How to Identify and Expand High-Performing Audiences on Meta
On Meta, audiences aren't targeted anymore. They're found through creative. How to tell which audiences are working, when they're saturated, and how to expand them.
For years, finding a new audience on Meta meant launching a new ad set. Lookalikes, interest stacks, a fresh custom audience. If you wanted to reach a different pocket of people, you changed the targeting.
That's not how it works anymore. And the teams still doing it are paying for audience expansion they could get for free.
Audiences on Meta aren't targeted now. They're found.
Meta said it plainly in March 2025: with AI-enabled ad tools, the focus has shifted from niche targeting to creative diversification as the best lever for finding relevant audiences. Creative diversification sits alongside broad targeting and the Conversions API as a core pillar of Meta's own Performance 5 framework.
The spend is following. Advantage+ Shopping Campaigns went from 26% of Meta retail and ecommerce spend during Cyber Five 2024 to 33% in 2025. Those campaigns automate targeting, placement, and budget. The advertiser controls the creative and almost nothing else.
Here's what that means mechanically. Since the Andromeda update, Meta's retrieval system decides who sees an ad based on predicted resonance. It matches each creative to the people most likely to respond to that specific creative. A raw founder-to-camera video reaches a different cluster than a polished product shot. A price-comparison hook reaches a different cluster than a testimonial.
So every genuinely distinct concept you run is, in effect, its own audience. The ad set is no longer where the audience lives. The ad is.
How to identify a high-performing audience
If the ad is the audience, then reading which ads work is reading which audiences work. Three steps.
Break performance down by persona and angle, not by ad set. Ad-set-level reporting tells you a blended number. What you want to know is which persona responded, which pain point pulled, and which format carried. Tag every concept with those three things before it launches, then read results against the tags. The persona-and-angle combinations that keep winning are your high-performing audiences, whether or not they share an ad set.
Check that the audience is actually still there. A winning concept can stop working for two different reasons, and they need opposite fixes. Meta's own research draws the line between creative fatigue, where the same people are tired of the same ad, and audience saturation, where you've reached everyone that creative can reach. Fresh creative fixes the first. It does nothing for the second.
The metric that separates them is one almost nobody watches: first-time impression ratio, the share of today's impressions reaching someone new. When prospecting frequency passes roughly 2.5 and first-time impression ratio drops below about 50%, delivery has shifted from finding new people to re-serving the saturated core. That's saturation. If frequency is fine and new reach is healthy but CTR is falling, that's fatigue.
Confirm it with the cheap test. Run a fresh concept into the same audience. If performance recovers, it was fatigue and the audience is still good. If it only recovers when you reach new people, the audience is done and it's time to expand.
Is this audience still performing?
Enter the last 7 days for one prospecting ad set. The verdict updates as you move the sliders.
Average times each person saw the ad
1.8
Share of impressions reaching someone new
65%
Use a 3-day average, not a single day
-5%
Ignore if there's a known auction event
+3%
Thresholds are practitioner benchmarks. The fatigue vs. saturation distinction is from Meta's own creative fatigue research. Confirm any verdict with the test described above.
How to expand a high-performing audience
Once you know an audience works, the instinct is to duplicate the ad set and raise the budget. That's the old playbook, and under Andromeda it mostly just raises frequency on the same people.
Expansion now means giving the retrieval system a new cluster to find. Three ways to do it, in order of how often they work.
Adjacent persona, same pain point. If the "skeptical first-timer" concept is winning, build the same argument for the "price-comparer" or the "gift buyer." Same product truth, different person. Andromeda reads the difference and retrieves against a new set of users.
Same persona, new format. Reels were 28% of Instagram ad impressions during Cyber Five 2025, up from 20%, equal to Feed. If the winning concept is a static, a video version of the same idea reaches people the static never did. Format is a retrieval signal now, not just a placement.
New angle, proven audience. The persona is responding, so try a different reason for them to act. A testimonial where you ran a demo. A problem-first hook where you ran a benefit-first one.
What doesn't count as expansion: more exports of the winning ad. Meta groups ads that look and read alike and treats them as one candidate, so cosmetic variation doesn't create a new cluster. It just spreads the same audience across more file names.
The floor for this to work is real. Practitioner guidance for Advantage+ is 8 to 15 genuinely different creatives before launch and 10 or more active. Below that, the system doesn't have enough distinct signals to find distinct audiences.
What to stop doing
Three habits from the targeting era that now cost money.
Stacking lookalikes to reach new people. Advantage+ already reaches beyond your targeting when it finds a better opportunity. Manual audience layers mostly constrain it.
Duplicating a winning ad set at a higher budget. You get auction overlap with yourself and a learning reset on the copy.
Judging individual ad CPA. Andromeda sends disproportionate spend to whichever ad is winning, which makes that ad's CPA rise fastest and a low-volume ad look artificially cheap. Read the persona and angle. Not the file.
How Campfire fits
Campfire briefs are built on persona, pain point, and format from the start, so every concept launches already tagged. Relay pulls live Meta performance and matches it back to the ads. Campfire Intel breaks results down by persona and pain point, which is the same thing as breaking them down by audience. Build from Report turns the winners into the next round of adjacent concepts.
The short version
You don't target a high-performing audience on Meta anymore. You find one by running distinct creative and reading which persona and angle responds. You confirm it's still there by watching first-time impression ratio, not just frequency. And you expand it by giving the retrieval system a new concept to match, not a duplicate ad set.
The audience lives in the ad now. Build accordingly.
FAQ
How do I find new audiences on Meta without lookalikes?
Run genuinely distinct creative. Since Andromeda, Meta matches each concept to the people most likely to respond to it, so a new persona, angle, or format reaches a new cluster without any targeting change. Meta's own guidance moved from niche targeting to creative diversification in March 2025.
How do I know if my audience is saturated or my creative is fatigued?
Watch first-time impression ratio alongside frequency. Frequency above 2.5 with fewer than half of impressions reaching new people is saturation, and new creative won't fix it. Frequency and new reach holding but CTR falling is fatigue, and a fresh concept for the same audience will. Confirm by running new creative into the same audience and seeing whether it recovers.
Should I duplicate a winning ad set to scale it?
No. Duplicating creates auction overlap with yourself and resets learning on the copy. Expand by adding a distinct concept aimed at an adjacent persona or a new format, and step the budget up 15 to 20% at a time on the original. More on that in How to Increase Your Facebook Ad Budget Without Lowering Performance.
