, “Judgment under uncertainty: Heuristics and biases,” Science, 1974. doi.org
  9. Eric J. Johnson and Daniel Goldstein, “Do defaults save lives?,” Science, 2003. doi.org
  10. Sheena S. Iyengar and Mark R. Lepper, “When choice is demotivating: Can one desire too much of a good thing?,” Journal of Personality and Social Psychology, 2000. doi.org
  11. Benjamin Scheibehenne, Rainer Greifeneder and Peter M. Todd, “Can there ever be too many options? A meta-analytic review of choice overload,” Journal of Consumer Research, 2010. doi.org
  12. Shane Frederick, Leonard Lee and Ernest Baskin, “The limits of attraction,” Journal of Marketing Research, 2014. doi.org
  13. Ron Kohavi, Alex Deng and Lukas Vermeer, “A/B testing intuition busters: Common misunderstandings in online controlled experiments,” Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022. doi.org
  14. Ron Kohavi, “Replication of four A/B testing patterns and the winner’s curse (KDD 2026 paper announcement, with abstract),” LinkedIn, 2026. linkedin.com
  15. Jamie Luguri and Lior Jacob Strahilevitz, “Shining a light on dark patterns,” Journal of Legal Analysis, 2021. doi.org
  16. Federal Trade Commission, “Bringing dark patterns to light,” 2022. ftc.gov
  17. Federal Trade Commission, “FTC secures historic $2.5 billion settlement against Amazon,” 2025. ftc.gov
  18. Federal Trade Commission, “FTC and states take action against Toast Inc. over hidden and unauthorized fees, deceptive subscription and cancellation practices and other unlawful conduct,” 2026. ftc.gov
  19. Evan Miller, “How not to run an A/B test,” 2010. evanmiller.org
  20. Zuko Analytics, “Form benchmarking: Form type benchmarking and online form statistics,” 2025. zuko.io
  21. Philip Walton, “Web Vitals,” web.dev, 2024. web.dev
  22. Philip Walton and Barry Pollard, “Largest Contentful Paint (LCP),” web.dev, 2025. web.dev
  23. Jeremy Wagner and Barry Pollard, “Interaction to Next Paint (INP),” web.dev, 2025. web.dev
  24. Milica Mihajlija and Philip Walton, “Cumulative Layout Shift (CLS),” web.dev, 2023. web.dev
  25. web.dev, “Vodafone: A 31% improvement in LCP increased sales by 8%,” 2021. web.dev
  26. Hayoung Lee, Linh Duong, Ryunosuke Akiba and Shogo Kashiwase, “How Rakuten 24’s investment in Core Web Vitals increased revenue per visitor by 53.37% and conversion rate by 33.13%,” web.dev, 2022. web.dev
  27. Deloitte Digital and Fifty-Five, “Milliseconds make millions: A study on how improvements in mobile site speed positively affect a brand’s bottom line,” Google, 2020. thinkwithgoogle.com
  28. Daniel An, “Find out how you stack up to new industry benchmarks for mobile page speed,” Think with Google, 2018. web.archive.org
  29. Jakob Nielsen, “Response times: The 3 important limits,” Nielsen Norman Group, 1993. nngroup.com
  30. Amit Kumar and Saurabh Rajpal, “How redBus improved their website’s Interaction to Next Paint (INP) and increased sales by 7%,” web.dev, 2023. web.dev
  31. Saurabh Rajpal and Swetha Gopalakrishnan, “The business impact of Core Web Vitals,” web.dev, 2021. web.dev
  32. John Morkes and Jakob Nielsen, “Concise, SCANNABLE, and objective: How to write for the web,” Nielsen Norman Group, 1997. nngroup.com
  33. Jakob Nielsen, “F-shaped pattern for reading web content (original study),” Nielsen Norman Group, 2006. nngroup.com
  34. Kara Pernice, “F-shaped pattern of reading on the web: Misunderstood, but still relevant (even on mobile),” Nielsen Norman Group, 2017. nngroup.com
  35. Harry Dry, “My step-by-step guide to landing pages that convert,” Marketing Examples, n.d.. marketingexamples.com
  36. Copy Hackers, “Customers buy the product they understand fastest,” X, 2026. x.com
  37. B. J. Fogg, “Stanford guidelines for web credibility,” Stanford Persuasive Technology Lab, 2002. credibility.stanford.edu
  38. Peep Laja, “Most homepages lead with reasons nobody picks,” LinkedIn, 2026. linkedin.com
  39. Peep Laja, “The words “agentic” and “AI-powered” are on most category-leader homepages,” LinkedIn, 2026. linkedin.com
  40. Talia Wolf, “So many brands are telling prospects to leave,” LinkedIn, 2026. linkedin.com
  41. Aurora Harley and Kim Flaherty, ““Get started” stops users,” Nielsen Norman Group, 2017. nngroup.com
  42. Dan Urban, “Should you remove navigation from your landing pages? Data reveals the answer,” HubSpot, 2025. blog.hubspot.com
  43. Paras Chopra, “A/B testing case study: Removing navigation menu increased conversions by 100%,” Wingify, 2011. wingify.com
  44. Casey Hill, “What does the data say about “sign up with Google”?,” LinkedIn, 2026. linkedin.com
  45. Michael Aagaard, “How to design call to action buttons that convert,” Unbounce, 2013. unbounce.com
  46. Casey Hill, “Pick any one of these 5 SaaS websites,” LinkedIn, 2026. linkedin.com
  47. Tom Collinger, Edward Malthouse, Ewa Maslowska, Ran Wang, Su Jung Kim and Georgios Askalidis, “How online reviews influence sales,” Spiegel Research Center, Northwestern University, 2017. spiegel.medill.northwestern.edu
  48. Ben Labay, “Which types of social proof work best? [Original research],” CXL Institute, 2016. cxl.com
  49. Benjamin Scheibehenne, Tahira Jamil and Eric-Jan Wagenmakers, “Bayesian evidence synthesis can reconcile seemingly inconsistent results: The case of hotel towel reuse,” Psychological Science, 2016. doi.org
  50. Gerd Bohner and Lena E. Schlüter, “A room with a viewpoint revisited: Descriptive norms and hotel guests’ towel reuse behavior,” 2014. socialpsychology.org
  51. Daniel Ben David, Yehezkel S. Resheff and Talia Tron, “Explainable AI and adoption of financial algorithmic advisors: An experimental study,” Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, 2021. arxiv.org
  52. Aurora Harley, “Trustworthiness in web design: 4 credibility factors,” Nielsen Norman Group, 2016. nngroup.com
  53. Hoa Loranger, “State the price to give B2B sites a competitive advantage,” Nielsen Norman Group, 2013. nngroup.com
  54. Jen Cardello, “Social proof in the user experience,” Nielsen Norman Group, 2014. nngroup.com
  55. Baymard Institute, “50 cart abandonment rate statistics 2026,” 2025. baymard.com
  56. Jamie Holst, “How users perceive security during the checkout flow (incl. new ‘trust seal’ study 2023),” Baymard Institute, 2016. baymard.com
  57. Christian Holst, “Which site seal do people trust the most? (2013/2016 survey results),” Baymard Institute, 2013. baymard.com
  58. Taylor Kohler, “12 design recommendations for calculator and quiz tools,” Nielsen Norman Group, 2024. nngroup.com
  59. Javier Bargas-Avila and Mirjam Seckler, “Simple is better: Making your web forms easy to use pays off,” Google Research, 2014. research.google
  60. Edward Scott, “Checkout optimization: 5 ways to minimize form fields in checkout,” Baymard Institute, 2024. baymard.com
  61. NP Digital, “How form length impacts conversion rate,” 2024. neilpatel.com
  62. Zuko Analytics, “8 surprising insights from Zuko’s webform benchmarking data,” 2025. zuko.io
  63. Brian Moloney, “Fewer fields in a contact form sharply increases conversions,” Imaginary Landscape, 2008. imagescape.com
  64. Nick Heath, “Expedia on how one extra data field can cost $12m,” ZDNet, n.d.. zdnet.com
  65. Edward Scott, “Usability testing of inline form validation: 31% don’t have it, 4% get it wrong,” Baymard Institute, 2024. baymard.com
  66. Katie Sherwin, “Placeholders in form fields are harmful,” Nielsen Norman Group, 2014. nngroup.com
  67. Edward Scott, “E-commerce checkouts need to mark both required fields and optional fields explicitly (only 14% do so),” Baymard Institute, 2018. baymard.com
  68. Raluca Budiu, “Marking required fields in forms,” Nielsen Norman Group, 2019. nngroup.com
  69. Rachel Krause, “10 design guidelines for reporting errors in forms,” Nielsen Norman Group, 2019. nngroup.com
  70. Luke Wroblewski, “Testing accordion forms,” A List Apart, 2010. alistapart.com
  71. Sumo, “Pop-up statistics: Findings from analyzing 2 billion pop-up examples,” 2019. web.archive.org
  72. Jon M. Jachimowicz, Shannon Duncan, Elke U. Weber and Eric J. Johnson, “When and why defaults influence decisions: A meta-analysis of default effects,” Behavioural Public Policy, 2019. doi.org
  73. Lunzheng Li, Zacharias Maniadis and Constantine Sedikides, “Anchoring in economics: A meta-analysis of studies on willingness-to-pay and willingness-to-accept,” Journal of Behavioral and Experimental Economics, 2021. doi.org
  74. Dan Ariely, George Loewenstein and Drazen Prelec, ““Coherent arbitrariness”: Stable demand curves without stable preferences,” The Quarterly Journal of Economics, 2003. doi.org
  75. Drew Fudenberg, David K. Levine and Zacharias Maniadis, “On the robustness of anchoring effects in WTP and WTA experiments,” American Economic Journal: Microeconomics, 2012. doi.org
  76. Alexander Chernev, Ulf Böckenholt and Joseph Goodman, “Choice overload: A conceptual review and meta-analysis,” Journal of Consumer Psychology, 2015. doi.org
  77. Xueming Luo, Xiaoyi Wang, Yue Wu and Takeshi Moriguchi, “From stopping to shopping: A field experiment on free return and free shipping retargeting policies in online retail operations,” Manufacturing & Service Operations Management, 2025. scholars.cityu.edu.hk
  78. Recurly, “State of subscriptions: The acquisition chapter,” 2022. recurly.com
  79. First Page Sage, “SaaS free trial conversion rate benchmarks,” 2025. firstpagesage.com
  80. Peep Laja, “The whole internet is your pricing page now,” LinkedIn, 2026. linkedin.com
  81. Kate Moran and Tim Dykes, “Comparison tables for products, services, and features,” Nielsen Norman Group, 2024. nngroup.com
  82. Ben Labay, “Pricing page optimization: How to order pricing plans [Original research],” CXL Institute, 2016. cxl.com
  83. Ben Labay, “The effects of highlighting a “recommended” pricing plan [Original research],” Speero, 2016. speero.com
  84. ChartMogul, “Billing is the invisible force behind your growth,” 2025. chartmogul.com
  85. Avner Strulov-Shlain, “More than a penny’s worth: Left-digit bias and firm pricing,” 2018. anderson.ucla.edu
  86. RevenueCat, “Money-back guarantee vs. free trial for subscription apps,” 2026. revenuecat.com
  87. Joel Huber, John W. Payne and Christopher P. Puto, “Let’s be honest about the attraction effect,” Journal of Marketing Research, 2014. doi.org
  88. Dan Ariely, “Are we in control of our own decisions?,” TED, 2008. ted.com
  89. Venkat Kuppuswamy and Barry L. Bayus, “Crowdfunding creative ideas: The dynamics of project backers in Kickstarter,” 2013. funginstitute.berkeley.edu
  90. Unlock SaaS, “Founding member pricing (cohort-locked lifetime discount),” 2026. unlocksaas.com
  91. Federal Trade Commission, “Rule concerning the use of prenotification negative option plans: Advance notice of proposed rulemaking,” 2026. federalregister.gov
  92. Taylor Kohler, “Calculators and quizzes: User expectations,” Nielsen Norman Group, 2024. nngroup.com
  93. Content Marketing Institute and ion interactive, “Deliver peak experiences with interactive content,” 2016. eu-assets.contentstack.com
  94. Harry Dry, “Rows’ homepage test that doubled conversions,” Marketing Examples, n.d.. marketingexamples.com
  95. GetResponse, “Email marketing benchmarks,” 2024. getresponse.com
  96. Omnisend, “Email marketing benchmarks 2026,” 2026. omnisend.com
  97. Kim Flaherty, “Marketing email and newsletters: UX findings then and now,” Nielsen Norman Group, 2017. nngroup.com
  98. Mailchimp, “Single opt-in vs. double opt-in,” n.d.. mailchimp.com
  99. Federal Trade Commission, “CAN-SPAM Act: A compliance guide for business,” n.d.. ftc.gov
  100. Christian Holst, “‘Touch keyboard’ implementations have improved just 9% since 2013 (60% still get it wrong),” Baymard Institute, 2015. baymard.com
  101. Aurora Harley, “Touch targets on touchscreens,” Nielsen Norman Group, 2019. nngroup.com
  102. W3C Web Accessibility Initiative, “What’s new in WCAG 2.2,” 2023. w3.org
  103. WebAIM, “The WebAIM Million: The 2026 report on the accessibility of the top 1,000,000 home pages,” 2026. webaim.org
  104. U.S. Department of Justice, “Extension of compliance dates for nondiscrimination on the basis of disability; accessibility of web information and services of state and local government entities,” 2026. federalregister.gov
  105. U.S. Department of Justice, “Guidance on web accessibility and the ADA,” 2022. ada.gov
  106. Stefan Thomke, “Building a culture of experimentation,” Harvard Business Review, 2020. hbr.org
  107. Jakub Linowski, “GoodUI patterns,” GoodUI, 2026. goodui.org
  108. Ron Kohavi, Diane Tang and Ya Xu, “Trustworthy online controlled experiments: A practical guide to A/B testing,” Cambridge University Press, 2020. doi.org
  109. Jakob Nielsen, “Why you only need to test with 5 users,” Nielsen Norman Group, 2000. nngroup.com
  110. Megan Chan, “Testing visual design: A comprehensive guide,” Nielsen Norman Group, 2024. nngroup.com
  111. Wynter, “A/B testing alternative: Message testing,” 2026. wynter.com
  112. Speero, “Researching with low traffic,” 2026. speero.com
  113. Ron Kohavi, Randal M. Henne and Dan Sommerfield, “Practical guide to controlled experiments on the web: Listen to your customers not to the HiPPO,” Proceedings of the 13th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2007. doi.org
  114. Arunesh Mathur, Gunes Acar, Michael J. Friedman, Eli Lucherini, Jonathan Mayer, Marshini Chetty and Arvind Narayanan, “Dark patterns at scale: Findings from a crawl of 11K shopping websites,” Proceedings of the ACM on Human-Computer Interaction, 2019. doi.org
  115. Federal Trade Commission, “Trade regulation rule on the use of consumer reviews and testimonials, 16 CFR part 465,” 2024. law.cornell.edu
  116. Federal Trade Commission, “Guides concerning the use of endorsements and testimonials in advertising, 16 CFR 255.2,” 2023. law.cornell.edu
  117. Federal Trade Commission, “The consumer reviews and testimonials rule: Questions and answers,” n.d.. ftc.gov
  118. Federal Trade Commission, “Rule on unfair or deceptive fees, 16 CFR part 464,” 2025. law.cornell.edu
  119. United States Court of Appeals for the Eighth Circuit, “Custom Communications, Inc. v. Federal Trade Commission, no. 24-3137,” 2025. ecf.ca8.uscourts.gov
  120. Federal Trade Commission, “Agency information collection activities: Negative option rule,” 2026. federalregister.gov
  121. State of California, “Business and Professions Code sections 17601 to 17602 (Automatic Renewal Law),” 2025. leginfo.legislature.ca.gov
  122. Federal Trade Commission, “Business guide to the FTC’s mail, internet, or telephone order merchandise rule,” n.d.. ftc.gov
  123. Peep Laja, “How to do conversion optimization with very little traffic?,” CXL, 2013. cxl.com

Want us to look at your business this way?

We measure where calls, leads and paperwork slip through, in your own records, before we build anything.

Book a free call
ResearchDownload the PDF30 pages