On 27 August 2026, the Advertising Standards Authority published a guide titled "Regulating the Illusion of Intelligence in Ads." Five days later, it banned an ad for an AI looksmaxxing app called Maxxing, citing exactly the framework the guide had just laid out. Nobody had complained about the ad. The ASA found it itself.
That detail matters more than the ruling.
What actually changed, and what didn't
No new rules were added to the CAP Code. The ASA is explicit about this in its own guide: the Code already applies to ads for AI products the same way it applies to any other ad, because "the CAP Code only applies to ads, not the products advertised." The regulator isn't ruling on whether an AI chatbot, dating app or looksmaxxing tool should exist. It's ruling on how that product is presented in the ad.
What changed is that the ASA has now published, in one place, exactly how it intends to apply existing rules (1.3, 1.7, 3.1, 3.7, 3.11 and 4.9 among them) to AI-specific ad content. And within days, it demonstrated it's actively hunting for ads that cross those lines rather than waiting for someone to report them.
The five categories the ASA says it's watching
The guide groups AI-product ads into five recurring failure patterns. None of the underlying principles are new to advertising law. What's new is seeing them named specifically for AI marketing, with the ASA effectively pre-announcing where it will look.
- Productivity and earnings claims. Ads that lean on a vague word like "productivity" often get away with puffery (the ASA's own example: a car insurance ad claiming it "really solves your problems" was ruled acceptable exaggeration). But ads claiming an AI tool can generate income with minimal effort, trading stocks or selling AI-generated art, are held to the same standard as multi-level marketing and business course ads that made unrealistic earnings claims. If the ad promises income, the product has to deliver it, and you need evidence.
- Homework helpers. Essay mills have been illegal in England since 2022. The ASA has already upheld complaints against ads implying students could submit AI-written essays as their own work without detection. An ad for an AI "homework helper" that looks more like a ghostwriter than a spellchecker is where the line sits.
- Companionship and dating advice apps. The ASA has ruled against ads for AI companion apps that objectified women, and flagged the risk that AI dating advice content can reinforce harmful gender stereotypes, particularly given the regulator's awareness of Ofcom's 2025 research into the "manosphere." An ad that mocks a man for needing dating advice, or reduces a woman to an idealised avatar, breaches the same gender stereotyping guidance that applies to any other ad category.
- Looksmaxxing and body image tools. This is the category that produced the Maxxing ruling. AI tools that analyse a user's face or body and return a score are treated as making an objective, testable claim, not a lifestyle observation. If you can't prove the before-and-after result is genuine and typical, the ad is misleading. If the ad targets people who could be under 18, the responsibility bar goes up further.
- Health, wellness and "scan and analyse" claims. AI tools claiming to assess nutrition, health, or wellness from a photo or input are judged on the accuracy they can prove, not the accuracy they imply. The ASA has already ruled that meal-replacement ads promising to replace all food were irresponsible, and that unqualified weight-loss timelines are not permitted. It's applying the same standard to AI-generated health advice, and has said plainly it does not want ads implying an AI chatbot can substitute for a qualified therapist, especially in a crisis. There's also a specific warning about "free scan, paid advice" business models: if the ad doesn't disclose that the useful part of the service sits behind a paywall, that's a misleading omission on its own.
The real change is proactive detection
Here's the mechanism worth sitting with. The Maxxing ruling states plainly that the ad "was identified for investigation following intelligence gathered by our Active Ad Monitoring system, which uses AI to proactively search for online ads that might break the rules." One complaint was eventually logged against the ad, but the investigation didn't start there. The advertiser never responded to the ASA's enquiries at all, and was banned and named regardless.
For years, the working assumption behind a lot of ad compliance has been simple: if nobody complains, nobody's watching. That assumption no longer holds for AI-related claims. The ASA published its interpretive framework on 27 August and used it days later, off the back of automated monitoring rather than a public complaint. The gap between "we published guidance" and "we're actively enforcing it at scale" has effectively closed.
What this means for your review process
A few practical takeaways for anyone marketing an AI-enabled product, or any product using AI-generated content in the ad itself:
- Treat any before-and-after, score, or percentage claim as an evidential claim, not a creative device. If you can't substantiate it with real, representative results, don't run it.
- Check "free scan" or "free analysis" flows for a hidden paywall. If the ad doesn't disclose that the meaningful output requires payment, that's a standalone breach regardless of whether the claim itself is accurate.
- Scrutinise anything close to gender stereotyping, particularly in companionship, dating, or self-improvement framing. This is a well-established CAP focus area, not a new invention for AI.
- Assume health, earnings, and productivity claims involving AI will be tested against the same evidence bar as any other category. There's no AI exemption anywhere in the guidance.
- Don't rely on the absence of complaints as a signal of compliance. The regulator that used to need one is now finding ads itself.
The bottom line
The ASA didn't write a new rulebook for AI ads. It wrote a clearer map of an old one, then built a system to walk it without waiting to be asked. For marketing and compliance teams, the operational question isn't just "does this ad comply with the CAP Code." It's "would this survive being found by a regulator before a single customer sees it." Increasingly, that's not a hypothetical.
Sources: ASA, "Regulating the Illusion of Intelligence in Ads," CAP News, 27 August 2026 (asa.org.uk). ASA Ruling on Glow Up LLC t/a Maxxing, 2 September 2026 (asa.org.uk).
