AI Ad Creatives
4 min read
Your best ad is still your best ad, and it's dying. CTR has slid from its peak, frequency is climbing, CPA is drifting up. By the time it shows up as a bad week in the dashboard, you're already scrambling for a replacement you haven't made yet.
The free Claude skill creative-fatigue-scanner catches the decline early and, more usefully, turns the autopsy into the brief for the next batch. The point isn't spotting the death. It's having the replacement ready before it happens.
This post covers the fatigue triggers, why the brief names what to keep, and how to run your first scan.
Key takeaways
Each ad is compared to its own peak week, not the account average, so a strong ad in decline still gets flagged.
Any 1 trigger means early fatigue, 2 or more means fatigued. Nothing is judged off a single bad day.
Every fatigued winner gets an iteration brief that names what made it win and what to keep identical in the next batch.
5 variant directions, each changing exactly 1 variable, so you don't accidentally replace the part that was working.
The 4 fatigue triggers
Trigger | Threshold |
|---|---|
CTR decline from the ad's peak week | Down 20% or more |
Frequency, prospecting | Above 3.0 |
Frequency, retargeting | Above 5.0 |
CPA rising while CTR falls | 2 consecutive weeks |
The comparison against the ad's own peak is what makes this work. Measured against an account average, a great ad in decline can still look above average and sail through unflagged. Measured against its own best week, the slide is obvious while there's still time to act on it.
Everything is week over week. A single soft day is noise, and the skill won't call fatigue on it.
Healthy, early, fatigued
Every ad lands in 1 of 3 grades, and the middle one is the reason to run this weekly. Healthy means no triggers hit. Early fatigue means exactly 1 trigger, which is a watch signal, not a pause signal: it tells you to start briefing now while the ad is still earning. Fatigued means 2 or more triggers, and that ad needs a replacement in the queue already.
Most teams only ever see the third grade, because it's the only one visible without a scan. Catching the second is what buys you the week of lead time that makes a calm creative refresh possible instead of a panicked one.
The iteration brief is the actual deliverable
Flagging a tired ad is easy. Knowing what to make next is the hard part, and it's where most creative refreshes go wrong: the team changes everything, the new batch underperforms, and nobody can say which change caused it.
So for every fatigued ad that was once a winner, the skill writes a brief covering what it won on, its angle, hook anatomy, format, and first 3 seconds, then what to keep identical, then 5 variant directions that each change 1 variable: new hook same angle, new format same hook, new opening visual, new social proof, new CTA.
That keep list is the part worth reading twice. Fatigue refreshes fail when a team accidentally rebuilds the winning variable out of the ad.
Produce the variants while the winner is still running
A brief is only useful if the batch gets made. Vibelets turns the direction into on-brand product ads in minutes, so the replacement is live before you rotate the winner out.
Rotate only once the replacement is live
The skill never recommends pausing a fatigued ad into an empty slot. The rotation rule is that the replacement goes live first, then the tired ad comes out.
There's a cost reason for the ordering too. Adding creative to a live ad set is a significant edit, which resets the learning phase. If you're going to pay for a relearn, pay for it once with the whole new batch, not 5 times across a fortnight of one at a time swaps.
How to set it up in Claude
Download the skill file from the skill's Google Drive folder, free, no email gate, then add it to Claude by uploading it as a skill under Settings or attaching it to a project or chat. Anthropic's Help Center guide to skills covers the upload path.
Export per ad weekly data for the last 4 weeks with CTR, frequency, and CPA or ROAS. 4 weeks is the minimum for peak comparison to mean anything. Run it weekly on any account spending daily.
Conclusion
Refresh the winner before it dies, not after. The scan tells you which ads are sliding, the brief tells you what to keep, and the replacement goes live before you rotate.
Catch fatigue while there's still time
Free Claude skill, yours in 1 click.
Frequently asked questions
What counts as creative fatigue on Meta?
Any of 4 triggers: CTR down 20% or more from that ad's peak week, frequency above 3.0 in prospecting or 5.0 in retargeting, or CPA rising 2 consecutive weeks while CTR falls. 1 trigger is early fatigue, 2 or more is fatigued.
Why compare an ad to its own peak instead of the account average?
Because a strong ad in decline can still sit above the account average and pass unflagged. Measured against its own best week, the slide shows up while there is still time to brief a replacement.
How often should I run the fatigue scan?
Weekly on any account spending daily. The triggers are week over week, and the scan needs 4 weeks of per ad data for the peak comparison to mean anything.
Should I pause a fatigued ad straight away?
No. The rotation rule is replacement live first, then rotate the tired ad out. Pausing into an empty slot just removes delivery without replacing it.
Will adding new creative reset my learning phase?
Yes, adding or swapping creative on a live ad set counts as a significant edit. That is why the skill briefs a whole batch rather than a single replacement: if you are paying for a relearn, pay for it once.
Sources
Make ads that sell
your Products.
Generate on-brand product ads in minutes, no designer or brief required.
Trusted by 400+ growing brands