Research

A/B test data across 1,000+ brands

Live aggregates from ABWatcher’s detection database.154 A/B tests observed at 104 brands, 11 in the last 30 days. Methodology below.

Top-line numbers

Total detections
154
Brands with ≥1 detection
104 of 1,032
Detections in last 30 days
11
Coverage period
5/12/2026 – 11/22/2026
Adoption rate (brands testing / brands monitored)
10.1%
Data as of
11/22/2026

Which surfaces get tested most?

Across every detected A/B test in the pool, hero leads with 52 detections. Full distribution:

A/B test detections by surface, all-time across the monitored pool
SurfaceDetectionsShare
hero5233.8%
layout3623.4%
form1711.0%
pricing149.1%
copy127.8%
cta117.1%
navigation85.2%
social_proof31.9%
other10.6%

Which A/B testing tools are used most across the pool?

No vendor detections yet.

Methodology

ABWatcher crawls each brand in the monitored pool from three isolated browser sessions with varying fingerprints (user agent, viewport, geo). We flag an A/B test candidate when any of three signals fire: a vendor SDK is actively serving variants, the DOM differs materially across the three same-run visits, or the week-over-week diff against last week’s snapshot exceeds a noise threshold on the hero, pricing, CTA, or form area. Every candidate is analysed by a vision model against a fixed prompt; analyses that fail the quality gate (or that trip a “no meaningful change” heuristic) are excluded from the aggregates on this page.

Sample: the 1,032 brands in the ABWatcher monitored pool at the time of query. Full company list is browsable at /companies.

Common questions about this data

How often is this data updated?

The aggregates on this page recompute hourly from ABWatcher's live detection database. The "as of" date at the top of the key-facts block reflects the last recomputation.

Where can I cite these numbers?

Cite them as "ABWatcher research, 11/22/2026" with a link to /research. The Dataset schema on this page makes the numbers machine-readable, and the methodology section covers what "an A/B test" means in this dataset.

Why do these numbers differ from published benchmarks?

Two reasons. First, this dataset is only tests we've actually detected in the wild — a lower bound. Any test on internal pages, mobile apps, or behind auth is invisible to us. Second, most published CRO benchmarks come from a testing tool's own customers (a self-selected group); this dataset is a curated pool of 1,000+ brands regardless of what tool they use.

How can I see the raw detections?

Every detection is browsable at /companies/[brand-domain]. The public per-brand pages show the detection date, page, surface, and what changed. The full feed across the pool is at /blog for detections we've written up in depth.