# Ad Creative Testing > Systematic creative testing frameworks for ad campaigns to find winning combinations. Use when testing ad creative across platforms. - **Type:** Skill - **Install:** `agentstack add skill-brainbytes-dev-everything-claude-marketing-ad-creative-testing` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [brainbytes-dev](https://agentstack.voostack.com/s/brainbytes-dev) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [brainbytes-dev](https://github.com/brainbytes-dev) - **Source:** https://github.com/brainbytes-dev/everything-claude-marketing/tree/main/skills/paid-ads/ad-creative-testing ## Install ```sh agentstack add skill-brainbytes-dev-everything-claude-marketing-ad-creative-testing ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Ad Creative Testing ## When to Activate Use this skill when: - Launching new ad campaigns that require creative testing - Optimizing existing campaigns with creative refreshes - Building a systematic testing framework for ad creative - Determining winning creative elements (hooks, formats, angles) - Detecting and addressing creative fatigue - Establishing a creative testing library and documentation process ## First Questions Before designing creative tests, clarify: 1. What platform(s) are being tested on? (Meta, Google, LinkedIn, TikTok, YouTube) 2. What is the current creative performance? (Baseline CTR, CPA, ROAS) 3. What creative assets are available? (Images, video, UGC, product shots) 4. What is the testing budget? (Determines number of simultaneous variants) 5. What is the current conversion volume? (Need enough data for statistical significance) 6. How quickly can new creative be produced? (Testing velocity depends on production speed) 7. What has been tested before? (Don't repeat failed experiments without new hypotheses) ## Core Rules 1. **Test concepts before details.** A new messaging angle matters more than a different font color. Start with big swings. 2. **Isolate one variable at a time.** If you change the image AND the headline, you don't know which made the difference. 3. **Let data decide, not opinions.** The creative your team loves and the creative your audience loves are rarely the same. 4. **Statistical significance is non-negotiable.** Small sample sizes lead to false conclusions. Don't declare winners too early. 5. **Winners decay.** Every winning creative eventually fatigues. Plan for continuous testing, not one-and-done. 6. **Document everything.** Without documentation, learnings are lost and tests get repeated. 7. **Testing is an investment, not an expense.** Allocate 15-20% of ad budget specifically for creative testing. 8. **Bad creative tested is better than good creative guessed.** Launch and learn. Overthinking kills momentum. ## Creative Testing Hierarchy Test from top to bottom. Higher-level tests have larger impact. ``` LEVEL 1: CONCEPT (Highest Impact) What: The core messaging angle, value proposition, or audience appeal Example: "Save time" vs "Save money" vs "Look professional" Impact: Can change performance by 2-10x LEVEL 2: FORMAT What: The creative format (video vs image vs carousel vs UGC) Example: Product demo video vs customer testimonial image vs carousel Impact: Can change performance by 50-200% LEVEL 3: HOOK / OPENING What: The first 1-3 seconds (video) or the visual/text hook (image) Example: "Stop wasting money on..." vs "The #1 mistake in..." Impact: Can change performance by 30-100% LEVEL 4: ELEMENT What: Specific design or copy elements within a proven concept Example: CTA button color, headline wording, background color, person in image Impact: Can change performance by 10-30% LEVEL 5: MICRO-OPTIMIZATION What: Small refinements to winning creative Example: Font size, text placement, thumbnail frame selection Impact: Can change performance by 5-15% ``` **Rule:** Never test at Level 5 before you've tested at Level 1. Find the right concept first, then refine. ## Testing Methodology ### A/B Testing (Split Testing) **What:** Two variants (A and B) shown to comparable audiences simultaneously. One variable is changed. **When to use:** - Testing one specific element (headline, image, CTA, hook) - When you need clean, conclusive data - When budget allows sufficient volume per variant **Rules:** - Change ONLY one element between A and B - Split audience evenly and randomly - Run both variants simultaneously (not sequentially) - Same budget per variant - Don't peek at results before reaching statistical significance ### Multivariate Testing **What:** Multiple variables tested simultaneously in various combinations (A1B1, A1B2, A2B1, A2B2). **When to use:** - When you have high traffic/budget volume - When you want to understand interaction effects between elements - When you have a production team that can create many variants **Rules:** - Requires MUCH more data than A/B testing (each combination needs sufficient volume) - Typically only viable on Meta or Google with large budgets ($5,000+/month for testing alone) - Use platform-native multivariate features when available ### Champion/Challenger Testing **What:** The current best-performing creative (champion) runs alongside 1-3 new challengers. If a challenger wins, it becomes the new champion. **When to use:** - Ongoing optimization of established campaigns - When you have a proven winner and want to find something better - For continuous improvement without disrupting performance **Rules:** - Champion gets 60-70% of budget; challengers split the remaining 30-40% - A challenger must outperform the champion by at least 10-15% to become the new champion (accounts for noise) - Replace the weakest challenger every 1-2 weeks with a new test ## Variable Isolation ### What to Test (One at a Time) **Visual Variables:** - Image vs video vs carousel - Product shot vs lifestyle shot vs UGC - Face vs no face - Color scheme - Text overlay vs no text overlay - Background (solid color vs environment vs white) **Copy Variables:** - Headline angle (benefit vs feature vs question vs statistic) - Hook line (first sentence of ad copy) - CTA copy ("Shop Now" vs "Learn More" vs "Get Started") - Copy length (short vs medium vs long) - Tone (formal vs casual vs urgent vs humorous) **Structural Variables:** - Ad format (single image vs carousel vs video) - Video length (15s vs 30s vs 60s) - Number of carousel cards - Landing page destination **Offer Variables:** - Discount percentage vs dollar amount - Free shipping vs discount - Free trial length - Guarantee vs no guarantee - Bonus inclusion vs price reduction ## Sample Size Requirements ### Minimum Data per Variant | Metric Being Tested | Minimum Events per Variant | Typical Duration | |--------------------|---------------------------|-----------------| | CTR | 1,000+ impressions | 2-5 days | | CPC | 100+ clicks | 3-7 days | | Conversion rate | 50+ conversions per variant | 7-14 days | | CPA | 30+ conversions per variant | 7-21 days | | ROAS | 50+ purchases per variant | 7-21 days | ### Statistical Significance - Aim for 95% confidence level before declaring a winner - Use a significance calculator (many are available free online) - If results are close (within 10%), the test may be inconclusive — that's valid data too - Don't stop tests early just because one variant looks ahead. Early leads can reverse. ### When You Don't Have Enough Volume - Extend the test duration (up to 21 days) - Increase budget to get more data faster - Test higher-funnel metrics (CTR instead of CPA) as a proxy - Use sequential testing: test creative A for a week, then creative B for a week (less reliable but better than nothing) - Combine platform data with directional landing page data ## Creative Testing Matrix Template Use this template to plan and document tests: ``` TEST ID: CT-[number] DATE: [Start date] — [End date] PLATFORM: [Meta / Google / LinkedIn / TikTok] CAMPAIGN: [Campaign name] HYPOTHESIS: [If we change X, we expect Y to improve because Z] VARIABLE BEING TESTED: [Specific element being changed] VARIANT A (Control): [Description of current creative] VARIANT B (Challenger): [Description of test creative] TARGET METRIC: [CTR / CPA / ROAS / Conversion Rate] BUDGET PER VARIANT: [$X per variant, $Y total] MINIMUM SAMPLE SIZE: [X conversions / Y clicks per variant] RESULTS: Variant A: [Metric value] (n=[sample size]) Variant B: [Metric value] (n=[sample size]) Winner: [A or B or Inconclusive] Confidence: [XX%] Lift: [+/- XX%] INSIGHT: [What did we learn? How does this inform future creative?] NEXT TEST: [What will we test next based on this result?] ``` ## Iterating on Winners ### The Iteration Process ``` WINNING CREATIVE FOUND | v ANALYZE: Why did it win? - Was it the visual, the hook, the offer, the format? - What audience responded most? - What placement performed best? | v ITERATE: Create 3-5 variations of the winning element - New hooks with the same format - Same hook with different visuals - Same concept targeted to different audiences - Extended or shortened version | v TEST: Run iterations against the original winner | v SCALE: Increase budget on the best iteration | v REPEAT: Continue iterating until performance plateaus ``` ### Iteration Ideas for Winners - **If a video hook won:** Test 5 more hooks with the same video body - **If a specific angle won:** Test different creative executions of that angle - **If UGC style won:** Test multiple UGC creators with the same script - **If a specific offer won:** Test different ways to present the same offer - **If a format won:** Adapt the concept to other formats (carousel version of a winning video) ## Creative Fatigue Detection ### Signs of Creative Fatigue | Signal | Threshold | Action | |--------|-----------|--------| | CTR declining | 20%+ drop from peak | Refresh creative or retire | | CPA increasing | 30%+ increase from best | Test new creative | | Frequency rising | Above 3 (prospecting) or 5 (retargeting) | Rotate or pause | | Engagement dropping | Comments and shares declining | New creative angle | | Relevance score dropping | Below 5/10 (Meta) | New creative | ### Preventing Creative Fatigue - Rotate 3-5 creative variants at all times - Refresh at least one creative element every 2-4 weeks - Have new creative in the pipeline before fatigue hits - Use dynamic creative (platform auto-combines elements) for variety - Expand audiences to show ads to new people (not just new creative to the same people) ### Creative Lifecycle ``` Launch (Days 1-3): Learning phase. Performance is volatile. Ramp (Days 4-10): Performance stabilizes. Best metrics achieved. Peak (Days 10-21): Optimal performance. Scale budget. Decline (Days 21-45): Gradual performance drop. Prepare replacements. Fatigue (Days 45+): Significant performance decline. Retire or radically refresh. ``` ## Performance Benchmarks by Industry ### Meta Ads (Facebook/Instagram) | Industry | Avg CTR | Avg CPC | Avg CPA | |----------|---------|---------|---------| | E-commerce | 1.0-2.0% | $0.50-1.50 | $15-50 | | SaaS/Tech | 0.8-1.5% | $1.00-3.00 | $50-150 | | Education | 0.8-1.5% | $0.80-2.00 | $20-80 | | Finance | 0.5-1.0% | $2.00-5.00 | $50-200 | | Health/Wellness | 1.0-2.0% | $0.50-2.00 | $20-80 | | B2B Services | 0.5-1.0% | $1.50-4.00 | $50-200 | ### Google Search Ads | Industry | Avg CTR | Avg CPC | Avg CPA | |----------|---------|---------|---------| | E-commerce | 2.0-4.0% | $1.00-3.00 | $30-80 | | SaaS/Tech | 2.0-4.0% | $2.00-8.00 | $50-200 | | Legal | 1.5-3.0% | $5.00-15.00 | $80-300 | | Finance | 2.0-4.0% | $3.00-10.00 | $50-200 | | Real Estate | 2.0-4.0% | $1.00-5.00 | $30-100 | **Note:** These are directional benchmarks. Your industry, offer, and targeting will create significant variation. Use your own historical data as the primary benchmark. ## Creative Library Management ### Organization System ``` Creative Library Structure: /Platform /Campaign Type (Prospecting, Retargeting, Brand) /Test Round (CT-001, CT-002, etc.) /Variant A /Variant B /Results Summary ``` ### What to Track Per Creative - Platform and placement - Launch date and retirement date - Performance metrics (CTR, CPA, ROAS) - Total spend - Audience it was tested with - Test result (winner, loser, inconclusive) - Tags (format, angle, hook type, offer type) - Link to source files ### Creative Library Rules - Archive every creative — even losers. Losing creative teaches you what doesn't work. - Tag creative by concept, format, and result for easy searching - Review the library monthly to identify patterns (which concepts keep winning?) - Share insights across the team — creative learnings should not be siloed - Keep source files (PSD, AI, PR, etc.) accessible for quick iterations ## Testing Documentation ### Monthly Testing Report Template ``` CREATIVE TESTING REPORT — [Month/Year] TESTS RUN THIS MONTH: [Number] TOTAL TESTING BUDGET: [$X] TOP FINDING: [One-sentence summary of the most impactful insight] TEST RESULTS SUMMARY: | Test ID | Variable | Winner | Lift | Confidence | |---------|----------|--------|------|------------| | CT-XX | [X] | [A/B] | +XX% | XX% | | CT-XX | [X] | [A/B] | +XX% | XX% | KEY INSIGHTS: 1. [Insight 1 — what worked and why] 2. [Insight 2 — what didn't work and why] 3. [Insight 3 — pattern or trend observed] CREATIVE FATIGUE STATUS: Active winners: [List] Approaching fatigue: [List] Retired this month: [List] NEXT MONTH TESTING PLAN: 1. [Test hypothesis 1] 2. [Test hypothesis 2] 3. [Test hypothesis 3] CREATIVE PRODUCTION NEEDS: [What new creative is needed for next month's tests] ``` ## Quality Gate Before launching a creative test, verify: - [ ] Is there a clear, written hypothesis? (If we change X, Y will improve because Z) - [ ] Is only ONE variable being changed between variants? - [ ] Is the test budget sufficient for statistical significance? - [ ] Is the test duration planned (minimum 5-7 days, ideally 14)? - [ ] Are baseline metrics documented for comparison? - [ ] Is the test documented in the testing matrix template? - [ ] Are other variables controlled? (Same audience, same budget per variant, same schedule) - [ ] Is there a plan for what to do with the winner AND the loser? - [ ] Will results be documented regardless of outcome? - [ ] Is the creative library updated with the new test variants? ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [brainbytes-dev](https://github.com/brainbytes-dev) - **Source:** [brainbytes-dev/everything-claude-marketing](https://github.com/brainbytes-dev/everything-claude-marketing) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-brainbytes-dev-everything-claude-marketing-ad-creative-testing - Seller: https://agentstack.voostack.com/s/brainbytes-dev - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.