How Many Companies Are Detectably A/B Testing, And Across How Many Experiments?
Lazyweb Research detected 4,814 before/after UI experiments across 362 distinct companies within its ~800 tracked-apps corpus [1][2]. Of those, 2,160 area-level annotations cover 1,126 experiments, and the activity is overwhelmingly mobile: 4,449 experiments (92.4%) are mobile versus 365 (7.6%) web [1][3]. Testing is broad but uneven — 362 of ~809 tracked apps show at least one detected experiment [2].
Lazyweb Research detected 4,814 UI experiments across 362 companies, 92.4% of them on mobile — July 2026.
The finding: broad, mobile-dominated activity
The corpus contains 4,814 detected before/after UI experiments across 362 distinct companies [1][2]. Area-level analysis covers a subset: 2,160 annotations over 1,126 experiments [1]. The platform split is lopsided — 4,449 experiments (92.4%) are mobile and 365 (7.6%) are web [3]. So the dataset is primarily a picture of mobile-app experimentation, with web as a thin supplement [3].
The headline denominators
The numbers that anchor every cut in this family [1][2][3]:
| Measure | Value |
|---|---|
| Detected experiments | 4,814 |
| Companies with >=1 experiment | 362 |
| Area annotations | 2,160 |
| Annotated experiments | 1,126 |
| Mobile experiments | 4,449 (92.4%) |
| Web experiments | 365 (7.6%) |
| Paywall CTA experiments | 795 |
Within annotations, 1,280 are mobile (147 companies) and 880 are web (106 companies) [4].
How to apply it
Use these denominators to weight everything else in this family: percentage cuts are honest only against the annotated subset (2,160), and web percentages are unreliable given just 365 web experiments [1][3]. When you cite an area or category share, state the base — e.g. 'of 2,160 annotated experiments' — so the reader knows it isn't drawn from tens of thousands of tests [1].
Caveats
Every experiment is an observed before/after screenshot pair with LLM-inferred rationale, not a measured A/B result [5]. The ~800-app corpus is not a census of all apps, so 'how many companies test' means how many show detectable changes on tracked surfaces [2][5]. Web is thin and gets absolute-count treatment [3].
The numbers
| Stat | Computed from |
|---|---|
| 4,814 detected experiments; 2,160 area annotations; 1,126 annotated experiments; 795 paywall CTA experiments | totals |
| 362 distinct companies with >=1 detected experiment, out of ~809 tracked apps | companies_with_experiments |
| Mobile 4,449 (92.4%) / web 365 (7.6%) of 4,814 detected experiments | platform_split_experiments |
| Annotations: 1,280 mobile (147 companies) / 880 web (106 companies) | platform_split_annotations |
| All experiments are observed before/after screenshots with LLM-inferred rationale, not measured A/B results | universe; smallSampleWarnings |
Sources & citations
- [1] Lazyweb Research analysis of 4,814 detected experiments (~800 tracked apps), July 2026. Headline denominators: experiments, annotations, annotated experiments, paywall CTA subset. ↩
- [2] Lazyweb Research analysis of 362 companies with detected experiments (~800 tracked apps), July 2026. Distinct companies with >=1 detected experiment out of ~809 tracked apps. ↩
- [3] Lazyweb Research analysis of platform split (~800 tracked apps), July 2026. Mobile vs web experiment counts. ↩
- [4] Lazyweb Research analysis of annotation platform split (~800 tracked apps), July 2026. Mobile/web annotation and company counts. ↩
- [5] Lazyweb Research methodology note (~800 tracked apps), July 2026. Experiments are observations, not measured A/B results. ↩
Source: Lazyweb Research — proprietary analysis of real, in-market app screens. Cite as Lazyweb Research, 2026-07-07.