What e-commerce brands' own A/B tests on Meta show
Published 13 Nov 2026. Data read Meta's EU reach read on 2026-11-12; groups and figures measured 2026-11-13 00:53-00:55 UTC. Source: Meta's Ad Library (EU transparency data), read by Poseidon, unless the text names another source.
In short, as read on 13 November 2026. Poseidon found 2,522 tests where an e-commerce brand put the same ad text on different videos, launched within 14 days of each other. In the 425 tests where Meta's EU reach is known for at least two of the videos, the best one reached a median 3.2 times more people than the weakest. In one test in three, the videos stayed within 2 times of each other.
None of the tools in our comparison shows brands' own tests on its public pages. Here is what the tests say, and what they can't.
What counts as a test here
- A test is a set of ads from one brand, launched within 14 days, with the same text (90% alike counts as the same) and different videos (a different preview image).
- Reach is Meta's own count of people in the EU who saw each ad, read by Poseidon on 12 November 2026. It's the only number of people on this page.
- A comparable test has Meta's reach for at least two of its videos. Poseidon reads reach once an ad has run 21 days, so a video switched off before that has no reach. That leaves 425 of the 2,522 tests comparable. Of these 425, 280 have a clear outcome through the reach gap and 145 stayed within 2×; the 282 video tests with a clear outcome add 2 decided by a stop, where reach was read for only one video.
- A clear outcome, Poseidon's label, means the best video reached at least twice as many people as the weakest, or one video is still running while another stopped. 282 tests have one. Across all three kinds of test, 288 have one: 278 through the reach gap and 10 through a stop (counted 13 November 2026). Most tests without an outcome simply have no reach read yet: in 1,944 of them, none of the versions has one.
How far apart the videos end up
Best video's EU reach divided by the weakest's, across the 425 comparable tests:
| Tests | Median gap | Within 2× | 3× or more | 10× or more | |
|---|---|---|---|---|---|
| All comparable tests | 425 | 3.2× | 33.4% | 52.2% | 22.6% |
| Videos launched the same day | 214 | 2.8× | 37.9% | 46.7% | 19.2% |
| Videos launched on different days | 211 | 3.7× | 28.9% | 57.8% | 26.1% |
- Half the time, one video reaches three times more people than another with the same words, from the same brand.
- Same-day tests are the fairest comparison, because every video had the same time to gather reach. There the median gap is 2.8×. When the videos started on different days, part of the gap is just time.
More versions, wider gaps
| Videos with a known reach | Tests | Median gap | Within 2× | 10× or more |
|---|---|---|---|---|
| 2 | 210 | 1.8× | 56.2% | 4.8% |
| 3 | 83 | 4.1× | 22.9% | 20.5% |
| 4 or more | 132 | 10.8× | 3.8% | 52.3% |
With two videos, more than half the tests end within 2× of each other. With four or more, half end 10 times apart or more. Part of this is arithmetic: the more videos you compare, the further apart the best and the weakest can be. It still means a brand that tests four videos on the same text usually finds one that reaches far more people than the rest.
Brands rarely switch off the weaker video
Of the 282 tests with a clear outcome, 270 still had every video running when Poseidon last read them. In 9, some videos had stopped. In 3, all had stopped. The gap comes from Meta showing one video much more than the others inside the same test, not from the brand cutting the losers.
Four tests, linked
Each launched all its videos on the same day, so time doesn't explain the gap. EU reach is Meta's total, read 12 November 2026.
| Shop | Main country | Videos | Best video | Weakest video | Gap |
|---|---|---|---|---|---|
| kijimea.it | Italy | 4 | 5,528,270 · ad | 32,131 · ad | 172× |
| lanuvi.com | Germany | 9 | 800,212 · ad | 3,196 · ad | 250× |
| nootri.pl | Poland | 4 | 364,953 · ad | 914 · ad | 399× |
| offre.glow25.fr | France | 5 | 277,180 · ad | 275 · ad | 1,008× |
Watch both videos of a pair on Meta: same brand, same day, same words.
Where the tests come from
The 282 tests with a clear outcome come from 116 shops. By the main country of the best video (the country holding the largest part of Meta's breakdown): France 58, Germany 51, Italy 48, Poland 47, Spain 31, and 47 in other countries. The shop with the most tests, nootri.pl, has 28 (10%).
What this can't tell you
- Sales. Reach is how many people saw an ad, not who bought. Meta publishes no spend or sales for commercial ads.
- Why a video won. This page compares reach, not what is in the videos.
- The tests brands cut early. A video stopped before 21 days has no reach in Poseidon, so 2,097 of the 2,522 tests can't be compared yet. If brands mostly cut their losers early, the gaps here understate the real ones.
- Text and hook tests. Poseidon found 115 tests of the same video with different texts (4 with a clear outcome) and 4 tests of a different first line (2 with an outcome). That's under 30, so no pattern is shown. They're listed one by one in the brands' own A/B tests.
How this was measured
- Source: Meta's EU transparency data on each ad (reach), read by Poseidon on 12 November 2026. Tests found by Poseidon's test finder (method "groups-v1": 14-day window, texts 90% alike are the same, the same preview image is the same video). Figures computed on 13 November 2026 between 00:53 and 00:55 UTC.
- Which brands: e-commerce shops that passed Poseidon's shop check.
- Gap: the highest EU reach among the test's videos divided by the lowest, each video counted once (its best-reaching copy).
- Minimum sample: no figure is drawn from fewer than 30 tests.
FAQ
How many videos should I test on the same text?
This page can't answer that for your brand. It shows that among brands testing four or more videos, half found one that reached at least 10 times more people than the weakest.
Does the winning video sell more?
Unknown. It reached more people because Meta showed it more. That is usually a sign it performs better for the brand, but it isn't a sales figure.
Can I see the tests for my niche?
Yes. Ask Poseidon "which videos won brands' own tests in weight loss in France?", or ask the same from your AI assistant through Poseidon's MCP server.
Ask Poseidon which version won. askposeidon.app