Skip to content
Log inGet started

What Are Companies A/B Testing on Their CTA Buttons?

  • By Ali Abouelatta
  • n = 232
  • Published
  • Updated
  • 2 min read
  • checkout · paywall · experiments · ux-patterns · web · mobile
Read as Markdown

Short answer

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. source[1]

The quotable stat
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.

Bar chart · Value

Value — How much, and where

One block is about 8%

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)

How much, and where

CTA testing at a glance: source[2]

MetricValue
Annotations232 (11.0% of 2,160)
Distinct experiments225
Mobile / web split141 / 91
Avg impact (1–5)3.16
Share scored 4+/527.2% (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. source[3]

Observed patterns

From CTA experiments Lazyweb Research scored impact 5: source[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. source[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. source[5]

Real in-market screens

Screens from real, in-market apps, tracked by Lazyweb Research.

  • Airchat Sign up screen — real in-market example tracked by Lazyweb Research
    Airchat1206 × 2622opens the full image
  • Spothero Account Login screen — real in-market example tracked by Lazyweb Research
    Spothero1206 × 2622opens the full image
  • Onx Hunt All plans screen — real in-market example tracked by Lazyweb Research
    Onx Hunt1179 × 2556opens the full image
  • Draftkings Login screen — real in-market example tracked by Lazyweb Research
    Draftkings1206 × 2622opens the full image

The numbers

#StatComputed from
[1]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
[2]CTA: 232 annotations, 225 experiments, avg impact 3.16, 63/232 scored 4+statpack area_CTA + high_impact_share_by_area
[3]795 paywall CTA experiments across 146 companies; CTA text changed in 344 (43.3%)statpack paywall_cta_experiments
[4]CTA impact-5 examples: Factory ('START BUILDING' vs CLI widget), Cycle ('Book a demo'→'Start free trial')statpack qualitative CTA entries
[5]Impact is a relative model score on observed diffs, not measured lift; 43.3% counts text changes not winsstatpack methodology + paywall_cta_experiments

Methodology

Universe: 232 CTA annotations across 225 detected experiments (subset of 2,160), plus a 795-row mobile paywall-CTA dataset (146 companies), July 2026, LLM-labelled from paired screenshots. Named companies are single observations; impact is relative signal, not measured lift.

Sources

  1. [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.

    Back to text

  2. [2]

    Lazyweb Research analysis of 232 CTA annotations, July 2026

    63 of 232 scored 4+/5.

    Back to text

  3. [3]

    Lazyweb Research analysis of 795 paywall CTA experiments (146 companies), July 2026

    CTA text changed in 344 of 795 (43.3%).

    Back to text

  4. [4]

    Lazyweb Research qualitative review of top-impact CTA experiments, July 2026

    Factory and Cycle — single observations scored impact 5 by the model.

    Back to text

  5. [5]

    Lazyweb Research methodology note, July 2026

    Impact is a relative model score; the 43.3% figure counts text changes, not measured wins.

    Back to text

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

Explore the evidence

Keep reading

Related questions

  1. What Are Companies A/B Testing on Their Hero Sections Right Now?

  2. Which Page Areas Get A/B Tested Most — Hero, Pricing, or CTA?

  3. Which Experiment Areas Look Highest-Impact in Detected A/B Tests?

  4. What Are Companies A/B Testing on the Offer (Discounts and Trials)?

  5. What Are Companies A/B Testing on Pricing Right Now?

  6. What Are Companies A/B Testing on Value Props and Feature Lists?

  7. Hero vs Pricing: Which Should I A/B Test First?

  8. How Many Companies Are Running Detectable UI Experiments, and Who Tests Most?

More in What companies A/B test, by page area →