Development
The UK's Committee of Advertising Practice has applied a familiar rule to a new category: an AI claim is still an advertising claim.
CAP's 27 August guidance focuses on how AI products are presented in non-broadcast advertising. The code does not assess a product merely because it uses AI. It assesses what the advertisement communicates to a likely consumer and whether the advertiser can support that meaning.
The guidance gives the example of an AI diet assistant promoted as able to calculate nutrition accurately from a photograph. A claim like that needs evidence. Without it, the advertisement could mislead and, depending on the potential harm, may also be irresponsible.
The underlying CAP Code is direct. Rule 3.7 requires documentary evidence before publication for objective claims capable of substantiation. Rule 3.11 prohibits misleading exaggeration of a product's capability or performance.
Why it matters
AI marketing often compresses several different claims into one sentence: what the system can detect, how accurately it performs, which audience it works for, and what result a user can expect. Evidence for one part does not automatically support the rest.
The practical test is sentence-level. What would a reasonable person understand this line to promise? Which product version was tested? Under what conditions? How often did it fail? Are material limits visible before the person acts on the claim?
Brand, ecommerce, product, and legal teams need one shared claim register while the campaign is still editable. For every objective line, record the evidence owner, tested version, audience, test date, known limitations, qualification, and final approval. If the evidence supports a narrower statement than the draft, narrow the statement.
What to watch next
This is not a new AI statute, a ruling on every AI product, or legal advice. The evidence burden depends on the product, audience, context, wording, and any sector-specific rules. For a specific campaign, use the current CAP Code and relevant ASA rulings, with qualified advice where needed.
The useful principle is stable: if the product cannot reliably deliver the promise being written, the fix is not a quieter disclaimer. The promise should change.