Lichtenstein Creative Testing

Based on Denys Lichtenstein's public method. Not affiliated with or endorsed by Denys Lichtenstein.

When to use

Use this skill to turn an account's creative bottleneck into a repeatable testing and production cycle. It supports a new testing agenda, a refresh plan for fatigued campaigns, UGC creator briefs, and a review of results that should inform the next wave. Produce a practical plan using templates/output-template.md, with hypotheses, briefs, measurement rules, responsibilities, and a learning log.

The core distinction is between Exploration, which discovers substantially different concepts, and Iteration, which improves an evidenced winner. Keep both visible. A backlog of cosmetic variations cannot substitute for new ideas, and continually replacing everything cannot substitute for exploiting what already works.

Treat the published volume ranges as starting points: 8 to 12 substantively distinct creatives per campaign, replenished through weekly waves of 3 to 5 variants. Adjust explicitly to budget, production capacity, and the signal each test can realistically receive. These are different counts, not requirements to launch 12 new ads every week. No universal budget split, CPA threshold, or conversion minimum is supplied by the public method.

Procedure

1. Establish the decision and evidence. Read the attribution boundaries in references/method.md. Record the product, offer, objective, existing results, capacity, and missing inputs. Separate account facts from assumptions. Use references/sources.md to check the origin of methodological claims, especially when adding a numerical rule.

2. Check measurement and economics. Follow references/method.md. Identify the conversion event, attribution window, CPA or other business target, and tracking gaps. Put account-level MER and contribution margin beside platform results. A measurement problem makes the test inconclusive rather than proving the creative failed.

3. Classify the creative inventory. Apply references/method.md. Label each asset as exploratory, an iteration of a named winner, or unproven. Count genuine concepts separately from exports and hook variants. Preserve parent asset IDs so learnings remain traceable.