Build a claims inventory before variants
The dangerous copy prompt is "make this more compelling." Claude can intensify a promise faster than your evidence can support it. Before asking for copy, create a claims inventory with approved product facts, proof, prohibited claims, substantiation, and review owner.
The FTC has repeatedly signaled that AI does not create an exemption from deceptive-practices law. For marketers, the practical rule is simple: if Claude writes a claim, your company still owns the claim.
Ask Claude to label each line of draft copy as fact, interpretation, aspiration, testimonial, comparison, or call to action. Then remove or rewrite anything whose proof status is weak.
There is no AI exemption.
Use a message brief with constraints
A useful copy brief has more negatives than most writers expect. Claude should know what not to say: no fake urgency, no fabricated results, no invented customer quotes, no unsupported comparisons, no medical or financial implication unless approved.
Give Claude the persona, pain, product, proof, objections, offer, channel, and voice. Then ask for distinct strategic angles before asking for final copy.
This separates thinking from wording. You can reject a strategy before it becomes a polished asset.
- Persona: situation, sophistication, urgency, objections.
- Product truth: what it does, what it does not do, what is unknown.
- Proof: data, demos, reviews, case studies, certifications, screenshots.
- Compliance: forbidden words, regulated claims, required disclaimers.
Generate variants by strategy, not by synonym
Claude is useful for exploring message space. Do not ask for twenty headlines that say the same thing. Ask for five strategies: risk reversal, speed, status, cost of delay, and clarity. Each strategy should include the claim, proof needed, likely objection, and sample headline.
For paid ads and landing pages, require a proof note under every variant. That note should tell the human reviewer what evidence must exist before the copy can ship.
Then choose one strategy and request a complete asset with a clear source trail.
Use research for market context, not fake certainty
Claude Research can help gather competitor language, customer vocabulary, and market context when web search is enabled. Treat those findings as inputs, not as final truth.
For citation-ready content, ask Claude to separate source quotations from interpretation. For customer language, use real interviews, support tickets, reviews you are permitted to analyze, or sales-call notes under your company policy.
The output should improve specificity. If the final copy sounds like any SaaS page, the research stage failed.
FAQ
Can Claude write ads?
Yes, but you should supply approved claims and require proof notes so generated ad copy does not invent substantiation.
Can Claude create testimonials?
Do not fabricate testimonials. Use real customer statements only with the permissions and disclosures required for your market.
What is the best copywriting use case?
Message exploration, objection mapping, structural critique, and variant generation tied to a claims inventory.
Primary sources
FTC enforcement examples for deceptive AI claims, fake reviews, and unsupported performance promises.
Anthropic Docs Prompting best practices Reviewed 2026-07-06Official guidance on clarity, context, examples, XML tags, roles, and long-context prompt structure.
Claude Help Center What are projects? March 16, 2026Official description of project workspaces, project knowledge, and project instructions.
Claude Help Center Use research on Claude June 2026Official note that Claude Research works with web search and returns checkable citations.