Purpose
Turn performance data and client objectives into a ranked backlog of 12 disciplined A/B test hypotheses — each isolating one variable, stating a directional prediction, and defining a success metric and sample size — so creative testing stops being random and starts compounding into learnings.
When to Use
Invoke this skill when the user asks to:
- Plan the next round of creative A/B tests for a client
- Build a creative testing roadmap or experimentation backlog
- Decide which creative variable to test next based on current results
- Turn a batch of underperforming ads into structured test ideas
- Justify a testing plan to a client with clear hypotheses and success criteria
- Diagnose whether a plateau is a hook, format, message, or offer problem
Inputs Required
Before starting, collect:
- Current creative performance: Top and bottom performers with metrics — CTR, hook rate, CPA, ROAS, thumbstop, conversion rate
- Platform(s): Meta, TikTok, YouTube, Google Demand Gen, etc.
- Client objective: The KPI that matters most — CPA reduction, ROAS lift, scale, lead quality
- What's already been tested: Known winners, known losers, tested angles to avoid repeating
- Audience: Who the ads target and their awareness level (cold / warm / retargeting)
- Creative constraints: Brand guidelines, mandatory elements, formats available (UGC, static, animation)
- Monthly spend / conversion volume: Needed to set realistic sample sizes
If conversion volume is missing, assume a mid-range budget and flag that sample-size guidance is directional until real volume is confirmed.
Instructions
Generate 12 Hypotheses Across 6 Creative Dimensions
Each hypothesis follows this exact structure:
- Hypothesis statement: "If we [change X], then [metric] will [direction] because [reasoning grounded in the data or audience psychology]."
- Variable isolated: The single element changing — everything else held constant
- Control vs. variant: What the current creative does vs. what the test does differently
- Predicted outcome: Directional prediction with rough magnitude (e.g., "hook rate +15–25%")
- Primary success metric: The one metric that declares a winner, tied to the client objective
- Minimum sample size / duration: Conversions or impressions needed, and days to run
- Confidence in hypothesis: High / Medium / Low, based on how much data supports it
Every hypothesis must change exactly one variable. If two elements move, split it into two hypotheses.
Dimension 1 — Hook / Opening (3 seconds) (3 hypotheses) The single highest-leverage variable on Meta and TikTok. Tests what stops the scroll.
- Test types: problem-first vs. benefit-first opener, visual pattern interrupt, on-screen text hook, question vs. statement, motion in first frame
- Tie predictions to hook rate / thumbstop rate, then downstream to CTR
- Highest priority when thumbstop or 3-second view rate is below platform benchmark
Dimension 2 — Message / Angle (3 hypotheses) The core promise or problem being framed. Tests what the ad is about, not how it looks.
- Test types: pain-agitation vs. aspiration, feature vs. outcome, social proof vs. authority, new mechanism vs. familiar framing
- Each angle must map to a distinct audience motivation — not a reworded version of the control
- Tie predictions to CTR and conversion rate; angle changes affect qualification, not just clicks
Dimension 3 — Format / Production Style (2 hypotheses) The container and production approach. Tests how the message is delivered.
- Test types: UGC vs. polished, static vs. video, talking-head vs. voiceover-over-b-roll, single vs. carousel, native vs. branded
- Format tests often reveal cost efficiency wins even when message is held constant
- Tie predictions to CPM, CPA, and completion rate
Dimension 4 — Offer / CTA (2 hypotheses) The ask and the value exchange. Tests the transaction, not the attention.
- Test types: CTA verb ("Shop" vs. "Learn"), urgency vs. evergreen framing, discount vs. bonus, hard offer vs. soft lead magnet
- Tie predictions directly to conversion rate and CPA — this is a bottom-funnel dimension
- Only test offers within what the client can actually fulfill
Dimension 5 — Social Proof / Trust (1 hypothesis) Credibility signals inside the creative. Tests trust as a conversion lever.
- Test types: testimonial vs. no testimonial, star rating overlay, review count, before/after, expert endorsement
- Strongest for warm and retargeting audiences and higher-consideration products
- Tie predictions to conversion rate and, where measurable, lead quality
Dimension 6 — Audience-Creative Fit (1 hypothesis) The interaction between message and who sees it. Tests whether a segment needs a different creative.
- Test types: same creative across segments vs. tailored creative per awareness level, demographic-specific messaging, cold vs. retargeting-specific hooks
- Tie predictions to segment-level CPA and ROAS
- Best when segment reporting shows one audience underperforming with shared creative
Rank All 12 by Expected Impact
After generating, order the full set into a priority ranking. Weight by:
- Leverage — how much the variable typically moves the target KPI (hook and angle usually outrank offer for cold traffic)
- Evidence — how strongly the current data points to this being the bottleneck
- Feasibility — production effort and sample-size reachability within the current budget
State the reasoning in one line per ranked item.
Recommend the First 3 Tests to Launch
Name the three to run first, why they lead, and whether they can run concurrently or must be sequential (dependent variables can't run in the same window).
Output Format
- Structure: 6 labeled dimensions, each hypothesis using the fixed 7-field template; followed by a ranked priority table and a "launch first" recommendation
- Length: 1,100–1,500 words
- Format: Structured — bulleted hypothesis blocks, a ranking table, short prose recommendation
- Sections: 6 dimension groups → Ranked priority table (all 12) → First 3 to launch
- Tone: Creative strategist — data-anchored, decisive, allergic to vague "let's try this and see"
Quality Checklist
Before delivering output, verify:
- [ ] All 12 hypotheses are present with the correct count per dimension (3, 3, 2, 2, 1, 1)
- [ ] Every hypothesis isolates exactly one variable — no compound changes
- [ ] Each hypothesis has a directional prediction AND a single primary success metric
- [ ] Every success metric maps to the stated client objective, not a vanity metric
- [ ] The ranking is by expected impact with one-line reasoning — not arbitrary order
- [ ] The "launch first" recommendation flags whether tests can run concurrently
Notes
- The most common mistake in agency creative testing is testing multiple variables at once — enforce single-variable isolation ruthlessly, even when the client wants to "just ship the new one."
- Hook tests almost always deliver the fastest, cheapest learnings on paid social — lead with them when thumbstop rate is weak.
- Sample-size guidance is only as good as the conversion volume provided; if the account can't reach significance in a reasonable window, recommend testing on an upper-funnel proxy metric (CTR, hook rate) instead of waiting on conversions.
- A losing test is still a win if it kills a bad angle cheaply — frame the roadmap as a learning engine, not a hit rate.
How to use this skill
This file is a Claude Code skill in Anthropic's standard format — markdown with YAML frontmatter at the top.
Installing in Claude Code:
~/.claude/skills/<folder-name>/SKILL.md— auto-loads in every session anywhere on your machine (recommended for skills you'll use across multiple projects)<your-project>/.claude/skills/<folder-name>/SKILL.md— auto-loads only when working in that project (recommended for project-specific skills)
Restart Claude Code or open a new session to register.
Using in Claude.ai, ChatGPT, Gemini, or other AI tools:
- Drag this file into a new chat, or copy and paste its contents
- Send your task or request as your next message — the AI will apply the skill where it fits
For Claude Code skill details: https://docs.anthropic.com/en/docs/claude-code/skills