What Are Companies A/B Testing on Their CTA Buttons?
CTA is the fifth most-tested area: 232 of 2,160 annotations (11.0%) across 225 experiments Lazyweb Research observed by July 2026. It splits 141 mobile / 91 web, with an average model impact of 3.16/5 and 27.2% of tests scored 4+/5 — mid-pack. On mobile paywalls specifically, the CTA text itself changed in 344 of 795 paywall-CTA experiments (43.3%). Observed web moves swap the whole hero action, not just the button label. [1]
The CTA is tested in 232 of 2,160 annotations (11.0%); on mobile paywalls the button text itself changed in 43.3% of 795 CTA experiments — Lazyweb Research, July 2026.
How much, and where
CTA testing at a glance: [2]
| Metric | Value |
|---|---|
| Annotations | 232 (11.0% of 2,160) |
| Distinct experiments | 225 |
| Mobile / web split | 141 / 91 |
| Avg impact (1–5) | 3.16 |
| Share scored 4+/5 | 27.2% (63/232) |
A separate paywall-CTA dataset (795 mobile paywall experiments across 146 companies) shows how often the wording itself moves: the CTA text changed in 344 experiments — 43.3%. So more than half the time the button changes without its label changing, meaning teams test placement, color, and size as much as copy. [3]
Observed patterns
From CTA experiments Lazyweb Research scored impact 5: [4]
- Replace the whole hero action. Factory swapped a CLI curl-install widget for a single 'START BUILDING' button and removed LOG IN from the nav — widening activation for enterprise buyers who aren't terminal-ready.
- Shift the funnel stage. Cycle switched its persistent nav CTA from 'Book a demo' to 'Start free trial,' moving from sales-led to product-led so self-serve teams reach the product immediately.
The pattern: the highest-impact CTA tests change what the button asks for (or removes competing actions), not just how it looks. [4]
How to apply it and caveats
The mid-pack impact score is a caution: a label tweak alone rarely lands high on the model's scale. The stronger observed plays change the funnel stage the CTA commits to (demo → trial) or remove a competing primary action. Since 43.3% of paywall CTAs change text and the rest change form, test both text and treatment. Caveat: impact is a relative model score on observed diffs, not measured lift; the 43.3% figure counts text changes, not wins. [5]
The numbers
| Stat | Computed from |
|---|---|
| CTA 232 annotations (11.0%), 225 experiments, 141 mobile / 91 web, avg impact 3.16, 27.2% scored 4+ | statpack area_CTA + high_impact_share_by_area |
| CTA: 232 annotations, 225 experiments, avg impact 3.16, 63/232 scored 4+ | statpack area_CTA + high_impact_share_by_area |
| 795 paywall CTA experiments across 146 companies; CTA text changed in 344 (43.3%) | statpack paywall_cta_experiments |
| CTA impact-5 examples: Factory ('START BUILDING' vs CLI widget), Cycle ('Book a demo'→'Start free trial') | statpack qualitative CTA entries |
| Impact is a relative model score on observed diffs, not measured lift; 43.3% counts text changes not wins | statpack methodology + paywall_cta_experiments |
Sources & citations
- [1] Lazyweb Research analysis of 232 CTA annotations (225 experiments) within 2,160 total, July 2026. CTA: 141 mobile / 91 web; avg impact 3.16. ↩
- [2] Lazyweb Research analysis of 232 CTA annotations, July 2026. 63 of 232 scored 4+/5. ↩
- [3] Lazyweb Research analysis of 795 paywall CTA experiments (146 companies), July 2026. CTA text changed in 344 of 795 (43.3%). ↩
- [4] Lazyweb Research qualitative review of top-impact CTA experiments, July 2026. Factory and Cycle — single observations scored impact 5 by the model. ↩
- [5] Lazyweb Research methodology note, July 2026. Impact is a relative model score; the 43.3% figure counts text changes, not measured wins. ↩
Source: Lazyweb Research — proprietary analysis of real, in-market app screens. Cite as Lazyweb Research, 2026-07-07.