Ad Creative Analysis

Analyze a directory of competitor or reference ad creatives. Produce a per-creative JSON analysis and a cross-creative pattern summary.

Step 1 — Accept Inputs

Expect one of:

• - A directory path containing image files (.jpg, .jpeg, .png, .webp, .gif) and/or video files (.mp4, .mov, .avi, .webm)
• An optional metadata.json file in that directory with fields per filename: platform, spend, duration_days, impressions, INLINECODE14

If no path is given, ask the user: "Please provide the directory path containing the ad creatives."

List all files in the directory. Separate into image ads and video ads. Log the count of each before proceeding.

Step 2 — Analyze Image Ads

For each image file, use vision/image analysis to evaluate the following.

Design Evaluation

Assess these five dimensions:

1. 1. Visual hierarchy — Is the eye drawn to the right element first? Is there a clear focal point?
2. Color usage — Does the palette create contrast, evoke emotion, and maintain brand coherence?
3. Text overlay readability — Is copy legible at a glance? Font size, contrast, placement?
4. CTA prominence — Is the call-to-action visually distinct, clearly placed, and easy to act on?
5. Brand consistency — Logo placement, color adherence, font alignment with brand identity.

Image Scores (1-10 each)

• - attention_grab — How fast and strongly does the creative stop a scroll?
• INLINECODE16 — How clearly is the core message communicated without needing context?
• INLINECODE17 — How compelling and action-oriented is the CTA?

Image Extraction

Extract:

• - primary_message — The single core thing this ad is communicating (one sentence)
• INLINECODE19 — One of: fear, aspiration, social_proof, urgency, curiosity, humor, trust, belonging, exclusivity
• INLINECODE20 — Inferred from visuals, copy, and context (e.g., "women 25-35 interested in fitness")
• INLINECODE21 — The first piece of copy the eye lands on (headline or main text)

Step 3 — Analyze Video Ads

For each video file, analyze the video directly using vision. If a transcript file exists alongside the video (same filename, .txt or .srt extension), read and use it.

Video Evaluation

Assess these four dimensions:

1. 1. Hook quality (first 3 seconds) — Does it immediately create curiosity, shock, or recognition? Would someone stop scrolling?
2. Script structure — Does it follow a logical persuasion arc (problem, solution, proof, CTA)?
3. Pacing — Is the editing rhythm appropriate for platform and audience? Not too slow or rushed?
4. CTA placement — Is the call-to-action clear, timed well, and repeated if needed?

Video Scale Score (1-10)

Assign a single scale_score representing the ad's viral and engagement potential at scale:

• - 9-10: Exceptional hook, tight script, clear CTA. Likely to perform well at high spend.
• 7-8: Strong fundamentals, minor weaknesses. Good candidate for testing.
• 5-6: Average execution. Needs a stronger hook or clearer CTA before scaling.
• 3-4: Core idea present but poor execution. Requires significant rework.
• 1-2: Unlikely to perform. Fundamental issues with hook, message, or CTA.

See references/analysis-framework.md for detailed scale score rubric.

Video Extraction

Extract:

• - hook_text — Exact words spoken or shown in the first 3 seconds
• INLINECODE27 — One of: question, boldclaim, painpoint, curiositygap, socialproof, beforeafter, demonstration
• INLINECODE28 — The core value proposition stated in the ad
• INLINECODE29 — One of: fear, aspiration, socialproof, urgency, curiosity, humor, trust, belonging, exclusivity
• INLINECODE30 — The exact CTA spoken or shown
• INLINECODE31 — When the CTA appears (e.g., "end", "middle", "repeated throughout")

Step 4 — Universal Metadata (All Ad Types)

For every creative, regardless of type, record:

• - filename — The file name
• INLINECODE33 — One of: single_image, carousel, video, story, reel
• INLINECODE34 — Detected or inferred (e.g., 1:1, 9:16, 16:9, 4:5)
• INLINECODE39 — Width x height in pixels if detectable
• INLINECODE40 — Inferred from content and CTA: awareness, consideration, or INLINECODE43
• INLINECODE44 — Which platforms this format and ratio suits best (e.g., ["Instagram Feed", "Facebook Feed"])

Step 5 — Output Per-Creative JSON

Output one JSON object per creative. Print all results together in a single JSON array.

Image ad example structure

CODEBLOCK0

Video ad example structure

CODEBLOCK1

Step 6 — Generate Cross-Creative Summary

After analyzing all creatives, produce a summary object appended to the output. Include:

• - total_analyzed — Count of creatives analyzed (split by type)
• INLINECODE48 — Filenames of the top 3 creatives by score (images by average score, videos by scale score)
• INLINECODE49 — Most frequently detected emotion appeal across all ads
• INLINECODE50 — List of recurring hook patterns or phrases observed
• INLINECODE51 — Most common CTA structures seen (e.g., "verb + free + urgency")
• INLINECODE52 — Most common inferred ad objective
• INLINECODE53 — Count per ad format
• INLINECODE54 — 3-5 actionable observations for improving or scaling these creatives

Summary example structure

CODEBLOCK2

Step 7 — Handle Missing or Unreadable Files

If a file cannot be analyzed (corrupted, unsupported format, too dark/blurry for vision):

• - Include the filename in the output with "status": "unreadable" and a brief "reason" field
• Continue analyzing remaining files, do not stop

Reference Material

Consult skills/ad-creative-analysis/references/analysis-framework.md for:

• - Detailed scoring rubrics per metric
• Ad psychology pattern definitions
• Hook formula templates
• Extended example output