AI-Generated Ads Under the CAP Code and the DMCC Act: What UK Advertisers Have to Hold

Quick answer:A UK advertiser running an AI-generated video ad has to hold three things, and none of them is an AI label. Under CAP Code rule 3.7 you must hold documentary evidence for every objective claim before the ad runs. Under rule 3.47 you must hold evidence that any testimonial is genuine, plus contact details for the person who gave it, "unless it is obviously fictitious". And since April 2025 the Digital Markets, Competition and Consumers Act 2024 bans fake consumer reviews in any medium; the CMA's guidance says a review "does not have to be written by a consumer to fall within the definition". The ASA is explicit that "There is no blanket legal requirement in the UK to disclose the use of AI in ads and there are many schools of thought, around the world, on the suitability and effectiveness of regulators insisting on such disclosure in all circumstances." The exposure is not the AI. It is a script that sounds like a customer's experience when no customer had it.

The rules are media-neutral, which cuts both ways

The ASA has said it twice, a year apart. May 2025: "Whilst the CAP and BCAP Codes do not contain AI-specific rules, our existing rules apply regardless of how content is generated, edited, or targeted." June 2026, in an article written for advertisers about to press run: "The Code is media-neutral, meaning the rules apply regardless of how the content was created."

Read that as reassurance and you have read half of it. Media-neutral means no AI checkbox, no AI rule and no AI defence. The June 2026 article closes the door on the one argument a buyer might reach for, quoting its own Stripe & Stare Ltd ruling as "a key reminder that even when ads are generated or distributed through automated AI platforms, responsibility does not shift." So the useful question is not what the Code says about AI. It is which existing rules an AI actor walks into most easily. There are three, and they sit a few lines apart in section 3.

Rule 3.7: evidence before the ad runs, not after the complaint

The rule reads: "Before distributing or submitting a marketing communication for publication, marketers must hold documentary evidence to prove claims that consumers are likely to regard as objective and that are capable of objective substantiation. The ASA may regard claims as misleading in the absence of adequate substantiation." The timing word is before.

AI video changes the working pattern here, and not because of anything the model does. A human creator improvises; an AI actor says what the brief says. On our own platform the talking-head prompt sent to the model instructs the actor to speak "ONLY these exact words to the camera, once, without repetition or improvisation", followed by your script. So every claim in the ad was typed by someone on your side. "Clears up breakouts in a week", "lasts twice as long", "dermatologist recommended": each is an objective claim you must be able to prove on launch day. The script is the checklist. Rule 3.9 adds: "Marketing communications must not mislead by omitting significant limitations and qualifications. Qualifications may clarify but must not mislead by contradicting the claims that they qualify." A condition on a claim belongs in the ad, not in a file at the office. For skincare and supplement claims specifically, the platform rules that sit on top of the Code are quoted in supplement and skincare claims in AI UGC ads.

Rule 3.47: the testimonial problem an AI actor cannot solve

The rule in full: "Marketers must hold documentary evidence that a testimonial or endorsement used in a marketing communication is genuine, unless it is obviously fictitious, and hold contact details for the person who, or organisation that, gives it."

Take a standard UGC line. "I've been using this for three weeks and my skin has honestly never looked better." Spoken by a paid creator who used the product, that is a testimonial with a person behind it, and you keep her details. Spoken by an AI actor, it is a testimonial with nobody behind it: no evidence it is genuine, because it is not, and no contact details, because there is no one to contact. The only exit is the clause after the comma, "unless it is obviously fictitious".

The fiction has to be obvious, judged from the viewer's side. A photorealistic presenter in a bedroom, holding the product, speaking in the first person, is the opposite of obvious fiction; the format exists to read as a real customer. The ASA's May 2025 article points at the way through while stopping short of a promise: "That said, potentially, making clear that deepfake content featuring in an ad is only being used for comedic effect or that an influencer is AI-generated, could well help to negate an otherwise misleading impression." "Could well help" is not "will cure".

The same article states the limit in one sentence: "It would almost certainly be against the rules to make a misleading claim in an ad, either directly or by implication, and then seek to 'disclaim' that message by disclosing that AI was used." An AI label is not a permission slip for a fake experience. And a synthetic presenter reading a real customer's review aloud runs into rule 3.50: "Marketing communications must not feature a testimonial without permission; exceptions are normally made for accurate statements taken from a published source, quotations from a publication or references to a test, trial, professional endorsement, research facility or professional journal, which may be acceptable without express permission." The words may be genuine; the author still has to have agreed to appear in an ad.

The DMCC Act: fake reviews, in speech, by anyone

The CAP Code is self-regulation. The second rulebook is statute. Schedule 20 to the Digital Markets, Competition and Consumers Act 2024 lists the commercial practices "which are in all circumstances considered unfair", and paragraph 13 bans "Submitting, or commissioning another person to submit or write" a fake consumer review. The CMA published its guidance on that paragraph, CMA208, on 4 April 2025, and its press release three days later confirmed the provisions "are now in force". Five sentences from CMA208 decide whether an AI UGC ad is inside the ban.

Put those together and the "customer experience" script is in an uncomfortable place. If an ad presents a synthetic person's scripted opinion as though it were a customer's genuine experience, it purports to be based on an experience nobody had. Paragraph 3.3 of the guidance names who is on the hook: "Anyone who engages in the commercial practice of either submitting or commissioning banned reviews will be in breach of the banned practice." Its own worked example of a fake review is a video: "Making a video that claims to demonstrate the results of using a specific product, but in fact using a different product." An AI demonstration showing results the product does not produce sits next to that example.

The consequence is what separates this from an ASA ruling. The CMA's April 2025 press release states: "Under the new consumer regime, if a company infringes consumer protection law, the CMA can fine them up to 10% of their global turnover." The same release says early fines are likely to be lower because the powers cannot be applied retrospectively, which is a comment on the first year, not a ceiling.

What the ASA has actually ruled on

As of this writing there is no published ASA ruling on an AI actor delivering a fabricated product testimonial. The ASA has published several rulings on AI-generated ads (Polyverse Inc, Cosmos Oyun Yazilim, Stripe & Stare Ltd, Saeta Tech Ltd), every one of them on harm, offence or social responsibility rather than misleadingness. The closest to an AI presenter is still instructive. On 6 August 2025 the ASA upheld a complaint against Pheon Inc over ads featuring "a video of an AI-generated young female character sitting on a bed" who introduced herself as the viewer's AI girlfriend. It found breaches of rules 1.3 (social responsibility) and 4.1, 4.8 and 4.9 (harm and offence) because the character appeared under 18 and was sexualised. The action was the standard one: "The ads must not appear again in the form complained of." The synthetic character was assessed exactly as a filmed one would have been; the AI itself was never the breach. Expect the first testimonial case to look the same: a ruling about rules 3.47 and 3.1, with "it was AI" in the background facts.

Real people and the line the ASA draws hardest

The June 2026 article reserves its bluntest sentence for likeness: "If an AI-generated celebrity 'endorsement' looks real enough to mislead consumers, it's likely breaching the Code." That is rule 3.1 ("Marketing communications must not materially mislead or be likely to do so.") at work, and it applies to a colleague, a competitor's founder or a local dermatologist as readily as to a celebrity. On our platform the AI actors are licensed characters, and our acceptable use policyrequires documented permission before any real, identifiable person is depicted and prohibits "fabricated endorsements or testimonials attributed to real people". Check that whichever tool you use draws the same line.

Two script shapes, side by side

Everything above collapses into a choice you make before you render anything. The same product, the same 15 seconds, written two ways.

The presenter script gives up some of the intimacy that makes UGC work. The trade is that it does not pretend to be anything. If you want the experiential register, the only exit 3.47 offers is "obviously fictitious". In practice that means a presenter who is plainly a character, an on-screen line saying the spokesperson is AI-generated and the scenario illustrative, or a format no viewer would take as a review. Whether a treatment is obvious enough is a judgement the ASA has not yet had to make in a testimonial case, so write down why you think yours is.

Fix the script before it exists as a video. A rewrite in the text box costs nothing; every re-render costs credits. The script templates we publish are written in the experience register, the honest default for a human creator, so move the claims into the presenter shape before running them on an AI actor for UK traffic.

How this sits next to the other regimes

A UK company rarely buys UK traffic only. In the US, the FTC's fake testimonial rule asks the same question of the script, and the FTC post works through it. The EU AI Act imposes the disclosure duty the UK does not: Article 50(4) applies from 2 August 2026 and the Commission's Guidelines treat a realistic AI presenter as a deep fake, which our EU AI Act post works through. New York requires a conspicuous disclosure when an ad produced for a commercial purpose knowingly contains a synthetic performer. TikTok requires the AIGC label, or a clear disclaimer, caption, watermark or sticker, on ad content that is completely AI-generated or significantly edited, and Meta applies its own AI label to what it detects.

The UK is the outlier because the ASA thinks a label answers the wrong question: "disclosure alone is very unlikely to mitigate the harm caused by a fundamentally misleading message." The UK asks what the ad claims and whether you can prove it, which is harder than ticking a box and survives whichever platform you run on.

A checklist for a UK AI video ad

• Read the script as a list of claims. For each objective one, name the document that substantiates it and have it in hand before launch (rule 3.7).
• Decide which shape the script is. A first-person experience needs a real, consenting customer behind it, or a treatment so obviously fictional that no viewer could take it as a review (rule 3.47).
• Never generate a presenter who resembles a real, identifiable person without documented permission (rule 3.1, and exposure outside the Code as well).
• Keep a log of which ads used AI, which script shape they used and what evidence backed them.

None of this is legal advice; it is a media buyer's reading of the primary documents.

Sources

• Committee of Advertising Practice, CAP Code section 03, Misleading advertising (rules 3.1, 3.7, 3.9, 3.47, 3.50)
• ASA, "Disclosure of AI in Advertising: Striking the Balance Between Creativity and Responsibility" (29 May 2025)
• ASA, "AI and Deepfakes: Four Things Advertisers Need to Know Before They Hit 'Run'" (11 June 2026)
• ASA, Ruling on Pheon Inc, G25-1293034 (6 August 2025)
• Digital Markets, Competition and Consumers Act 2024, Schedule 20, Commercial practices which are in all circumstances considered unfair (paragraph 13)
• Competition and Markets Authority, Fake reviews: guidance on the prohibition under paragraph 13 of Schedule 20 (CMA208) (4 April 2025, read as the published PDF)
• Competition and Markets Authority, "CMA to boost consumer and business confidence as new consumer protection regime comes into force" (7 April 2025)
• UGC Vids AI, Acceptable Use Policy (likeness and consent, prohibited content), quoted as our own document

All seven external documents were read directly on 4 September 2026.

If the claims are the problem rather than the presenter, how to script a 30-second UGC ad covers the structure. The lip-sync model comparison explains which models speak your script word for word and which paraphrase it, which matters once the script is the compliance record. If you want the presenter shape rendered on a licensed AI actor instead of a real customer, that is what UGC Vids AI does: plans start at $49 a month for 5,000 credits, and the first 3 days are free.