Your frequency number is hiding which of three problems you have

Frequency is impressions divided by reach. That is Meta's own definition, and it is the thing that turns a scary number into a diagnosable one. A frequency level is not a diagnosis, because the same reading comes from opposite movements: rising frequency with rising reach means you are buying more coverage; rising frequency with flat reach means the market got more expensive; rising frequency with falling reach means your pool is closing. Three problems, three fixes, one identical number in the column.

The industry took that definition and turned it into a rule of thumb — refresh creative at 2.0, panic at 3.0, rotate every 10 to 14 days. The rule acts on the numerator and ignores the denominator. It also misreads the source. Meta's glossary does not say high frequency means fatigue. It says that if performance begins to drop as your frequency numbers rise, your audience may be experiencing fatigue. The conditional is the whole sentence. Almost nobody quotes it.

That matters more this quarter than usual, because the auction is being rebuilt underneath the metric.

Same number, three different accounts

Here is the arithmetic, using the definition and nothing else. Three accounts all read 4.7 this morning:

These are three arithmetic cases, not three advertisers. The point is that the metric cannot tell them apart, and the wrong response costs a quarter. Account A should probably keep going. Account B needs its exclusions removed. Account C needs a bid decision. Send all three the same new creative set and only one of them improves.

Read the two inputs, then the result

Before touching anything, pull four numbers per ad set for the same date range: spend, impressions, unique reach, and results. Then compute two derived figures.

• Cost per 1,000 people reached = spend ÷ reach × 1,000. This is what a new person actually costs you. Meta shows it directly in reach campaigns; in conversion campaigns you calculate it.
• Results per 1,000 people reached = results ÷ reach × 1,000. This tells you whether new people are behaving differently from the people you already saturated.

Those two numbers, tracked week over week, separate growth from repetition far more cleanly than frequency does. Frequency tells you the ratio. These tell you the price of moving the denominator.

The control test nobody runs

This is the cheapest diagnostic in paid social and almost nobody does it: pick three campaigns or ad sets you have not edited in at least 14 days.

• If their frequency and CPM moved in the same direction as the account you are worried about, the cause is outside your account. It is the auction or the season, and creative work will not touch it.
• If only the campaigns you edited moved, the cause is inside your account. It is your budget, your audience, or your creative.

An untouched cohort is your only control group in a live account, and it costs one filter. Run it before you commission new creative, not after.

Here is the full mapping once you have both reads:

Three of the four rows are not creative problems, and the industry default answer — new ads — only helps in the last one.

You cannot cap frequency on a conversion campaign

The most common advice in every November planning thread is "cap the frequency." Check what buying type you are in before you spend an afternoon looking for the setting.

Meta's glossary is explicit: a frequency cap is a feature of reservation buying, where it defaults to two impressions every seven days and can be adjusted. Auction buying — the buying type behind conversion campaigns, app campaigns, and Advantage+ — has no frequency cap. There is no field to find. The cap you are hunting for exists in a different product.

What auction buying gives you instead:

• Budget, which controls impressions directly.
• Audience definition and exclusions, which control reach.
• Placement, which changes which inventory pool you draw from.
• Schedule, which changes when those people are reachable.
• The offer, which changes whether the same impression is worth anything.

Four of those five move the denominator. That is the real reason to widen an audience before you re-cut the ad.

Why the auction got denser: Meta's own numbers

Meta published the mechanics twice in the last three weeks, and both posts are worth reading before you blame your creative.

On 28 October, Meta's engineering team described the current Andromeda retrieval system: it surfaces top candidates at over one billion ads per second, roughly a 100x throughput lift over the 2024 system, and scores every eligible ad with near ranking-level model quality. The post also says why the candidate pool grew — generative AI creative tools helped advertisers rapidly create hundreds of millions of new ads, pushing it to "an unprecedented scale." Andromeda alone has delivered a 2.4% conversion increase on Facebook since 2025.[1]

On 13 October, the GEM post reported 18.9% conversion improvement and 12% click-through improvement across Meta's surfaces and objectives, with conversion uplift of 5.8% on Facebook, 4.1% on Instagram and 1.6% on off-site conversions for targeted users over the past year.[2]

Put those together and you get the exact pattern that gets mislabelled as fatigue. A much larger pool makes the marginal impression more contested, which pushes CPM up. A better matching model means the impressions you do win perform well, so click-through and conversion rate hold flat. Result: frequency climbing, cost climbing, performance unchanged. Every input points at the auction and away from your creative.

Here is the part the vendor posts will not say. Meta publishes these numbers to explain why its inventory is worth more, not to explain your account. The company has every reason to describe the system as better. It has no reason to tell you whether the extra cost is worth it for your margin. That part stays with you, and it is a pricing question, not a production question.

The peak-season decision is not a creative decision

From mid-November through early December the third diagnosis is close to guaranteed. Auction density rises, marginal reach gets more expensive, and frequency climbs whether or not you change anything. So the budget decision in front of you is not "what do we shoot next." It is "what is the most we will pay for one more new person."

Two things make that decision survivable:

Pre-commit the number before peak week. Work out your cost per 1,000 new people reached today, and your loaded value per conversion — margin per order times close rate. Divide. That gives you a ceiling. When the cost of a new person crosses it, more budget buys repetition, not growth, and you should say so in the room before the numbers go bad rather than after.

Change the offer, not the edit. Inside peak week, one lever still changes who the system can find: price, bundle, threshold, guarantee. A re-cut of the same ad with a new hook does not change the addressable pool; it changes which of your own ads wins the same impression. Teams that swap creative during the last ten days usually spend the week relearning a model they already taught.

And give any change enough room before you judge it. You cannot read a peak-week result off three days of data — the same sample-size logic that applies to when to kill a Meta ad set applies here, only with more expensive variance.

What to do on Monday

1. Filter for untouched campaigns. Three ad sets, no edits in 14 days. Compare frequency and CPM direction against the ones you are worried about.
2. Decompose the ratio. Spend, impressions, reach, results. If reach is falling while spend is flat, stop looking at creative.
3. Compute cost per 1,000 new people reached. Write the number down with the date. It is your baseline for the ceiling above.
4. Fix the denominator first. Exclusions and narrow geographies are the most common self-inflicted pool shrink we see. Then decide whether the remaining pool is worth the marginal price.

If the answer after all four steps is "the auction is just expensive and the pool is real," that is a legitimate finding. It is a re-pricing conversation, not a production brief — and it is the one most teams never have because they were too busy rotating creative against a number that was never a diagnosis.

For the reporting side of the same problem — why your CPM and a public benchmark can both be true and still not describe your account — that inconsistency has its own write-up. And if you do decide the pool is the problem, creative diversity is the lever that changes which people you can reach, not creative volume.

If you want this run against your own account before the expensive weeks, our paid media team does exactly this audit — the ratio decomposition, the untouched-campaign control, and a written ceiling for peak. If you would rather start with a conversation than a brief, book a call and bring your reach and spend numbers for the last four weeks.