Own Creative Diagnosis

Diagnoses the user's own ad creative performance: which ads perform strongly and what patterns are worth testing in variations. Adapted from the Meta Ads Creative Analysis framework — the numeric half (Hook/Hold/CTR) works from any complete data source; the qualitative half requires creative content from the MCP creative endpoint or files/descriptions supplied by the user because CSV exports do not contain it.

Read-and-recommend only. Produce analysis and recommendations for the user to review and apply manually; do not execute changes in the ad account.

When to use this skill

• "Why isn't this ad working?"
• "What's working across my top ads?"
• "Help me plan variations of [winning ad]"
• As the "own creative" input into winning-pattern-synthesis, alongside competitor-ad-intelligence

Step 0: Data Inventory

Identify the source before anything else:

State what's available and missing in one short line before proceeding. If GoMarble MCP is connected, default to it — never ask the user for a CSV export first, as it exists already.

Default account: unless the user specifies otherwise, use their default Meta ad account. In all output, reference campaigns/ad sets/creatives by name only — never surface account IDs or account names.

Set the account's Primary Conversion Metric (PCM) before comparing performance: use ROAS for ecommerce and cost per lead/result for lead generation or custom conversion accounts. Treat ROAS as weak when it is below target and cost per result as weak when it is above target; never compare mixed conversion types in one baseline.

Record the PCM target and an explicit, current analysis window before applying an action rule. Use 30 days by default when the user has not requested another period. If either the target or dated window is missing, diagnose the creative but do not recommend pausing it.

Step 1: Establish the high-spend review set

Sort ads by spend descending and include rows until cumulative spend first reaches or exceeds 90% of total, including the row that crosses the threshold. This is the high-spend review set, not a top-performer set; performance is assessed in the following steps.

Step 2: Format-agnostic first pass

Apply to every ad in the high-spend review set:

Pause evidence gate: never recommend pausing an ad younger than 7 days or the highest-converting ad in its ad set. For an ad with conversions, require at least 3 ad-level conversions in the analysis window before using weak PCM as pause evidence. For an ad with zero conversions, require spend above 2× the target cost per result (or 2× a known break-even acquisition cost for ROAS accounts). If the applicable threshold cannot be computed, classify the ad as WATCH.

Step 3: Video-specific diagnostic scenarios

For video ads, classify using Hook Rate / Hold Rate / CTR:

• Hook Rate (%) = 3-second views ÷ video plays × 100. Good ≥40%, average ≥26% and <40%, poor <26%.
• Hold Rate (%) = ThruPlays ÷ 3-second views × 100, benchmarked against a 90-day account average. Poor is below the baseline, average is from the baseline to <1.25× baseline, and good is ≥1.25× baseline (a relative 25% lift, not 25 percentage points).
• CTR (%): good ≥1.25%, average ≥0.65% and <1.25%, poor <0.65%.

Compare unrounded values and round only for display. If a denominator is zero or missing, mark the metric unavailable instead of classifying it.

These scenarios localize where performance may weaken; they do not prove a root cause. Treat the diagnosis as a hypothesis and check audience, placement, offer, delivery, and landing-page differences before presenting a causal explanation.

Step 4: Format-specific rules

• Single image: same high-CPM/low-CTR pause logic as Step 2; good metrics → variation candidate
• Single video: apply Step 3 scenarios
• Catalog/Advantage+ ads: Meta doesn't provide per-product conversion data — use spend+CTR to build a high-spend product review set. High CPM+weak PCM → test a different product set or remove products pushing CPM up. Low CTR+weak PCM → identify and remove low-CTR products.

Step 5: Pattern recognition across strong performers

Across the high-spend set's good-performing ads, extract and log the actual creative content. Use facebook_analyze_ad_creative_by_id_or_url when MCP is connected; otherwise ask the user for video/image files or descriptions because CSV exports do not include the creative:

• Common angle (which selling points repeat)
• Common format (video vs image vs carousel)
• Common hook mechanism
• Common copy structure/tone/length

This output is the direct input to winning-pattern-synthesis.

Guardrails

• Never recommend pausing the ad with the highest conversion volume in its ad set, regardless of budget share — that's the algorithm working correctly.
• Before judging an ad set's efficiency trend, require 3–5 conversions/week; the zero-conversion pause exception instead uses the explicit spend gate in Step 2.
• Apply the ad-level pause evidence gate in Step 2; the ad-set minimum does not substitute for ad-level evidence.
• Budget/scale recommendations are out of scope for this skill — it diagnoses creative, not delivery mechanics.
• Never invent Hook/Hold/CTR numbers if the data isn't available — state what's missing and ask for it, or proceed on qualitative creative review only and say so.
• Do not execute pause, edit, or launch actions; recommendations are for manual user review and application.

Output Format

### Creative Diagnosis: [Account/Campaign scope]

High-spend review set: [N ads, cumulative spend first reaches/exceeds 90%]

Per-ad diagnosis:
| Ad | Hook | Hold | CTR | PCM | Scenario | Verdict |
|---|---|---|---|---|---|---|

Performance patterns (shared across strong performers):
- Angle: ...
- Hook: ...
- Format: ...
- Copy: ...

Recommendations:
1. [ad-specific fix, with metric justification]
2. ...