Ad Fatigue Detector

"My ads are fatigued" is the most over-diagnosed condition in paid social. Four very different problems produce the same symptom — results got worse — and each needs an opposite response:

Getting this wrong is expensive in both directions: refreshing creative that was fine wastes a production cycle and resets learning, while blaming the auction for a genuinely dead ad burns budget for weeks.

Getting the data

You need time-series data, not totals. A single lifetime row cannot show a trend, and the whole diagnosis is about trend.

1. Connected account (Meta / TikTok / Google connector or MCP) — pull daily or weekly rows per ad for at least the last 30 days, ideally 60. Request: date, ad name, spend, impressions, clicks, conversions, and frequency where the platform provides it (Meta and TikTok do; Google Search does not).
2. File or paste — an export broken down by day or week. In Meta Ads Manager the breakdown is Reports → Export table data with Day as a breakdown; a lifetime-only export can't be used for this and you should ask for the day breakdown rather than guessing.
3. Critically: get the account or campaign totals too, not just the one ad in question. Without a baseline you cannot tell fatigue from a market-wide shift, and that distinction is the entire value of this skill.
4. Establish the target. Ask for (or confirm) the target CPA or ROAS before diagnosing. Step 2's "never worked" verdict is meaningless without it — there is no principled line between "never worked" and "worked, then decayed". If they don't know, build one from unit economics and say you've done so.

The diagnostic sequence

Run these in order and stop at the first that fits.

1. Is there enough data? (the impatience check)

An ad needs meaningful volume before any trend is real. Reasonable floors: at least 2 weeks running, 50+ conversions total or 1,000+ clicks, and enough spend to have exited the platform's learning phase (Meta's rule of thumb is ~50 optimization events per week per ad set).

Below those, the honest answer is "there is no signal here yet" — say that plainly. Also flag the specific impatience pattern of judging an ad inside its learning phase, when performance is expected to be unstable and usually improves.

2. Did it ever work? (the dud check)

Compare the ad's first stable week (after learning) against its target. If it never hit target CPA/ROAS, there was never a peak to decay from. This is a creative that didn't land — kill it and move on. Calling it "fatigue" implies a refresh of the same angle will help, and it won't.

3. Did everything decline together? (the baseline check — the one people skip)

Compute the change for the ad and for the account (or at least the campaign) over the same window. Compare:

• Ad CPA up 40%, account CPA up 38% → not fatigue. The auction got more expensive, or a tracking break, seasonality, or a landing page change hit everything. Refreshing creative will not fix it, and you'll conclude the new creative "also fatigued".
• Ad CPA up 40%, account CPA flat → this ad specifically decayed. Continue to step 4.

Also sanity-check CPM separately: if CPM rose sharply while CTR held, that's auction pressure, not creative wear.

4. Does it match the fatigue signature?

Genuine creative fatigue has a recognizable shape. Look for these together, not individually:

• Frequency climbing past roughly 3–4 on a 7-day lookback for prospecting (retargeting tolerates much higher — 6–8 over 7 days is normal, so don't flag retargeting on frequency alone). Read this from the trailing-window pull, never from day rows: daily frequency is almost always near 1.0 and structurally cannot cross the threshold.
• CTR decaying meaningfully — 25%+ measured as the ad's best 7-day window against its most recent 7 days — while the audience stays the same size. Use 7-day windows rather than single days so one bad day can't masquerade as a peak.
• CPA/CPM rising as a consequence, following the CTR decline rather than preceding it.
• Reach saturating — the ad is cycling the same people rather than finding new ones. This needs reach and audience size, which are not in a standard export: pull reach from the platform and ask for the audience definition. If you have neither, say you're judging on frequency and CTR alone.

CTR falling before frequency rises usually isn't fatigue; suspect an audience or placement change instead.

5. Quantify how far along it is

If it is fatigue, say where in the curve it sits and roughly how long is left at current spend, based on the rate of decline. "CTR is down 31% over three weeks and still falling — you have maybe 10 days before it's under target" is far more useful than "fatigued: yes".

What to produce

Lead with the verdict in one sentence, because that's the decision:

FATIGUE DIAGNOSIS — [ad name], [date range]

VERDICT: [Genuine fatigue | Never worked | Account-wide shift | Too early to tell]

Evidence:
  CTR       [first period] → [recent period]  ([%] change)
  Frequency [first] → [recent]
  CPA       [first] → [recent]  ([%] change)
  Account CPA over the same window: [%] change   ← the control
  Volume: [conversions] conv / [clicks] clicks over [days] days

WHAT TO DO: [one specific action]

Then two or three lines of reasoning, and only if the verdict is genuine fatigue, what to change: which element to vary first (hook/opening 3 seconds usually carries most of the effect on video, primary image on static), and what to hold constant so the test stays readable.

Judgment worth applying

Frequency alone proves nothing. It's the most cited and least reliable fatigue signal. High frequency with stable CTR means the ad is still working on a small audience — that's fine. Only frequency plus CTR decay means wear.

Broad audiences fatigue slower than you think; small ones burn out fast. Always report frequency against audience size, not as an absolute.

Seasonality is often mistaken for fatigue — post-holiday drops in January, summer B2B slumps. If last year's data is available, check whether the same dip happened then.

Refreshing the whole creative is usually overkill. Fatigue is often specific to the hook. Recommend the smallest change likely to work before a full production cycle.

A refresh resets learning. New creative re-enters the learning phase with unstable performance for days. Say so — otherwise the person panics at day 3 of the replacement and the cycle repeats.

Never pause or edit live ads directly, even with write access. Present the diagnosis and the recommended action, and let the person execute or explicitly approve it.