Which Apps A/B Tested Adding A Paywall Close Button?

Lazyweb Research detected 11 paywall CTA experiments across 8 companies that explicitly changed a dismissal affordance — adding or removing a close button, swapping an X for 'Maybe later', or adding a skip button ^[1]. Named examples include SoundCloud adding a close button to its Go+ paywall and Replit repeatedly A/B'ing a dismissible modal-with-X against a full-screen 'Maybe later' ^[2]. These are observed UI changes with inferred rationale, never measured lift ^[1].

11 detected paywall experiments across 8 companies explicitly changed a dismissal affordance — Lazyweb Research, July 2026.

Lazyweb Research · n=11 · Published 2026-07-07

paywallexperimentsmonetizationmobileux-patterns

The finding

Across the corpus, 11 detected paywall CTA experiments from 8 companies changed a dismissal affordance according to their what-changed/learning text — adding or removing a close button, switching an X for a 'Maybe later' link, or introducing a skip button ^[1]. Dismissal is a real, if uncommon, thing teams test ^[1].

Named teardowns

SoundCloud added a visible close button to its Go+ paywall while leaving the offer unchanged — inferred to reduce perceived coercion ^[2].
Replit repeatedly A/B'd a dismissible modal with an X against a full-screen interstitial with 'Maybe later' (and ran the reverse), trading focus against reduced friction — 4 detected experiment rows ^[2].
NGL tested a skip button replacing the X on a dark-mode paywall ^[2].
Photomath paired a stronger 'Try Plus' CTA with a 'Not Now' exit ^[2].
Disney overlaid an exit-intent modal offering 'Finish later' / 'Resume' on a restart-subscription screen ^[2].

How to apply it and caveats

Use these as pattern precedents, not proof: the two canonical soft-paywall dismissal patterns teams actually test are 'dismissible modal + X' and 'full-screen + Maybe later', and Replit ran both directions — a sign neither is universally right ^[2]. Exit-intent modals ('Finish later') are a distinct third pattern for cancellation flows ^[2]. Critical caveat: these are LLM-inferred rationales from observed before/after UI, with n=11/8 companies — treat them as named anecdotes, never as prevalence or lift ^[1].

The numbers

Stat	Computed from
11 dismissal-affordance experiments across 8 companies	dismissal_experiments_n: 11 rows, 8 companies
named: SoundCloud, Replit (4 rows), NGL, Photomath, Disney, Apple Fitness, Rootd, AI chatbot	qualitative teardown entries

Methodology. Experiment count is a strict keyword match over paywall CTA experiment rows (11 rows, 8 companies). All examples are observed UI changes with LLM-inferred rationale; the corpus never measures conversion lift. July 2026.

Sources & citations

[1] Lazyweb Research analysis of 11 paywall CTA experiments (8 companies, mobile app corpus), July 2026. Strict keyword match on what_changed/learning; observed before/after UI with inferred rationale, never measured lift. ↩
[2] Lazyweb Research qualitative paywall teardowns (mobile app corpus), July 2026. Named-company examples of dismissal-affordance changes; inferred rationale only. ↩

Source: Lazyweb Research — proprietary analysis of real, in-market app screens. Cite as Lazyweb Research, 2026-07-07.