Ad copy prompt — where the law catches AI-written marketing

What makes this hard

Ad copy's problem is that it writes *too* well. AI is good at persuasive sentences, so unsupported efficacy claims and superlatives blend in naturally. 'Industry #1', 'your skin will change'. Attractive as copy, but published without substantiation and an as-of date, they sit exactly where advertising regulation bites.

Requests without target and channel are common too. The same product needs different writing for a search ad, an Instagram post, and a landing page. Who reads it, where, and what they should do after. Without those three you get a middle-of-the-road text usable nowhere.

What to add to the request

1. Fix target, channel, and conversion goal

2. Demand sources for numbers and claims

Write: no figures without a source; superlatives only with substantiation and an as-of date. Same for competitor comparisons. Without this line, plausible numbers walk into the copy, and filtering them out becomes your job.

3. Make room for disclosure and regulation checks

For sponsored posts, put the material-connection disclosure at the top of the body, not in a hashtag pile. That is what the FTC's Endorsement Guides require in the US, and equivalents exist elsewhere. For fields with their own review regimes, such as medical, supplements and finance, don't let AI assert the rules: have it name which regulation may apply and leave it as a check item.

The four parts of the finished instruction

• Core message — if the copy could say one thing, what is it
• Variants per channel — how the same message changes between a search ad line, a post, and a landing page
• CTA — what the reader should do after reading
• Banned-expressions list — words the brand doesn't use and words regulation catches

What a channel variant is

The same core message takes its length and order from the channel. A search ad is one line that has to connect to the query within a character limit; a social post has to stop the scroll in its first sentence; a landing page puts claim, evidence, and the action button in sequence. The instruction produces a variant per channel while the core message and the banned-expressions list are shared across all of them.

Where a figure goes in, its source goes in with it. '98% satisfaction' isn't used without the survey body, period, and sample, and superlatives and competitor comparisons appear only with substantiation and an as-of date. That rule doesn't change when the channel does.

Checklist for writing it yourself

1. 1Did you state target reader, channel, conversion goal
2. 2Did you forbid unsourced figures, superlatives, efficacy claims
3. 3For sponsored posts, did you pin the disclosure to the top of the body
4. 4Did you give the brand's tone and banned-words list
5. 5Did you leave a check slot for whether a regulated field applies