TL;DR
- Agency executives told Marketing Brew on 28 September 2026 that Meta will not rank an ad as highly if it shares the same look or opening hook as the advertiser's other creative.
- One agency executive said two videos of the same person that differ only in on-screen text count as the same creative.
- Meta spokesperson Alisha Swinteck called creative diversification a longstanding best practice and said Meta's guidance is outcome-based, not model-specific.
- Meta has tested creative insights metrics in its Insights API with a small number of agencies/GCR resellers since June 2026; they are not widely available.
- On Meta's July 2026 Q2 call, CFO Susan Li said user-behaviour model changes plus the Gem model raised ad clicks 8.3%.
Meta has given advertisers updated best practices for getting ads ranked highly by its models, including specific advice to diversify creative, Marketing Brew reported on 28 September 2026. According to ad executives who spoke to the outlet, Meta's systems will not rank an ad as highly if it shares the same look or the same opening hook as other creative from that advertiser, because the models treat it as essentially the same ad. Meta also confirmed it has been testing creative insights metrics in its Insights API with a small group of partners since June 2026.
What Marketing Brew reported on 28 September 2026
The report, by Jasmine Sheena at Marketing Brew, is based on interviews with agency executives and an emailed statement from Meta spokesperson Alisha Swinteck. It covers three things: what Meta has been telling advertisers about creative diversity, how Meta itself describes that guidance, and a limited test of new creative metrics in the Insights API.
The clearest part of the story is a single example. A paid media director at a social agency told Marketing Brew that two videos of the same person, identical except that one carries on-screen text reading "Text A" and the other "Text B", would be considered the same creative. The executive said this clarification helped the agency structure creative testing for clients.
A second agency executive said Meta has been giving more detailed advice to agency clients on how to diversify: mix video and static assets, and vary messaging and tone by using different hooks, among other tips.
What Meta said, and what it would not say
Meta did not confirm the mechanics the agencies described. Swinteck told Marketing Brew that Meta's guidance is designed to be broad, not tuned to the behaviour of any one model, and called diversifying ad creative a "longstanding best practice" because it delivers measurable results.
In her emailed statement, she said Meta does not get into how its ranking models represent individual creative assets, and that its advice is "deliberately outcome-based (not model-specific), like 'diversify your creative, test, and let the system optimize.'" She added that this is what improves performance, rather than trying to engineer around a single model's behaviour.
That gap matters for reading the story. The "same look, same hook, same ad" rule comes from agency executives describing what Meta has told them. Meta's public position is the general principle, with no detail on how similarity is judged.
Creative insights metrics in the Insights API
Meta is also testing creative insights metrics in its Insights API, the interface that gives advertisers ad performance data. Swinteck said these insights "are not widely available" and are part of regular tests to make the API more useful to partners. According to Swinteck, a "small number of agencies/GCR resellers" have been testing the metrics since June 2026. Marketing Brew did not list the metrics, and Meta gave no date for wider release.
The model changes behind the guidance
Marketing Brew linked the guidance to Meta's continuing investment in its ads models. On Meta's second-quarter earnings call in July 2026, CFO Susan Li said Meta changed its models for understanding user behaviour so they could better analyse organic activity and match people with suitable ads. Li said those changes, together with Meta's recommendation model Gem, increased ad clicks by 8.3%.
The report does not say the creative guidance was caused by those model changes. It places them side by side as part of the same investment.
The facts at a glance
The table separates what came from Meta from what came from agency executives, because the two carry different weight.
| Claim | Who said it |
|---|---|
| Ads with the same look or same opening hook as other creative from the advertiser will not rank as highly; the models see them as the same ad | Agency executives, describing Meta's guidance |
| Two videos of the same person that differ only in on-screen text count as the same creative | A paid media director at a social agency |
| Mix video and static; vary messaging and tone with different hooks | An agency executive, describing Meta's advice |
| Diversifying creative is a "longstanding best practice"; guidance is outcome-based, not model-specific | Alisha Swinteck, Meta spokesperson |
| Creative insights metrics tested in the Insights API with a small number of agencies/GCR resellers since June 2026 | Alisha Swinteck, Meta spokesperson |
What Meta did not say
The report is short, and several things a reader might assume are not in it:
- No similarity threshold. Nobody quoted in the story said how different two ads must be to count as separate, beyond the on-screen text example.
- No penalty language from Meta. The agency executives said near-duplicates would not rank as highly. Meta did not say duplicates are suppressed, removed, or charged more.
- No number of creatives. There is no recommended count of ads per ad set or campaign.
- No launch date for the guidance. The report says Meta "has provided some updated best practices" without dating them.
- No geography. Nothing in the report says the advice or the Insights API test is limited to, or excludes, any market, including Thailand.
Relevant Audience analysis: how to build creative that counts as different
The rest of this article is Relevant Audience's practical reading of the report. It is analysis, not reporting from the source, and it stays within what the source supports: if the models group near-identical ads, testing only pays off when the variants are genuinely different.
Concept first, variations second
A useful way to plan is to separate concepts from variations. A concept is a distinct idea about why someone should buy: a different problem, a different benefit, a different proof. A variation is the same idea with a changed headline, colour or overlay. Based on the example in the report, a set of five variations of one concept may behave like one ad, not five. The planning question then becomes "how many distinct concepts do we have?" rather than "how many assets do we have?"
A simple test matrix
One way to build a matrix is to change more than one visible element between concepts, using the dimensions the agency executives named:
- Format: video against static image, which Meta reportedly advised mixing.
- Opening hook: what the viewer sees and hears first. A question, a product close-up, a person speaking to camera and a before-and-after shot are different openings. The same opening with new text on top, by the report's example, is not.
- Message: price, quality, convenience, social proof or a specific use case are different angles.
- Tone: serious, humorous, instructional or testimonial.
- Look: a different person, setting or visual style, since the report says the same look is read as the same ad.
Three to five concepts that each differ on several of these lines is a more meaningful test than twenty assets that differ only in copy. The report gives no ideal number, so the count here is a planning choice, not a Meta rule.
What counts as a distinct hook
The report uses "opening hook" without defining it. A practical reading is the part of the ad that decides whether someone keeps watching or scrolls past. If two ads open on the same shot of the same person, changing the caption probably does not create a new hook. Opening on the product in use instead of the presenter, or on a customer instead of the brand, is a different start. This is an interpretation of the report's example, not a Meta definition.
Reading the results
If the system treats near-identical ads as one, then results should be read at the concept level. Group ads by concept in your reporting and compare cost per result across concepts, not across small copy tweaks. If delivery concentrates on one ad inside a group of look-alikes, that is consistent with the report's description and not necessarily a sign that the others were worse. Give each concept enough budget and time to produce stable results before deciding, and keep the call to action and landing page constant during the test so the creative is the only thing changing.
Why this may matter more in Thailand
If a campaign runs the same video in Thai and English, or swaps only the on-screen promotion for each sales event, the report's logic suggests those edits may not register as new creative. Language versions for different audiences are still needed, but they should not be counted as creative diversity for testing purposes. Planning a fresh concept for each major sale period, rather than only a new price overlay, is the more direct response to what the agencies described. Thai advertisers running Facebook and Instagram ads can apply the same concept-level grouping to existing campaigns before producing anything new.
What this means for Thai marketers
Nothing in the report requires an advertiser to change settings or meet a deadline. The practical change is in the creative brief. Ask for distinct concepts, check that each opens differently, and report by concept. Because the Insights API creative metrics are in a limited test, most advertisers will keep judging creative from the metrics they already have in Ads Manager for now. For brands that run creator content or user-generated video across platforms, the same discipline carries over to social ads on other platforms.
Keep Meta's own framing in view. The company told Marketing Brew its advice is outcome-based and that advertisers should test and let the system optimise. The agency descriptions give a sharper picture of what "diversify" means in practice, but they are second-hand. Treat them as a hypothesis to test in your own account.
FAQ
Does Meta penalise ads that look too similar?
Not in those words; agency executives told Marketing Brew that ads with the same look or opening hook as the advertiser's other creative won't rank as highly because the models treat them as the same ad. Meta itself only said diversifying creative is a longstanding best practice and declined to explain how its models represent creative.
Do different headlines or on-screen text count as different creative?
According to one agency executive quoted by Marketing Brew, no: two videos of the same person that differ only in on-screen text are considered the same creative. Meta did not comment on that example.
What are the new creative insights metrics in the Insights API?
Meta has not said what they are. A spokesperson confirmed that a small number of agencies and GCR resellers have tested creative insights metrics in the Insights API since June 2026, and said they are not widely available.
Is this guidance live for advertisers in Thailand?
The source does not say. Marketing Brew did not name any market for the guidance or the API test, and Meta gave no rollout details.
How many creatives should I run?
Meta gave no number. The report points to variety between ads, not volume, so the useful measure is how many genuinely distinct concepts are running.
The short version: if two ads look and open the same, plan as though Meta will treat them as one ad. Relevant Audience can help audit an existing Meta account by concept and plan a test that gives each idea a fair run.
