What the AI marketing tools advertising rules mean for campaign sign-off

What to take away

The CAP Code applies to an AI-written ad exactly as it applies to one written by a person, so the tool is never your defence.
Rule 3.1 covers misleading claims, and rule 3.7 requires documentary evidence to exist before publication.
Responsibility rests with the advertiser, so name one signatory before the campaign runs.
The ICO guide "Explaining decisions made with AI" explains how to describe automated decisions to the people affected.
Nothing forces you to label copy as AI-written, but leaving the fact out can still mislead.

Which rules bind an ad written with AI tools

The CAP Code governs non-broadcast advertising across the UK, and the BCAP Code covers broadcast ads. The Advertising Standards Authority enforces both. Neither text exempts automated or AI-assisted work, so the wording on the page is judged on its own merits. The current texts sit on the ASA's advertising codes page.

Two clauses carry most of the weight. Rule 3.1 says a marketing communication must not materially mislead. Rule 3.7 requires documentary evidence for claims that are objective and capable of substantiation, held before the ad runs. A vendor's own performance report is not your substantiation.

A complaint is judged on the ad as it ran, so a quiet edit afterwards does not repair the original claim. The ASA publishes its rulings, which makes the evidence file a reputational matter as well as a regulatory one.

If you are still working out how advertising sits beside data protection and consumer law, our guide to AI marketing tools rules and compliance maps the pieces before you brief an agency.

Building the evidence file before sign-off

Start the file before the campaign goes live, not after a complaint lands.

The five lines every evidence file needs

Claimthe promise the ad makes, written as one sentence.
Methodthe test, sample or source behind it.
Datewhen that evidence was created.
Reviewerthe named person who checked the claim.
Retentionhow long the file is kept, and who owns it.

Evidence file before publication

Support for response-rate claims
Method and date for comparisons
Ad judged as audience saw it
Named rival claims need method
Post-complaint exports do not count

Sign-off question / What to hold

Is each claim substantiated?: Method, sample, date, reviewer
Does the tool score or rank people?: Explanation of logic, data and consequences
Who answers if the copy misleads?: Supplier contract and named signatory
What did the audience see?: Original creative with a timestamp

Fill this in before launch. A dashboard screenshot pulled together after a complaint will not do.

Measurement claims belong in the same file. Our AI marketing analytics tools buyer's checklist covers what to ask a vendor about reporting before you rely on its numbers.

Transparency duties attached to automated decisions

Data protection law gives people rights over decisions taken about them without meaningful human involvement. Targeting that scores or filters individuals can fall inside that duty.

The ICO's guidance on explaining decisions made with artificial intelligence, published with the Alan Turing Institute in 2020, sets out how to describe the logic, the data used and the likely consequences.

For example, a tool that drops a customer from a campaign because of a predicted low value is making a decision about that person. Write the explanation before someone asks for it, and record who reviewed the outcome.

If a tool scores, ranks or excludes customers, the explanation is a sign-off item, not a support task.

Explaining automated decisions

Tool scores or filters a customer
Describe the logic in plain language
Note the data used
State likely consequences
Record the human reviewer

Platform terms and supplier evidence

The Online Safety Act 2023 places duties on regulated online services, and Ofcom's codes of practice shape the content and targeting features platforms offer. A campaign built on one platform feature is only as stable as that feature.

Check terms before you commit, and treat a mid-campaign change as a reason to revisit the plan.

The government's guidelines for AI procurement were written for public bodies, but the questions carry over. Ask what the system does and what data it needs. Then ask who is accountable when output is wrong, and how you extract your data at the end.

Our article on AI marketing tools commercial contracts in England covers indemnities, data processing and termination, which usually decide the argument when output goes wrong. Have a solicitor read anything you sign.

Common questions

Do these rules differ in Scotland, Wales or Northern Ireland?

No. The CAP Code, the BCAP Code, UK data protection law and the Online Safety Act 2023 apply across the UK, so one checklist covers all four nations.

Who is liable when a tool writes a misleading claim?

The advertiser. The ASA assesses the ad as published, so the process behind it does not move responsibility away from the brand that paid for it. A platform may also act under its own terms.

Must we tell people an ad was written with AI?

There is no general labelling duty for AI-written marketing copy. Disclosure becomes necessary where an automated decision affects someone, or where silence would mislead the audience.

What should stay on file, and for how long?

The brief, the substantiation for each claim, the human review record and the vendor's own evidence. Keep them while the campaign could still attract a complaint. A specific case needs a qualified adviser.