Product evidence
Roughly one in three people who see a form never type a single character. They are not lazy — they are budgeting, and your blank fields just asked for an unspecified withdrawal.
The research, distilled
In the jam study, the 24-jar display pulled in more traffic than the 6-jar one (60% vs 40% stopped) but converted about ten times worse (roughly 3% vs 30% of stoppers bought). The detail nobody quotes: both groups tasted the same number of jams — 1.50 versus 1.38. Nobody explored the big assortment and got tired. They priced the job by looking at it and left.
Iyengar & Lepper (2000), Journal of Personality and Social Psychology 79(6)When a company switched its 401(k) to automatic enrolment, participation went from 37% to 86%. The part almost nobody cites: 61% of auto-enrolled employees stayed on both the default contribution rate and the default fund — a combination only about 1% of self-selecting employees ever chose for themselves.
Madrian & Shea (2001), Quarterly Journal of Economics 116(4)Across 93 million form views, the vendor's own conclusion was that the number of inputs is not the primary factor — and cases exist where adding fields raised conversion. Prefilling removes the decision without removing the field, which is why the rebuild below keeps four of five inputs and still converts better.
Zuko Analytics telemetry (739 forms, 93,022,997 views)Placeholder-as-label fails in seven documented ways: the hint vanishes as you type, entries can't be verified before submitting, error correction means deleting your answer to re-read the question, and grey-on-white routinely fails contrast requirements. Labels belong above the field, and they stay there.
Nielsen Norman Group, form design usability researchEverything in this deep-dive
This topic was published before we started running our own screen audits, so it has two script versions: the original, preserved unchanged, and v2 with the audit data folded in.
Original data
We sampled 64 form screens across 60 top iOS apps via Mobbin — signup, booking, checkout, and profile setup — and tallied every one by hand. The split between what the industry does for its bookings and what it does to its own signups is the finding.
The whole thesis on one screen: stepper preset to 1 adult, first slot preselected, price and scarcity ('$25/guest', '10 spots left') on every option, and a CTA showing the live payoff '$25 for 1 guest'.
Zero-typing checkout: email, address, shipping method, and payment all prefilled; 6 payment options as a fully visible radio list; the CTA dynamically names your chosen method ('BUY WITH CREDIT CARD').
Nothing hidden: full visible calendar, guest number chips, segmented date control, '2' guests and today preselected, and an always-active 'Search for tables' CTA.
Auto-generates a username ('sacramentoUser1859419007') with a clear-X, so the naming decision is optional rather than blocking — a smart default applied to onboarding, where they are rarest.
Seven live address suggestions appear while typing; one tap replaces four fields. The frontier pattern: eliminating the field instead of decorating it.
Date, guests, ticket option, and time slot all preselected, with visible payoff at the CTA: '1 option available' and 'Total: $51.30 (2 Adults x $25.65)' — one of only three live-payoff CTAs in the sample.
The only signup in the sample with smart defaults — country preset to 'Singapore' and dial code '+65' from locale — proving it's possible. Yet it still stacks 6+ fields and a 4-bullet password-rule wall on one screen.
A wall of six consecutive dropdowns, but the content is regulator-mandated and resists chip-ification. Its header copy is the sample's only honest acknowledgment of form fatigue.
Admirably light — one text field, photo upload marked optional, active CTA — but pre-checks 'I agree to Notion marketing communications', a default deployed for the company rather than the user.
Inverts the dark pattern by pre-checking 'I don't want to receive marketing emails' — a privacy-positive opt-out default — but still ships placeholder-as-label fields, one with no visible label at all.
Blank free-text signup with no app-supplied defaults, yet finds room for one preselection: the marketing opt-in checkbox arrives pre-checked. Defaults deployed only where they benefit the company.
Placeholder-as-label on underlined fields, 'Send me promotions' pre-checked, and a grayed disabled-until-complete CTA — the blank signup wall with a dark-pattern garnish.
The churn-surface archetype: nine free-text fields (even State), every label a disappearing placeholder, no defaults, and a grayed 'Subscribe' CTA disabled until all nine validate.
Hides a ~10-team fan choice behind a dropdown where chips would show the whole grid, doesn't locale-default the country select, and grays the CTA until complete.
Textbook NN/g failure, shipped: once its placeholder-labeled dropdowns are filled, 'Senior' and 'Social media' sit as answers with no visible questions — plus a 'how did you hear about us' survey field.
Six placeholder-labeled pill fields with format masks crammed into the placeholders ('Phone number (XXX) XXX-XXXX*', 'Birthday (MM/DD)') and a disabled CTA; only the birthday field's 'So we know when to celebrate' shows payoff thinking.
How to read this data: Curated, relevance-ranked sample — not random. Source: Mobbin design database via MCP (search_screens, iOS), 5 queries at limit 12-14 each, deduplicated, collected 2026-08-10. Unit of analysis is the screen, not the app: 64 unique screens across 60 distinct apps (Marriott Bonvoy, Best Buy, Artsy, On, and Turo appear twice in different flows). Mobbin indexes design-notable top apps and returns best matches per query, so the sample skews toward well-regarded design; batch 5 was deliberately seeded with 'dropdown', inflating the overall dropdown count (the excluding-batch-5 figure is reported alongside). Only visible evidence was tallied; ambiguous filled values were conservatively counted as not prefilled, so the 47% smart-default figure is a floor for visible-prefill states. Several CTAs were cut off by the capture viewport and excluded from CTA denominators (n=46). Percentages on n=64 carry roughly +/-12pt margins at 95% confidence; treat them as descriptive of this sample. Full per-item observations are in the audit.
Apply it to your app
This prompt distills everything above into instructions for an AI coding session (Claude Code, Cursor, or similar). It interviews you about your app first — so nothing changes until it understands your context — then audits against the research and implements the fixes with your design system.
You are a senior product engineer applying peer-reviewed research on decision fatigue and choice architecture to the forms in my app (signup, booking, checkout, settings — any screen that asks for input). The rules below come from primary sources: the Iyengar & Lepper jam study read correctly (the overload was anticipatory — people priced the job by looking at it), Chernev's 2015 meta-analysis moderators, Madrian & Shea's 401(k) defaults research (61% stayed on a default combination only ~1% would choose freely), Zuko's 93M-form-view telemetry (roughly 1 in 3 viewers never start; the password field is the highest-abandonment field), and NN/g's placeholder-label findings.
BEFORE YOU CHANGE ANYTHING, ask me these questions and wait for my answers:
1. Which form(s) should we improve first, and what does a successful submission unlock for the user?
2. What platform and stack is this (web/iOS/Android; framework, form library, design system)?
3. Do you have data on the most common value users pick for each field (or a sensible default per field)? If not, what's your best guess per field?
4. Which fields are required by the business vs. genuinely required to deliver the outcome? For each field, can you name what changes downstream based on the answer?
5. Any compliance constraints (GDPR/consent, regulated data) I should know about?
THEN audit the current form against these rules and show me a table (field → issue → fix → evidence) before writing code:
- Never show an empty field you could have filled. Prefill every field with the most common answer, visibly changeable in one tap/click. Defaults are read as advice — so default to the most COMMON value, never the most profitable one (that's the dark-pattern line; in the EU, pre-ticked consent is legally void — CJEU Planet49).
- Convert decisions to confirmations: replace free-text with structured controls where option counts are small — steppers for quantities, visible chips/segments for 2–5 options (never a dropdown that hides 3 options), smart date shortcuts ("Tomorrow", "Fri 7pm").
- Delete any field where nothing changes based on the answer. If it's analytics-only, move it out of the critical path (progressive profiling after activation).
- Persistent labels above fields; placeholder text is never the only label (users can't verify entries; eyes are drawn to empty fields).
- Keep the CTA alive and show the payoff in it ("See 12 available tables"), not a dead disabled button.
- Reduce password friction: offer Sign in with Apple/Google where appropriate (password fields have the highest field-level abandonment; Baymard: up to 19% of existing-customer checkout abandonment traces to password resets). If you add Google sign-in on iOS, Apple guideline 4.8 requires an equivalent privacy-preserving option.
- Multi-step only if each step gets lighter; field count matters less than decision count.
THEN implement: propose the change list ranked by expected-impact-vs-effort, get my approval, implement using my existing design system and components, and keep each change small and reviewable. Where my platform supports it, wire the changes behind a feature flag and tell me exactly which metrics to compare (form view→start rate, start→completion, per-field abandonment, and downstream retention — not just completion).
Finally, list anything you wanted to do but couldn't verify was safe from my answers, so I can decide.
See it, click it
Verre, a fictional restaurant app. On the left, five empty fields and a dead button; on the right, the same booking pre-answered and changeable in a tap. Open the pickers, change the party size, and watch the CTA count update.
From Build With Kris
Subscribe to catch the teardown when it drops — the science, the audit, and the rebuild, screen by screen.