How many ad creatives should you test at once on Meta?

How many ad creatives should you test at once?

Three to five per ad set if you spend under $50 a day, and never more than five, the number Meta’s own structure guidance gives.^[1] At $50 to $100 a day, run two ad sets of three to five, or one of five on a more frequent event. The bound is results: an ad set usually settles after about 50 in the week after its last significant edit,^[3] and Meta’s ad-volume page says that when an advertiser runs too many ads at once, each delivers less often, fewer exit the learning phase, and more budget is spent before the system can optimize.^[4]

Count concepts, not files. A concept is a different idea: a different problem, proof, person or offer. A variation is the same idea in another crop or color. Meta’s December 2025 note on creative diversification draws the line the same way: iteration “might produce two ads with identical visuals, but different text CTAs”, diversification “two distinctly different pieces of creative”.^[7] Five of the first kind is one test. Five of the second is five.

What does Meta say about ads per ad set and the learning phase?

Meta gives the number in two places and the mechanism in a third. Its page on simplifying ad set structure says, in the step on adding creative: “We recommend only including images or videos that are distinct from each other. We recommend running no more than five ads per ad set for optimal delivery.”^[1] A post on Meta’s Government, Politics and Nonprofits blog repeats “Run no more than five ads per ad set for optimal delivery” and adds that “High ad set volume is a leading contributor to ad sets not exiting the learning phase.”^[2]

The mechanism is the learning phase. Meta’s help page says an ad set exits it “after about 50 results in the week after the ad set’s last significant edit”, and that performance is less stable until then.^[3] Its page on significant edits lists “Adding a new ad to your ad set” and “Any change to ad creative” among the edits that restart it,^[6] so dripping in a new ad every day keeps the set learning for weeks. Its ad-volume page says to “Decrease ads per ad set, but maintain diverse creative assets per ad set”, pointing to multiple text optimization and dynamic creative, where one ad holds up to 10 assets, so variations live inside one ad rather than as extra ads.^[4]

What did Andromeda change?

Andromeda made the creative the thing Meta’s system reads first. In a December 2024 engineering post, Meta described it as the retrieval stage of its ads system, which selects “from tens of millions of ad candidates into a few thousand relevant ad candidates”.^[8] Meta’s December 2025 note on creative diversification says what that means for an account: Andromeda keeps “only the most relevant and distinct creatives”, and the system “groups together ads that share key visual and thematic attributes”. Ads that look or feel alike “are seen as variations of the same creative”, with learnings shared at the creative level.^[7]

So near-duplicates in one ad set are one bet, not five, and they compete for the same slot. Meta now measures this: a help page describes a Creative diversity metric, marked “in development”, that compares the images and video thumbnails in an ad set and labels it High, Medium or Low,^[9] and Account Insights flags “creative similarity” when images or videos “appear too visually identical”, read from static ads active in the last 28 days.^[10]

Concepts or variations: what counts as a different ad?

A different ad is one Meta’s system would not group with the others: a different idea, not a different finish. Meta’s own example of diversification is one ad built for Gen Z in Reels and another made with a creator who speaks to new mothers; its example of iteration is the same visual with a different call-to-action button.^[7] 303 London’s test is blunt: if you removed the text overlays, would your active ads all look the same?^[12]

The numbers that circulate are practitioner guidance, not Meta’s. Segwise’s 2026 playbook suggests 8 to 12 distinct concepts per campaign with 2 to 3 variations each, and a “Creative Similarity Score” under 40%, a figure it credits to 303 London.^[11] 303 London’s own guide gives no percentage; it suggests five concepts a week and, for a new account, “3-5 distinct concepts in one ad set”.^[12] Meta has published no threshold and no concept count.^[9]

A worked example for a $40 sleep supplement. Concept one: the 3 a.m. wake-up (problem). Two: a customer’s review in her own words (proof). Three: the ingredient list beside a rival’s (comparison). Four: the founder on why it exists (origin). Each gets a 4:5 and a 9:16 crop, placement versions of one ad, not extra ads. Four ads in one ad set, each a different idea in a different layout: four bets, not one.

A service business gets the same split from its own calendar. The plumber, HVAC and med spa pages each list the ad ideas for that trade, with the offer and the photo each one needs, so the four bets can be four different jobs rather than four crops of one picture.

The budget arithmetic: how many results does each creative get?

Divide the week’s spend by your cost per result, then by the number of ads, and most small accounts cannot read more than three ads in a week. Meta’s threshold is for the ad set, about 50 results in seven days,^[3] but you judge ads one at a time, and an ad with four results tells you nothing. Our rule, not Meta’s: ten results per ad before you call it.

Under $50 a day, optimizing for a $25 purchase leaves one readable ad a week and an ad set that never reaches 50, which is why Meta’s fixes for a learning-limited set are a bigger budget, fewer ad sets, or an event that happens more often, such as add to cart.^[5] The week is the unit: judging an ad on its first two days judges the learning phase, not the ad. How long to run Facebook ads before judging covers when to call it.

A testing structure for a $30 to $100 a day account

1. 1

   Pick an optimization event you can afford

   Divide the week’s spend by your cost per result. If it is under 50, Meta’s advice is a more frequent event, a bigger budget or fewer ad sets, not more ads.^[5]
2. 2

   Write three to five concepts, not variations

   Each a different problem, proof, person or offer. If two would look the same with the text removed, merge them.^[12]
3. 3

   Put them in one ad set with one broad audience

   No more than five ads, each distinct.^[1] Use asset customization for the 4:5 and 9:16 crops so they stay one ad.^[4]
4. 4

   Leave it alone for seven days

   Adding an ad or changing creative restarts learning.^[6] Collect the next batch; do not drip it in.
5. 5

   Read the ads with ten or more results, then swap the losers in one edit

   Replace the weakest concepts with new ones in a single edit, so the set re-enters learning once. Keep the winner running.
6. 6

   Make the next batch different from the winner, not like it

   Meta says look-alike ads share one learning path,^[7] so a recolor of the winner adds a file, not a test. Showfolk returns four varied versions of every ad, each a different concept rather than a recolor, so a small account can test real alternatives. How the AI ad generator does it.

Questions people ask

How many ads should be in a Facebook ad set?

No more than five, by Meta’s own guidance, and only images or videos that are distinct from each other. Under $50 a day, three to five is the practical number, because every ad added splits the roughly 50 weekly results the ad set needs to leave the learning phase.

Is it better to test creatives in one ad set or in separate ad sets?

One ad set, for a small account. Meta says that when similar ad sets run at the same time each gets fewer opportunities to learn, and that high ad set volume is a leading reason ad sets never exit the learning phase. Split only when each ad set can reach about 50 results a week on its own.

Does Meta penalize similar ads?

Not with a published penalty. Meta says its system groups ads that share key visual and thematic attributes and treats them as variations of the same creative, so near-copies share one learning path instead of testing several ideas. Meta flags creative similarity in Account Insights but has published no threshold.

What is the 40% creative similarity rule?

Vendor guidance, not Meta’s. Segwise recommends keeping a creative similarity score under 40% and credits the figure to 303 London, whose own guide gives no percentage. Meta’s help page labels an ad set’s creative diversity only High, Medium or Low, and says the metric is still in development.

How many ad creatives does Meta’s Andromeda want?

Meta has not given a number. Its December 2025 guidance asks for ads that are truly distinct in visuals, messaging and formats, and its structure guidance still says no more than five ads per ad set. The 8 to 12 concepts figure and the 25-creative test come from agencies; neither is on a Meta page we could find.

Sources

Every figure above points at one of these. Facts were last checked on October 7, 2026.

1. [1]Meta for Business: Simplify your ad set structure. “We recommend only including images or videos that are distinct from each other. We recommend running no more than five ads per ad set for optimal delivery.”
2. [2]Meta for Government, Politics and Nonprofits: Smarter campaigns: how machine learning drives ad performance on Meta technologies. “Run no more than five ads per ad set for optimal delivery” and “High ad set volume is a leading contributor to ad sets not exiting the learning phase”
3. [3]Meta Business Help Center: About the learning phase. Ad sets usually exit after about 50 results in the week after the last significant edit; performance is less stable during it
4. [4]Meta Business Help Center: About managing ad volume. Too many ads at once means each delivers less often and fewer exit the learning phase; decrease ads per ad set but keep diverse assets; one ad can hold up to 10 creative assets; asset customization; the June 2024 dynamic creative note
5. [5]Meta Business Help Center: About learning limited. Causes include running too many ads at the same time; fixes are combining ad sets, raising budget, or choosing a more frequent optimization event such as add to cart
6. [6]Meta Business Help Center: Significant edits and the learning phase. “Adding a new ad to your ad set” and “Any change to ad creative” restart the learning phase
7. [7]Meta for Business, December 16, 2025: Demystifying creative diversification. Iteration vs diversification, the Andromeda sentence, ads that look or feel alike “are seen as variations of the same creative”, creative fatigue vs creative similarity, the Gen Z and new-mothers example
8. [8]Engineering at Meta, December 2, 2024: Meta Andromeda: supercharging Advantage+ automation with the next-gen personalized ads retrieval engine. Retrieval selects a few thousand candidates from tens of millions; +6% recall and +8% ads quality
9. [9]Meta Business Help Center: Creative diversity. The metric compares images and video thumbnails in an ad set, labels High, Medium or Low, and is marked in development
10. [10]Meta Business Help Center: About Account Insights in Meta Ads Reporting. Creative similarity is when images or videos appear too visually identical; based on static ads active in the last 28 days
11. [11]Segwise: Meta Andromeda update 2026, creative strategy playbook. Vendor guidance: 8 to 12 concepts with 2 to 3 variations, similarity under 40% (credited to 303 London), and the 25-creative test (credited to Logical Position’s playbook)
12. [12]303 London: Complete guide to creative diversity for Meta Andromeda. Agency guidance: the text-overlay test, five concepts a week in two or three formats, 3 to 5 concepts in one ad set to start; no similarity percentage on the page