Use case · diagnose
Creative Fatigue Detection — Find Dying Ads Before Budget Burns
Your Meta CTR is dropping, CPA is creeping up, but you don't know which ads are dying. Manually checking 50 creatives every week is the kind of work no human should do.
For: Performance marketers running 10+ Meta creatives in parallel · Time: 5 minutes
Used by paid media teams at
The prompt
Copy this into Claude (with GoMarble's MCP connected). Replace act_XXXXXXXXXX with your Meta ad account ID.
Lead-magnet prompt · free
I want a creative fatigue report for my Meta ad account act_XXXXXXXXXX. Use these signals to classify every ad with >$500 spend in the last 21 days: 1. CTR decline — compare last 7 days vs previous 7 days. >20% drop = fatigue signal. 2. Frequency creep — average frequency in the last 7 days. >4.0 = fatigue signal. 3. CPA increase — last 7 days vs previous 7. >25% increase = fatigue signal. 4. CVR decline — last 7 days vs previous 7. >15% drop = fatigue signal. 5. Spend shift — has Meta pulled spend away from this ad week-over-week? Yes = late-stage fatigue signal. Classify each ad as: - HEALTHY (0-1 signals) - EARLY WARNING (2 signals) - FATIGUED (3 signals) - DEAD (4-5 signals or already paused by Meta optimization) Return a structured table grouped by status. For every FATIGUED and DEAD ad, generate a refresh brief with: - The ad's hook, format, persona - Why it's fatigued (cite the signals) - 3 specific refresh angles to test (new hook, new visual treatment, new persona angle) Then output a paste-ready prompt I can give to Higgsfield (for video regen) or ChatGPT (for image regen) to produce the refresh creatives.
What you get back
Format: Markdown table grouped by status (HEALTHY / WARNING / FATIGUED / DEAD), plus per-ad refresh briefs
Sample output
| Ad ID | Status | Signals | Refresh angles |
|---|---|---|---|
| 120201234... | FATIGUED | CTR -32% · Freq 4.8 · CPA +41% | 3 |
| 120201567... | EARLY WARNING | Freq 4.2 · CPA +28% | 0 |
| 120201890... | HEALTHY | none | 0 |
Next step: Pipe the FATIGUED refresh briefs to Higgsfield MCP for AI video regeneration, or to ChatGPT for image refresh.
GoMarble MCP tools this prompt uses
-
•
facebook_get_details_of_ad_account -
•
facebook_get_adaccount_insights (with breakdowns + action_attribution_windows) -
•
facebook_list_ads -
•
facebook_list_adsets -
•
facebook_get_ad_creative_details -
•
facebook_analyze_ad_creative_by_id_or_url (for refresh brief input)
All tools shown exist in GoMarble's hosted MCP at apps.gomarble.ai/mcp-api/sse. Claude routes the prompt to the right tools automatically.
One-time setup (5 minutes)
- 1.Sign up at apps.gomarble.ai
- 2.Connect your ad accounts in Integrations (Meta, Google, etc.)
- 3.Open
claude.ai/settings/integrationsand add a Custom Connector with URLhttps://apps.gomarble.ai/mcp-api/sse - 4.Open Claude (web or Claude Code), paste the prompt above, replace placeholders, and run
Variations
Cross-channel version
Add Google Ads to the same prompt. GoMarble's hosted MCP runs the same fatigue logic against Google Ads creative if it's a Demand Gen or PMax with creative assets.
Scheduled weekly run
Move from a one-off prompt to a recurring weekly job via GoMarble's Agent Mode. Get the fatigue report in Slack every Monday morning.
Hook-pattern overlay
Cross-reference fatigued creatives with their hook archetype (sensory_demo / transformation / problem_statement / direct_address / concept_reframe). Identify which hook types are fatiguing fastest.
Want this automated?
Run the prompt above every Monday morning automatically. GoMarble's Agent Mode + scheduled jobs deliver the report (Slack / email / in-app) without re-prompting Claude. No more manual checks.
Frequently asked
How does this work without performance data in the API?
Why >$500 spend threshold?
What if I don't have Higgsfield or ChatGPT for creative refresh?
Can I run this on a schedule?
Does this work on TikTok or Google Ads creatives?
Skip the prompt — let GoMarble do this for you.
Sign up, connect your ad accounts, and creative fatigue detection runs on every account, every week, automatically.