Two sets of rules govern AI-generated marketing in the UK, and neither bans it. The advertising codes, enforced by the ASA with CAP guidance, contain no AI-specific rules: the existing rules apply to whatever AI produces, the advertiser is responsible for it, and CAP’s test is whether the audience would be misled if AI use were not disclosed, with the caveat that disclosure cannot remedy a fundamentally deceptive message. Google’s position is that it rewards helpful, original content regardless of how it was produced and treats scaled content generated primarily to manipulate rankings as spam. The safe workflow is the same for both: AI drafts, a person checks every claim, image and testimonial against evidence, disclosure is added where its absence would mislead, and volume is limited to what a person can actually check.
- ASA and CAP
- No AI-specific rules; existing rules apply
- Quality decides; scaled abuse is spam
- Workflow
- AI drafts, a person checks every claim
The advertising rules
CAP has said that the UK Advertising Codes do not contain AI-specific rules and that the existing rules apply regardless of how content is generated. The advertiser remains responsible for the content AI produces. On disclosure, CAP suggests asking whether the audience would be misled if the use of AI were not disclosed, and whether disclosure would clarify or contradict the message; it is clear that disclosure cannot remedy fundamentally deceptive messaging. The ASA has said advertisements both for and using AI are on its radar, and CAP has described proactive monitoring of AI in advertising, with further guidance expected. In practice: an AI-generated image of a product that does not exist as shown, a testimonial that was written rather than given, or a claim the model invented, is a breach whether or not it says “made with AI”.
Google’s rules
Google’s guidance for generative AI features in Search says the fundamentals continue to apply and that it wants unique, expert-led content beyond common knowledge; its spam policies name scaled content abuse, meaning many pages generated primarily to manipulate rankings rather than help users, whether by AI or by hand. So AI-assisted writing that is accurate, specific and checked is fine, and a hundred thin AI pages on a hundred keyword variations is the thing the policy describes. The practical tell is whether a person could vouch for every page.
What has to be true of every piece
Six conditions, each checkable by the person who signs the piece off.
- 01
Every factual claim is checked against a source a person can produce, and figures are attributed. Invented statistics are the most common AI marketing failure and the easiest to catch.
- 02
Testimonials and reviews are real and given by identifiable customers. AI does not write them, and AI-written ones are not published as if given.
- 03
Images of products, premises, people and results show what exists. AI-generated imagery is used for illustration, not evidence, and is labelled where a reader would otherwise take it as a photograph.
- 04
Comparisons with competitors are accurate and substantiated, and the tone is measured.
- 05
Disclosure is added where its absence would mislead: an AI voice presented as a person, an AI image presented as a photograph, an AI chat presented as a human.
- 06
Personal data used to personalise content follows PECR and UK GDPR; the model does not decide who is contacted.
A workflow that stays inside the rules
Brief the model with the facts, the sources and the firm’s tone, using a shared template. Draft. A named person checks every claim, image and quote against the evidence and signs it off, with the check recorded. Add disclosure where needed. Publish at a volume the checker can sustain, which for most small firms is a handful of pieces a week rather than a hundred. Keep the sources with the piece so a complaint can be answered in an afternoon. This is also what produces content AI search will cite, because it is specific and sourced.
RelatedPrompting for business teams AI acceptable-use policy template
What to avoid
Bulk-generated location or service pages that differ by a place name. Reviews written by a model. AI images of “our team”. Claims about AI in your own product that the product does not do, which the ASA has specifically flagged. And the assumption that a “generated by AI” label makes any of that acceptable, because CAP has said it does not.
Common questions
Is AI-generated advertising legal in the UK?
Yes. The advertising codes have no AI-specific rules; the existing rules apply and the advertiser is responsible for whatever AI produces. Misleading content is a breach however it was made.
Do I have to disclose that content was made with AI?
CAP’s test is whether the audience would be misled without disclosure. An AI image presented as a photograph or an AI voice presented as a person needs it; a checked, accurate blog post does not. Disclosure never cures a deceptive message.
Does Google penalise AI-written content?
No. Google rewards helpful, original content regardless of how it was produced and treats scaled content generated to manipulate rankings as spam. Quality and checking decide, not the tool.
Can AI write customer reviews or testimonials?
No. Reviews and testimonials must be real and given by identifiable customers. AI-written ones published as genuine breach the advertising rules and platform policies.
Can we use AI images in our marketing?
For illustration, yes. Not as evidence of a product, premises, person or result that does not exist as shown, and labelled where a reader would otherwise take it as a photograph.
How much AI content should we publish?
As much as a named person can check claim by claim. For most small firms that is a few pieces a week. Volume beyond checking capacity is where both the ASA’s and Google’s rules bite.