← Back to Sahilverse Humane Interfaces

An independent inquiry · Interface ethics · Updated 24 October 2026

Design with a door out.

Every screen asks you to do something. This is a field guide to the screens built so that saying yes is effortless and leaving is expensive — and to what the honest version of the same screen looks like.

In about ten minutes you can learn to spot six common traps, test any interface with five plain questions, and redesign one yourself.

Start exploring → Read the argument
Two buildings, two exits On the left, a building with a wide lit entrance and a tiny exit reached through a winding corridor with four checkpoints. On the right, the same building with an entrance and an exit of equal size, side by side. TRAP Join exit? 1 step in · 5 steps out HUMANE Join Leave 1 step in · 1 step out same building · different promise
The whole argument in one picture. Neither building forbids leaving; one of them makes leaving a project.

EvidenceEstablished frameworkInterpretationHypothesisIllustrationFiction
These labels mark how strongly each claim is supported. Every product, company and number inside the demo screens is invented.

01 · The argument

Leaving should cost about what joining cost.

IllustrationPicture two doors in the same building. The entrance is wide, well lit and opens as you approach. The exit is behind a curtain at the end of a corridor, and it opens only after you explain to a clerk why you want to go, decline a voucher, and wait for a manager who works weekdays. Nobody locked you in. Every step was, technically, optional. And yet the building has made a decision on your behalf.

Most of what people call “dark patterns” has this shape. Not outright lies so much as asymmetries: one path is designed with great care, tested and polished; the other path is left to fend for itself, or is quietly obstructed. A pre-ticked box, a decline button written as an insult, a cancellation that requires a phone call, a notification switch that bundles security alerts with sales messages — each is small. Together they move very large numbers of people in a direction they did not choose.

An interface is humane when the way out is as legible as the way in.

InterpretationThat sentence is the claim of this page, and it is narrower than it sounds. It does not say interfaces should be neutral; no interface can be. It does not say friction is bad; friction before deleting four years of photographs is a kindness. It says the ethical weight of a design lives less in its button colours than in three relations: whose goal each path serves, how much each path costs, and who can afford that cost.

Same mechanism, different ethics

IllustrationTake the humble default. A donation page that pre-selects “add ₹10 to plant a tree” and a checkout page that pre-selects “share my email with partners” use the identical mechanism: a box that is already ticked. Yet most people would judge them differently. The first is visible, small, obviously connected to what you were doing, and easy to untick; the money goes where it says. The second is often far from the decision you came to make, and its cost — a long future of messages from strangers — arrives after the moment you could have noticed it.

Or take confirmation. Three “are you sure?” screens before deleting a shared family album protects people who share that account. Three “are you sure?” screens before cancelling a subscription protects revenue. Same widget; different beneficiary. This is why a checklist of forbidden components will never be enough. You have to ask what the screen is for, and for whom.

What this changes for an ordinary reader

Once you see the asymmetry, you start counting. How many taps did it take to sign up? How many to leave? Could I say out loud what I just agreed to? Is the “no” as visible as the “yes”? Those questions do not require knowing the law of any country, and they work on apps, forms, kiosks and paper alike. The five door tests below turn them into a habit.

HypothesisSahil’s working hypothesis, not an established finding: products that keep exit as legible as entry may give up some short-term retention in exchange for longer-term trust, fewer complaints and easier re-subscription. It is plausible, and some regulators clearly think the trade is worth requiring. But this page has not tested it. A fair test would compare otherwise similar products over a long period and measure more than the dashboard’s favourite number.

02 · A usable framework

Five door tests for any screen.

InterpretationThese are editorial tests synthesised from the research and guidance listed in Sources, not a validated instrument. They give no score. A screen that fails one test has started a conversation, not received a verdict.

  1. Symmetry

    Is the way out roughly as easy as the way in? Unsubscribe, cancel, refuse and delete should not take many more steps, channels or office hours than subscribe, buy, accept and create.

    Count the steps to join. Count the steps to leave.

  2. Legibility

    Could the person explain, in their own words, what they just agreed to? Double negatives, bundled purposes and settings named for something else all fail here.

    Read it aloud to a tired friend. What did they hear?

  3. Reversibility

    Can the action be undone, and is the undo visible at the moment someone would need it? Where an action truly can’t be undone, does the screen say exactly what will be lost?

    If this was a mistake, where is the way back?

  4. Alignment

    When the default is accepted, who benefits? Would the team be comfortable explaining this design choice to the person it is aimed at — a version of the “publicity” idea Thaler and Sunstein discuss in Nudge?

    Would we put this decision in the release notes?

  5. Margins

    Does the honest path still work for someone tired, rushed, using a screen reader, navigating by keyboard, reading in a second language, or on an old phone with a cracked screen?

    Who is this hardest for — and is the hard part the “no”?

03 · Control room

Six screens, two versions each.

Each exhibit shows the same component twice: on the left, a trap version with its tactics labelled; on the right, a humane redesign. Both sides work — tick, toggle, click through, try to leave. Use Inspect to switch the analysis between the pattern and the alternative.

FictionEvery brand, price and number inside these screens is invented. Nothing here collects information: there are no real email fields, no network requests, and the cookie demo sets no cookies. Where real-world traps use pale grey text or tiny buttons, we kept ours readable and tappable and pointed at the problem instead — this page should not commit the harm it describes.

Exhibit 01 · Newsletter opt-in

The box you didn’t tick

A fictional tea shop’s checkout. You came to buy tea. What else are you agreeing to?

⚠ Trap patternfictional

Kettlewick Tea · Checkoutstep 3 of 3

Order total: ₹640 · 2 tins of Darjeeling

  • Preselection: both boxes start ticked.
  • Trick wording: the second box is a double negative; leaving it ticked shares your email.
  • Confirmshaming: the decline option mocks you, and “Place order” is welded to “join”.

✓ Humane alternativefictional

Kettlewick Tea · Checkoutstep 3 of 3

Order total: ₹640 · 2 tins of Darjeeling

Optional emails (both start off)

You can change these any time under Account → Emails.

  • Separate, plainly worded purposes; nothing pre-ticked.
  • One “Place order” button that does only that.
  • Tells you where to change your mind later.

Why teams ship it · incentive

List size is easy to measure and often tied to marketing targets. Pre-ticked, confusingly worded boxes reliably grow the list, and partner sharing can be sold or traded. On a dashboard this looks like a win.

What it costs people

Unwanted mail, an email address now held by companies you never chose, and the small corrosive feeling of having been handled. For the business: unengaged subscribers, spam complaints and readers who learn to distrust the checkout.

What the alternative changes

Each purpose stands alone, starts off, and is described by what actually happens (“we would share your email”). The purchase and the subscription are separate acts, so “yes” to one is not smuggled into the other.

What it honestly costs the business

Probably a smaller list. The bet is that a smaller list of people who asked is worth more than a larger list of people who were caught — a bet worth testing, not assuming.

Nuance

Not every default is a trap. A pre-selected “email me my receipt” is expected and serves the buyer. The questions are whether the default serves the person, whether it is visible at the moment of choice, and whether undoing it is easy. India’s 2023 dark-pattern guidelines name confirm shaming and trick question among their specified patterns; the scholarly taxonomies call these preselection, trick wording and confirmshaming.

Exhibit 02 · Destructive confirmation

“Are you sure?” is not a design

A fictional notes app. You are about to delete a project that holds 14 notes and 3 photos.

⚠ Trap patternfictional

FieldnotesProjects

Cancel deletion?

This action cannot be undone.

  • Trick wording: “Yes” keeps the project; “No” deletes it.
  • Missing consequences: no name, no count, no hint of what will be lost.
  • Misleading emphasis: the destructive answer is the bold one.

✓ Humane alternativefictional

FieldnotesProjects

Delete “Monsoon garden notes”?

14 notes and 3 photos will move to Trash for 30 days, then be removed permanently. Shared links stop working now.

  • Names the thing and counts what will be lost.
  • Buttons say what they do: “Keep project” / “Delete project”.
  • An Undo arrives at the moment it’s needed; Trash gives 30 days.

Why teams ship it · incentive

Usually nobody intends harm. Confirmation dialogs get written last, by whoever is closest, from a generic component that only offers Yes/No. The dialog exists so the team can say “we warned them”.

What it costs people

Lost work, lost memories, support tickets nobody can resolve. People learn to click through confirmations without reading — which makes every future warning weaker.

What the alternative changes

It replaces a quiz with information. Verb-labelled buttons can be understood without the question; the count makes the stakes concrete; the undo and the Trash make a mistake survivable. Where something truly cannot be reversed, a humane design names it and may ask for a deliberate act, such as typing the project’s name.

What it honestly costs the business

A Trash means storage and a little engineering. It is usually cheaper than the support hours spent on irrecoverable mistakes.

Nuance

This exhibit is the counterexample to “friction is bad”. Here friction protects the person. WCAG 2.2’s success criterion 3.3.4 asks that pages which modify or delete user-controllable data make submissions reversible, checked, or confirmed — and a confirmation only counts in spirit if people can understand it.

Exhibit 03 · Unsubscribe / cancel

One tap in, five steps out

A fictional streaming membership. Signing up took one tap in both versions. Now try to leave.

⚠ Trap patternfictional

Tidepool Plus · AccountPremium

Your membership

Premium · ₹499/month · renews 12 Nov

Help · Terms ·

  • Hard to cancel / obstruction: the exit hides under “Manage”, then runs a gauntlet.
  • Forced action: a reason is required before you may continue.
  • Channel switch: the last step demands a phone call in office hours.

✓ Humane alternativefictional

Tidepool Plus · AccountPremium

Cancel membership

If you cancel today, you keep access until 12 Nov and won’t be charged again.

  • The consequence (access until 12 Nov) is stated before you decide.
  • One honest alternative (pause), offered once, at equal weight.
  • The reason is genuinely optional.

Why teams ship it · incentive

Monthly churn is a headline metric. Each extra step converts some would-be leavers into “retained” users. A/B tests reward whichever flow keeps more people — and the test cannot see resentment, chargebacks or the friend you warned.

What it costs people

Money paid for something unwanted, time, and a lesson that the company is an adversary. People with less time, less confidence or less phone credit pay most.

What the alternative changes

Symmetry: leaving takes about as long as joining did. A retention offer still exists — once, honestly framed, and equally easy to decline. The person leaves knowing exactly when access ends.

What it honestly costs the business

Some people who would have been worn down will leave. Those were people paying for something they had decided not to want.

Nuance

Retention offers are not inherently manipulative; a pause option genuinely helps people whose budgets or schedules change. The trap is the accumulation: hidden entry, required justification, repeated offers and a channel switch. The FTC’s 2022 staff report discusses subscription sales among its contexts, and India’s 2023 guidelines list a “subscription trap”.

Exhibit 04 · Notification settings

The switch that sulks

A fictional social app’s notification settings. Try turning off the things you don’t want.

⚠ Trap patternfictional

Murmur · SettingsNotifications
Account & security Includes tips, trending posts and offersOn
ActivityOn
Recommended for youOn
PromotionsOn
  • Bundling: marketing rides inside “Account & security”, so turning it off feels unsafe.
  • Nagging: every “off” triggers a plea to reconsider.
  • Defaults: everything starts on; the only break is a one-hour snooze.

✓ Humane alternativefictional

Murmur · SettingsNotifications
Security alerts New logins and password changes. Always on — they protect your account.On
Direct messagesOn
Replies & mentionsOn
RecommendationsOff
PromotionsOff
  • Security separated from marketing, with a reason it can’t be switched off.
  • Low-value categories start off; a daily digest and quiet hours are offered.
  • Turning something off simply… turns it off.

Why teams ship it · incentive

Notifications drive return visits, and return visits are what many products are measured on. Every switch someone turns off is a lost channel, so the settings page becomes a negotiation the user was never meant to win.

What it costs people

Interrupted attention, notification fatigue, and — the worst outcome — people who silence the whole app and then miss the login alert that actually mattered.

What the alternative changes

Categories are organised by what they do for the person, not by what they do for engagement. Batching non-urgent items into a digest respects attention without going silent.

What it honestly costs the business

Fewer pushes, fewer reflexive opens. Possibly more trust in the pushes that remain.

Nuance

Reminders can be among the most humane things an app does: a medication reminder, a bill due tomorrow, a trial about to convert into a paid plan. The alternative to a notification loop is not silence; it is notifications the person chose, can understand and can stop. The interruption counts in this demo are invented assumptions, labelled as such.

Exhibit 06 · Reading-time screen

A feed with a floor

A fictional social feed. Scroll it — or use the “Scroll the feed” button. Does it ever let you finish?

⚠ Trap patternfictional

Scrollbook · For you∞

Posts appear here when JavaScript is available.

  • No stopping cue: more posts load before you reach the bottom.
  • No sense of position: no count, no “caught up”, no time.
  • Ranking for engagement mixes friends with whatever holds attention.

✓ Humane alternativefictional

Scrollbook · Followingsince yesterday

Posts appear here when JavaScript is available.

  • Chronological posts from people you follow, with a count.
  • A clear “You’re caught up” floor; older posts only on request.
  • An optional reading budget that you set, with a gentle signal.

Why teams ship it · incentive

Time spent and sessions per day are common product metrics, and advertising inventory grows with every additional screenful. Removing the end of the feed removes a natural decision point.

What it costs people

Interpretation Time that slides rather than being spent. People often describe this as not having decided to stay so much as not having been asked to leave. We make no claim here about clinical effects.

What the alternative changes

It restores a floor — a point at which the honest thing to say is “that’s everything new”. Continuing becomes a choice rather than a default, and the reading budget belongs to the reader, not the app.

What it honestly costs the business

Shorter sessions for some people. Perhaps more return visits from people who don’t feel worse when they close the app — a hypothesis, not a finding.

Nuance

Infinite scroll is a fine pattern for search results, photo archives and catalogues, where the person is looking for something and knows when they have found it. The trouble comes from combining it with engagement-optimised ranking, autoplay and no stopping cue. WCAG 2.2’s 2.2.2 (Pause, Stop, Hide) also asks that auto-updating content can be paused, stopped or hidden.

04 · Redesign bench

Your turn: fix a screen.

Pick one of the six trap screens, choose the fixes you would ship, and compare the before and after. Some fixes are real; at least one per screen is cosmetic — it looks like an improvement without restoring the choice. There is no score. The comparison describes what changed, what remains, and what each fix honestly costs.

Fixes to apply

⚠ Beforetrap version

✎ Afteryour fixes

05 · How we got here

From “persuasive technology” to “deceptive design”.

The vocabulary has shifted over two decades, and the shift itself is instructive: from computers that persuade, to choices that are architected, to patterns that deceive, to rules that name them.

  1. 2003FrameworkB. J. Fogg’s book Persuasive Technology frames computers as tools for changing attitudes and behaviour, and treats the ethics of that persuasion as part of the field.
  2. 2008FrameworkRichard Thaler and Cass Sunstein’s Nudge popularises “choice architecture”: since every presentation of options influences choices, the designer cannot avoid influencing — only choose how.
  3. 2010EvidenceDesigner Harry Brignull coins “dark patterns”, as the US FTC’s 2022 staff report records. His site, now called Deceptive Patterns, still catalogues types, laws and enforcement cases.
  4. 2018EvidenceGray and colleagues at Purdue analyse practitioner-collected examples and propose five strategies: nagging, obstruction, sneaking, interface interference and forced action.
  5. 2019EvidenceMathur and colleagues crawl roughly 53,000 product pages on about 11,000 shopping sites and find 1,818 dark-pattern instances on about 11% of sites — more often on the more popular sites — plus 22 third-party providers offering such features.
  6. 2022EvidenceThe FTC staff report Bringing Dark Patterns to Light groups common tactics into four kinds: inducing false beliefs, hiding or delaying material information, leading to unauthorised charges, and obscuring or subverting privacy choices.
  7. 2023EvidenceThe European Data Protection Board adopts final guidelines on “deceptive design patterns” in social-media interfaces (February). India’s Central Consumer Protection Authority notifies guidelines naming 13 specified dark patterns (30 November). WCAG 2.2 becomes a W3C Recommendation.

InterpretationNotice what changed and what didn’t. The techniques are old — a shop assistant can confirmshame you in person. What changed is scale and measurement: a single A/B test can tune a screen for millions of people, and a third-party plug-in can install the same countdown timer on thousands of sites. The ethics did not get more complicated; the feedback loops got faster.

06 · The strongest objection

“But all design persuades.”

Yes. That is the strongest objection to this whole page, and it is correct as far as it goes. Framework The core insight of choice architecture is that there is no neutral screen. Something has to be listed first. Some box has to be the default. Even alphabetical order is a decision that helps companies whose names begin with A.

And persuasion often helps. Consider designs that deliberately steer people:

  • Defaults that serve the person. Automatic enrolment in a savings plan with a simple opt-out is the signature example in Nudge: people stay free to leave, but inertia now works for their future selves.
  • Reminders. “Your free trial becomes a paid plan in three days” uses urgency — truthfully, and in the customer’s favour.
  • Protective friction. “You’re about to email 400 people — send?” and the type-the-name-to-delete pattern in Exhibit 02 slow people down exactly where speed causes harm.
  • Salience for safety. Making a security alert louder than a promotion is a deliberate visual hierarchy, and a good one.

So the line this page draws is not “persuasion versus neutrality”. It is closer to: can the person see the influence, refuse it at a cost similar to accepting it, and reverse it afterwards? The door tests do not require the designer to know what is good for someone. They require that the person can find, understand and use the other door.

The paternalism problem, honestly

Even that line has a contested edge. A cooling-off period on a high-interest loan, a forced break in a gambling app, a reading budget that nags you to stop — each overrides a choice “for your own good”. Reasonable people disagree about when that is care and when it is condescension, and the answer can depend on who is likely to be harmed and how badly. This page’s position: protective friction is easiest to defend when it is transparent, proportionate to the risk, and can be switched off by the person it protects.

A counterexample worth keeping

InterpretationHumane design can also fail. Equal-weight buttons on every trivial choice make simple tasks slower; a settings page with forty honest toggles can be as paralysing as one with four dishonest ones; “we respect your choice” banners can become their own kind of noise. Not every decision deserves a ceremony. The goal is proportion, not maximal choice.

07 · Incentives and power

Incentives, not villains.

FictionImagine the team behind Tidepool Plus. Their quarterly goal is to reduce monthly cancellations. They test five cancellation flows. The flow with a required survey, two offers and a phone-only final step “wins” — fewer people completed cancellation. Nobody on the team set out to trap anyone. Each step had a reasonable-sounding story: we need to learn why people leave; we should offer a discount; complicated accounts need a human.

What the dashboard could not see: people who gave up and disputed the charge with their bank; support calls; one-star reviews mentioning “can’t cancel”; friends who were warned off; and the members who stayed, resentfully, until the day a competitor arrived.

InterpretationThis is why blaming individual designers misses most of the problem. Dark patterns are frequently what an optimisation process produces when it is pointed at a narrow metric and nobody measures the costs that fall on users. The evidence also shows the patterns are packaged and sold: Mathur and colleagues found third-party providers offering ready-made features of this kind, which means a small shop can adopt a manipulative countdown without anyone there designing it.

Power changes who pays

The same friction lands differently. A five-step cancellation is an annoyance to someone with time and confidence; to someone working shifts, navigating in a second language or using a screen reader, it can be a wall. When a platform controls the screen, the data and the experiment, and the user controls only their next tap, “they could have just read more carefully” is not a fair account of responsibility.

Counter-metrics worth trying

HypothesisIf a narrow metric creates the trap, broader measurement may help dismantle it. Teams could track, alongside retention:

  • the share of cancellations completed in a single visit, and the median time to cancel;
  • complaints, chargebacks and support contacts that mention cancelling, unsubscribing or consent;
  • how often people who left cleanly come back;
  • the steps-in versus steps-out ratio for every subscription and consent flow.

Whether these actually change team behaviour is an open, testable question — not something this page can confirm.

IllustrationA physical shop makes the point plain. A shopkeeper who locked the door behind each customer would be obviously wrong. Online, the lock is rarely a lock; it is a corridor, a form, a phone number that only answers on weekdays. The corridor is easier to build and much harder to see.

08 · Standards

Accessibility is part of consent.

A choice a person cannot perceive or operate has not really been offered to them. If the “reject” link fails contrast, a person with low vision is shown only “accept”. If a toggle has no accessible name, a screen-reader user cannot tell whether marketing is on or off. Consent architecture inherits every accessibility failure underneath it.

Established frameworkThe Web Content Accessibility Guidelines (WCAG) 2.2 are a W3C Recommendation. Several success criteria map directly onto the traps in the control room:

WCAG 2.2What it asks (paraphrased)Where it bites
1.4.3 · AANormal text needs a contrast ratio of at least 4.5:1.Pale “no thanks” links and grey fine print on consent screens.
1.4.1 · AColour must not be the only way information or actions are distinguished.“Green means safe” buttons; switches whose state is shown only by colour.
2.5.8 · AAPointer targets at least 24×24 CSS pixels, with stated exceptions (new in 2.2).Tiny close “×” buttons on pop-ups and overlays.
4.1.2 · AControls expose their name, role and state to assistive technology.Custom toggles that look on/off but announce nothing.
3.3.4 · AAActions with legal or financial commitments, or that modify or delete user data, are reversible, checked or confirmed.Exhibit 02’s delete dialog; purchases bundled with subscriptions.
2.2.2 · AMoving or auto-updating content can be paused, stopped or hidden.Autoplay carousels, countdowns and self-refreshing feeds.
3.2.6 · AHelp mechanisms repeated across pages stay in the same relative order (new in 2.2).Help and contact links that move or vanish inside cancellation flows.

InterpretationThe converse is just as important: passing WCAG does not make a screen ethical. An accessible, perfectly labelled, high-contrast cancellation maze is still a maze. Standards set a floor for perception and operation; they do not decide what a screen is for.

This page targets WCAG 2.2 AA and has been checked with automated tools and keyboard testing, but it has not had a formal accessibility audit or testing with assistive-technology users. Please report anything that gets in your way.

09 · Misconceptions

Six things people get wrong.

  • Friction is always bad.

    Friction before irreversible or risky actions protects people. The question is where friction is placed and whom it serves — not whether it exists.

  • If it’s in the terms, it’s consented.

    Agreement is not the same as understanding. A disclosure nobody could reasonably find, read or act on does little to make a choice informed.

  • Dark patterns are always illegal.

    Legality depends on jurisdiction, sector and facts. Some practices are clearly prohibited in some places; many sit in grey areas. Ethics and law overlap but are not the same map.

  • Only big tech does this.

    Mathur and colleagues found dark patterns more often on popular shopping sites — and also found third-party providers selling them as features, which spreads them to small sites too.

  • Make the right button green and you’re done.

    Colour is cosmetic. If refusing still takes four screens, a green “Reject” on the fourth screen changes very little. (The redesign bench includes cosmetic fixes for exactly this reason.)

  • Humane design means fewer business results.

    It may mean smaller headline numbers in the short run. Whether it pays off over the long run is a genuine open question; it should be tested rather than assumed in either direction.

10 · Take it with you

Checklists: for makers and for everyone.

Two practical lists. The first is a pre-release review for anyone who designs, writes, builds or approves screens. The second is for anyone who has ever tried to cancel something.

Before you ship a screen

  1. Count the steps in and the steps out. Write down why any asymmetry exists.
  2. For each default: who benefits if it’s accepted? Would you show this choice in a screenshot to the person it affects?
  3. Buttons say what they do (“Delete 14 notes”, not “OK”). No double negatives.
  4. Decline options have equal visual weight and neutral wording. No confirmshaming.
  5. Ask for consent per purpose, off by default for non-essential uses, and make withdrawal as easy as giving it.
  6. Urgency and scarcity claims must be true, specific and checkable — or removed.
  7. Destructive actions name the consequence; prefer undo to warnings; confirm what can’t be undone.
  8. Separate security notifications from marketing. Offer digests and quiet hours.
  9. Give feeds a floor (“You’re caught up”). Avoid autoplay; respect reduced-motion settings.
  10. Test the “no” path with a keyboard and a screen reader, and check its contrast and target size.
  11. Measure what the trap hides: cancellation completion, complaints, chargebacks, support contacts.
  12. Review risky flows with people from design, product, legal and support together — the EDPB guidelines recommend interdisciplinary teams for exactly this.

When a screen is working against you

  1. Before agreeing, look for the second button. It is often smaller, lower or styled as a link.
  2. Scan checkout pages for pre-ticked boxes and untick before paying.
  3. Starting a free trial? Note the conversion date and set your own reminder.
  4. When you subscribe, find the cancellation path straight away and screenshot it.
  5. If cancelling is buried, check your phone’s app-store subscription settings or your payment provider; app-store subscriptions are usually managed there.
  6. Read double negatives slowly. Rewrite them in your head as “I want X: yes or no”.
  7. Switch notifications to a digest where offered — but keep security alerts on.
  8. Decide a stopping point before you open a feed. Your attention is a budget too.
  9. Complain specifically, with screenshots, to the company and to your country’s consumer-protection authority. In India, the CCPA’s 2023 guidelines list named patterns — such as “subscription trap” and “confirm shaming” — that you can cite.

For longer essays on behaviour, culture and the small decisions people make every day, see Sahil’s writing archive on Medium.

11 · Evidence and sources

Follow the sources, and their limits.

Links were checked on 24 October 2026. Each entry says what the source establishes and what it does not. Nothing here is quoted at length; please read the originals.

  1. Gray, C. M., Kou, Y., Battles, B., Hoggatt, J., & Toombs, A. L. (2018). “The Dark (Patterns) Side of UX Design.” Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, Paper 534. doi.org/10.1145/3173574.3174108 Establishes: a widely used five-strategy taxonomy (nagging, obstruction, sneaking, interface interference, forced action) grounded in practitioner examples. Does not establish: how common or harmful each strategy is.
  2. Mathur, A., Acar, G., Friedman, M. J., Lucherini, E., Mayer, J., Chetty, M., & Narayanan, A. (2019). “Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites.” Proceedings of the ACM on Human-Computer Interaction 3 (CSCW), Article 81. doi.org/10.1145/3359183 · arXiv preprint Establishes: measured prevalence on shopping sites in 2019 (1,818 instances across about 11% of ~11K sites), higher prevalence on popular sites, and a market of third-party providers. Does not establish: prevalence today, in apps or non-shopping contexts, or the size of harm to individuals.
  3. US Federal Trade Commission (2022). Bringing Dark Patterns to Light (Staff Report, September 2022). ftc.gov (PDF) Establishes: a regulator’s grouping of common tactics into four kinds, and contexts including e-commerce, cookie banners, children’s apps and subscriptions. Does not establish: legal rules outside the United States; a staff report is not itself a regulation.
  4. European Data Protection Board (2023). Guidelines 03/2022 on deceptive design patterns in social media platform interfaces: how to recognise and avoid them, Version 2.0, adopted 14 February 2023. edpb.europa.eu Establishes: how EU data-protection authorities connect deceptive interface design to GDPR principles, with best practices and a recommendation for interdisciplinary review. Does not establish: rules for non-social-media services or outside the EU; the guidelines say their lists are not exhaustive.
  5. Central Consumer Protection Authority, India (2023). Guidelines for Prevention and Regulation of Dark Patterns, 2023, notified 30 November 2023 under the Consumer Protection Act, 2019. consumeraffairs.nic.in (PDF) Establishes: India’s definition of dark patterns and 13 specified patterns (false urgency, basket sneaking, confirm shaming, forced action, subscription trap, interface interference, bait and switch, drip pricing, disguised advertisement, nagging, trick question, SaaS billing, rogue malware), applicable to platforms, advertisers and sellers. Verification note: the official PDF could not be fetched automatically while building this page (the site blocks automated access), so these details were cross-checked against several independent law-firm and policy summaries. Please open the official document to confirm.
  6. W3C (2024). Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation (current edition dated 12 December 2024). w3.org/TR/WCAG22 Establishes: testable accessibility success criteria, including those tabled above. Does not establish: whether a design is ethical, or legal accessibility obligations in any particular country.
  7. Brignull, H., and contributors. Deceptive Patterns (formerly darkpatterns.org), a public catalogue maintained since 2010. deceptive.design Establishes: a practitioner vocabulary of pattern types with examples, and links to laws and enforcement. Does not establish: prevalence; it is a curated catalogue, not a sample.

Background books (cited from general knowledge, not link-checked for this page): B. J. Fogg, Persuasive Technology: Using Computers to Change What We Think and Do (2003); Richard H. Thaler and Cass R. Sunstein, Nudge (2008). They are referenced here for frameworks, not for specific findings.

13 · An open question

What would change if products were judged, in part, by how well people can leave them?

And a sharper version that is arriving quickly: as software agents begin to click, accept, subscribe and cancel on people’s behalf, a confirmshaming line will not embarrass them, and a five-step exit will not tire them. Some traps may simply stop working. Others — those that exploit what an agent is instructed to optimise — may be harder to see. Who designs the door for a reader that never gets tired?

If you have a counterexample, a better test or a screen worth adding to the control room, Sahil would be glad to hear it.

← Back to the Sahilverse Email Sahil Back to top ↑