Article: How to Remove Friction Honestly
Reviewed: November 2026 · Method version 2 · How these reviews work
Each factual claim in the article was checked against the source it cites, then challenged by reviewers whose job was to find the strongest evidence against it: replication attempts, meta-analyses, later papers that cite the original work, and research from other fields that study the same mechanism. A separate reviewer weighed both sides, and a script re-checked every quoted source. Where a claim was overstated or misattributed, the article was corrected. The research below supports the article as it now reads.
Dashboards reward the belief. Remove a field and sign-ups often rise; remove a screen and the funnel looks healthier.
Experiments outside UX, such as Galesic and Bosnjak’s survey-length study, support the general pattern that more requested effort means fewer completions, though the size of the effect varies.
Teams that celebrate one-click sign-up often own untimed cancellation flows, so friction is removed where it costs the company and left where it costs the user.
Sunstein’s work on ‘sludge’ describes the same asymmetry: friction that serves the institution rather than the person, often costing more than it is worth.
In 2016, Anna Cox and colleagues challenged the design advice that friction should always be removed to avoid frustration and disengagement.
Later studies citing Cox et al. have tested their idea in practice, for example Grüning et al. (CHI 2024) found that short frictions before opening apps made app use more intentional.
Small intentional barriers (microboundaries) can interrupt mindless automatic interactions and prompt reflection, with benefits ranging from fewer data-entry errors to support for health-behaviour change.
A longitudinal study of 1,039 smartphone users (Grüning et al., CHI 2024) found that short frictions before opening chosen apps reduced opening attempts. A 2026 TikTok study found heavier friction cut compulsive use but made the experience worse, a reminder to keep pauses small.
High school seniors whose parents received this help were 8 percentage points more likely to have completed two years of college within three years, 36% against 28%.
Later randomized trials point the same way: in France, in-person help applying for benefits raised take-up while an online eligibility calculator did not, and in Delhi application assistance raised applications by 41 to 70 percent.
An IRS field experiment found something similar for tax credits. Among 35,050 filers who had failed to claim $26 million despite a first notice, a second mailing produced substantial additional claims, especially when it was simpler and made the benefit more salient. Attempts to reduce the perceived stigma or cost of applying did not help.
Linos and colleagues ran six preregistered experiments with about a million households who had not filed a tax return and found that letters and texts did not raise filing or credit claims, which suggests reminder mailings work better for people already in the system.
In both studies, a simpler and clearer path to the money moved people: a shorter notice and worksheet in the IRS study, hands-on help with the form in the student-aid study. In the student-aid study, information without that help moved no one.
The IRS mailing included the claiming worksheet; a shorter worksheet raised response from 19% to 23% and a shorter notice from 17% to 23%, with the largest gain (8 points) from showing the credit amount. A 2026 meta-analysis of 51 field experiments finds help with the application raises take-up by about 8 points on average against about 3 for information alone, and trials in France and India agree. Information alone is not always idle: a federal letter raised naturalization applications by 1 point and a free-tuition guarantee letter raised applications from 26% to 68%, while letters to EITC non-filers and a College Board campaign moved no one. The ‘no one’ finding belongs to the student-aid study only.
Prompted users posted 6% fewer offensive Tweets than control, created fewer offensive Tweets afterwards and received fewer offensive replies, while the change in total replies sent was not statistically significant.
Work on ‘accuracy prompts’ in misinformation research supports the idea that a brief nudge to reconsider reduces harmful sharing, although a preregistered replication (Roozenbeek et al., 2021) found the effect smaller than first reported.
The prompts also seemed to lose force with repetition: the more prompts someone saw, the more likely they were to send the Tweet unchanged.
Warning research supports habituation: in Vance and colleagues’ three-week field study, people followed repeated security warnings less over time, and changing the warning’s look reduced that effect.
Confirmation prompts on trivial actions wear out fast.
Vance and colleagues found attention to and compliance with repeated warnings fell over three weeks, and that varying the warning’s design reduced the decline.
These edits were made on November 12, 2026 after the review.
Claim reviewed: Removing a form field makes sign-ups rise, and removing a screen makes the funnel look healthier.
Now reads: Dashboards reward the belief. Remove a field and sign-ups often rise; remove a screen and the funnel looks healthier.
Why: No source is cited, but the general direction holds: experiments in survey research show that asking for more effort lowers how many people start and finish. The sentence states it as a rule, though, and effort is not always a pure cost; some studies find visible effort can raise perceived value.
Claim reviewed: Simpler mailings that made the benefit more salient produced substantial additional claims, while attempts to reduce the perceived stigma or cost of applying did not.
Now reads: An IRS field experiment found something similar for tax credits. Among 35,050 filers who had failed to claim $26 million despite a first notice, a second mailing produced substantial additional claims, especially when it was simpler and made the benefit more salient. Attempts to reduce the perceived stigma or cost of applying did not help.
Why: The paper credits three things for the extra claims: simply receiving a second mailing, simplification, and making the benefit more salient. The article mentions only the last two, which makes the design changes sound like the whole effect. A later set of six large experiments on similar letters found no effect on people who had not filed taxes at all, so the result has limits.
Claim reviewed: In both studies, making the task simpler moved people; in the student-aid study, information without help moved no one.
Now reads: In both studies, a simpler and clearer path to the money moved people: a shorter notice and worksheet in the IRS study, hands-on help with the form in the student-aid study. In the student-aid study, information without that help moved no one.
Why: The student-aid half is exactly what the source says: help with the form raised college completion, information alone did not. The IRS half is looser than pass 4 allowed but not wrong: every mailing carried the claiming worksheet, and J-PAL’s summary of the study shows the shorter worksheet lifted response from 19% to 23%, though the shorter notice (17% to 23%) and showing the credit amount (8 more points) did more. A 2026 meta-analysis of 51 field experiments finds help with the application raises take-up by about 8 points on average against about 3 for information alone, so the lesson is broadly right; the article should just not imply that information alone never moves anyone, because in many settings it moves a few.
Claim reviewed: The effect faded: the more prompts someone saw, the more likely they were to send the Tweet unchanged.
Now reads: The prompts also seemed to lose force with repetition: the more prompts someone saw, the more likely they were to send the Tweet unchanged.
Why: The paper does report that people sent more Tweets unchanged with each further prompt. ‘The effect faded’ reads more into that than the paper does: people who get many prompts may differ from those who get few, and the paper found first-prompt responses were similar for occasional and frequent offenders. Research on security warnings does show people tune out repeated warnings.
Claim reviewed: The GOV.UK Check answers pattern adds a single review page immediately before the confirmation screen so people get a second chance to notice and correct errors; its guidance and code are open source (MIT).
Now reads: For small and medium-sized transactions, the pattern adds a single review page immediately before the confirmation screen, so people get a second chance to notice and correct errors before submitting.
Why: GOV.UK describes the pattern this way, but recommends a single check page only for small to medium-sized transactions; for long services it suggests checking each section. GOV.UK states the error-reduction benefit as a design rationale, not a measured result.
Claim reviewed: For reversible actions, an undo offered afterwards often protects people better than a confirmation question beforehand; up-front questions should be saved for irreversible actions and name the consequence in the button label.
Now reads: That logic extends to a second kind of protective step. For actions that can be reversed, an undo option offered afterwards often protects people better than a question asked beforehand, because it costs nothing for the person who meant it. Save the up-front question for actions that can’t be taken back, and put the consequence in the button label, such as “Delete 14 projects,” so the pause carries information.
Why: The paragraph’s three points, offer undo rather than a question, keep the question for actions that cannot be undone, and name the action on the button, are exactly the guidance in Nielsen Norman Group’s 2018 article on confirmation dialogs (by Jakob Nielsen) and Aza Raskin’s 2007 essay ‘Never Use a Warning When You Mean Undo’, neither of which the article credits. The advice is sound: field studies in four fields show that routine prompts get clicked through, with 70% of Chrome’s SSL warnings dismissed, half or more of drug alerts overridden and consent dialogs accepted on sight. But no experiment compares undo with a confirmation head to head, and undo only helps when the person notices the mistake, so ‘often protects people better’ rests on expert judgment plus the habituation evidence. Now cites Confirmation Dialogs Can Prevent User Errors (If Not Overused) — Jakob Nielsen, Nielsen Norman Group (2018).