Short answer: Meta's creative testing tool lets you put 2 to 5 new ads into an existing ad set and hold their spend roughly even for a set number of days, so each one gets a fair read instead of Meta piling the money onto whichever ad it liked first. It is useful when you have a proven ad and want to know which new angle beats it. On a small budget it is thin: Meta suggests setting aside no more than 20% of the budget for the test, and at $50 a day that is often not enough for any test ad to reach a purchase. When the test ends, nothing is promoted for you; the ads keep running and the decision is yours.
I work on Everix, and this week the same three questions kept coming up under a new video about the feature: how the account should be set up, how much spend a result needs before you trust it, and what to do with the test ads afterwards. The feature details below come from Jon Loomer's walkthrough of the tool (October 2025) and Meta's own list of edits that restart learning.
What the tool actually does
- You start it at the ad level, from an ad you are editing in an existing or draft campaign. It sits in its own Creative testing section after the ad creative.
- It creates 2 to 5 test ads as duplicates of the ad you are editing, which you then change. The ad you started from is not part of the test.
- You set a test budget: an amount taken out of the campaign or ad set budget for the test. Meta's suggestion is no more than 20%, so the ads that are already working keep most of the money.
- You set how long it runs; 7 days is the default. During that window Meta keeps spend roughly even across the test ads.
- You pick the metric that decides it. Cost per result is the sensible choice for most stores.
- It needs the Highest volume bid strategy. Cost per result goal, bid cap and ROAS goal are not supported. Daily and lifetime budgets both work.
Why it exists: Meta does not split spend evenly
There is no per-ad budget in Meta. Inside one ad set, delivery leans hard on the ad it predicts will do best, often decided in the first day or two, and the other ads get scraps. That is fine when you trust the favourite. It is a problem when you are trying to learn whether a new angle works, because an ad that got $3 of spend has told you nothing. The test tool is Meta's answer to the old workaround of one ad per ad set (CBO vs ABO covers that route), and to the question of how many ads one ad set can really test.
What it costs on a small budget
Do the arithmetic before you start. At $50 a day, 20% is $10 a day, or $70 over the default 7 days. Split across 3 test ads, each gets about $23. If your cost per purchase runs around $30, every test ad ends the week with less spend than one purchase costs, and the 'winner' is whichever one got lucky.
The bar I use to judge any ad is spend of about 2 to 3 times your target cost per result (how much to spend before turning an ad off goes through why). Work backwards from that: 3 ads at 2 times a $30 target is $180 of test spend. On a small budget that leaves three honest options:
- Test fewer ads: 2 that are genuinely different angles, not 5 versions of one.
- Run the test longer than 7 days so the same daily slice adds up to enough per ad.
- Judge on an earlier step: cost per add to cart or cost per checkout started, which arrive several times more often than purchases. Let the survivor prove itself on purchases afterwards.
Raising the test share above 20% also works, but every dollar moved into the test comes out of the ads that are currently paying for themselves.
Does it restart learning?
Meta lists adding a new ad to an ad set among the significant edits that send it back into learning, and the test adds 2 to 5 new ads. Plan for it the way you would when adding ads by hand: put the whole batch in at once rather than one ad every few days, and do not judge the rest of the ad set on the first days after the test starts.
When the test ends
The test ads keep running and delivery goes back to normal: Meta stops holding spend even and allocates the way it always does. It does not hand the budget to the test winner. What you do next is your call:
- A test ad whose cost per result came in above 2 to 3 times your target: turn it off. Clearing out clear losers keeps the next test easy to read.
- A test ad close to target that Meta then stops funding: you can leave it on. It costs nothing while it is not being delivered.
- A clear winner that Meta starves again once the test is over: that is the sign it needs its own ad set, with its own budget, rather than more ads crowded around it.
When to use it, and when to skip it
Use it when you have an ad that is already working and 2 to 3 new angles you want to compare at equal spend, and your budget can fund roughly twice your target cost per result for each of them within the test. Skip it when the ads differ only in a headline or a colour, when the test slice cannot buy each ad a couple of results, or in the week before a sale you cannot afford to have the ad set relearning.
Everix does not use the test tool. Each lane is one broad ad set, Meta's delivery does the allocating, and Everix reads every ad on complete days. An ad that takes a real share of the spend but costs at least twice as much per result as the best ad in the set, with enough results on both sides, for three days in a row is retired. A new ad Meta has barely funded is never retired for that: after three days with almost no spend, Everix tells you Meta is backing the older ads rather than judging the new one, and offers two options, wait a few more days or move it into a small lane of its own. An ad Meta starves that still converts well under your target is proposed as a lane of its own too, which you approve before anything is spent. Only an ad that has had enough spend to be judged and still has no results is cleaned out.