meta-creative-fatigue-detection

Per-ad creative fatigue diagnosis on Meta Ads. Scores every active ad as Healthy / Early Warning / Fatigued / Dead using CTR decline, frequency creep, CPA spike, CVR decline, and spend-shift signals.

SKILL.md

GitHub

Meta Creative Fatigue Detection (Read-Only)

Diagnose creative fatigue across the ad account the user names. Default lookback: 14 days. Read-only — never pauses or modifies ads.

If the user didn't name an account, list their connected accounts and ask which one to use.

Fatigue scoring signals

Score each active ad on:

Signal Healthy Early Warning Fatigued Dead
CTR decline vs first-week-of-life < 10% drop 10–25% drop 25–50% drop > 50% drop
Frequency < 2.5 2.5 – 3.5 3.5 – 5 > 5
CPA spike vs ad-set median < 10% above 10–25% above 25–50% above > 50% above
CVR decline vs first-week < 10% drop 10–25% drop 25–50% drop > 50% drop
Spend shift Stable or rising Flatlined Declining Reaching cliff

Final classification = worst signal score across all five dimensions.

Output

Section 1 — Fatigue summary table

| Ad name | Score | CTR drop | Freq | CPA delta | CVR drop | Days live |

Sort: most fatigued first. Cap at top 20.

Section 2 — Per-ad recommendations

For each Fatigued or Dead ad, one line:

  • 🪦 [Ad name] — Dead. Recommendation: retire. The hook is exhausted; do not relaunch as-is.
  • 🔻 [Ad name] — Fatigued. Recommendation: pause this variant; the underlying concept is still winning, remix the hook.
  • ⚠️ [Ad name] — Early Warning. Watch — likely needs a refresh in 7–14 days.

For each Healthy or Early Warning winning ad with strong CVR:

  • ✨ [Ad name] — Healthy winner. Recommendation: extract the pattern, brief 2–3 variants on the same concept.

Section 3 — Creative refresh brief

3–5 bullet points on what themes/hooks/formats deserve fresh creative based on the patterns you see in the Fatigued + Healthy-winner cohorts.

End with: _All recommendations — no changes applied. Ask explicitly to pause anything._