Back to blog
AI Advertising10 min read17 September 2026

AI-Generated Ads and the CAP Code — What the ASA Will Hold You To

Adyft Guide

AI Advertising

This is not legal advice and anything in a regulated category should go past someone qualified. What it is: a plain account of where AI-generated advertising creates ASA exposure for a UK business, written because the tools produce exactly the copy patterns the CAP Code restricts and will never warn you.

The principle

The advertiser is responsible for the advertisement. Not the tool, not the agency, not the freelancer. "It was AI-generated" carries no weight, and the ASA publishes rulings with company names attached, which means the reputational cost lands before any formal sanction does.

💡 Worth internalising: AI copy reads with confidence, and confidence is precisely what makes unsubstantiated claims dangerous. A model asked for persuasive ad copy produces superlatives and outcome promises by default. It has no concept that those words are regulated.

The five patterns AI produces that CAP restricts

1. Superlatives without substantiation

"The best", "the UK's leading", "number one" — generated constantly, and each one is an objective claim requiring evidence you can produce on request. "Leading" is not a figure of speech to the ASA; it is a claim about market position.

The fix is not to soften them into vagueness. It is to replace them with something specific and true: not "the best boiler service in Leeds" but "same-day callout across Leeds, Gas Safe registered".

2. Invented statistics

The most common and the most dangerous. Ask for ad copy and you will get "trusted by 5,000 customers", "saves businesses 40% on average", "rated 4.9 by our clients". None of these are retrieved from your records — they are generated to fit the sentence rhythm.

Every one of them is a claim you must be able to evidence. Several of them, if you cannot, are straightforwardly misleading advertising.

3. Health, beauty and wellbeing outcomes

CAP restricts claims about what a treatment or product will do to a person's body. AI writes these freely — "permanent results", "clears acne", "reverses hair loss". Cosmetic and aesthetic advertising is among the most-ruled-on areas the ASA covers.

If you run a clinic, an aesthetics practice or a supplement business, treat every generated line as a draft requiring the same review any other advertisement would get.

4. Price and availability

"From £99" requires that a meaningful number of customers can actually get it at £99. "Limited time only" requires the time to actually be limited. "Free" requires no hidden condition. AI generates all three as stock phrases without knowing whether any is true of your business.

5. Testimonial framing, including synthetic people

This one is newer and catches people using AI video tools. An AI avatar speaking to camera about how good a product is reads as a customer testimonial. If it is not a real customer and is not clearly presented as an actor or illustration, that is a problem — testimonials must be genuine and held on file.

💡 The same applies to AI-generated images of "customers" and to review-style copy. There is a hard line worth adopting regardless of what the code technically permits: never publish a review, testimonial or customer name that is not real. It is not worth the exposure and it is not worth what it does to a brand when noticed.

Meta enforces its own rules, automatically

Separate from the ASA, and faster. Three patterns get rejected constantly and AI copy produces all three by default.

  • 1Implied personal attributes. "Are you struggling with debt?" is rejected; "Our service helps people manage debt" is not. AI addresses the reader directly by default, which is exactly the trigger.
  • 2Before-and-after imagery in health and cosmetic categories, regardless of how genuine the result.
  • 3Unrealistic outcome claims. The threshold is not published precisely, which makes appeals slow and frustrating.

Rejections are not compliance incidents, but a pattern of them affects account standing, which is a real operational cost. Screening AI copy for these three before submission is quicker than appealing afterwards.

A workable process

The one-hour rule

No claim, figure or superlative goes into an ad unless you could produce the evidence within an hour. This single habit removes most of the risk.

Specific beats superlative

Replace "best" with something checkable. Specific claims are also better advertising, so this costs you nothing in performance.

Look at every image

Seconds of work. Catches synthetic testimonials, before-and-afters, and anything that reads as a real person who is not.

Keep a substantiation file

If the ASA asks, they ask for evidence, and they ask promptly. A folder with the sources behind each claim turns a stressful week into an email.

What to want from a tool

Most do nothing about this and you should assume yours is one. The features that genuinely help are unglamorous: everything previewed before publication rather than automatic launching, and copy that describes what you offer rather than promising what the customer will get. A tool that publishes without showing you the creative has removed the only review step you had.

The review step stays yours

Adyft shows you every headline and image before anything publishes, so the review step stays yours. Fourteen-day free trial, no card.

Start free — no card needed
A

Adyft Team

Published 17 September 2026

© 2026 Adyft Technologies Pvt. Ltd.