What 72,400 ads reveal about compliance monitoring

On 15 October 2026, the ASA published two enforcement reports covering paid advertising on Meta and Google between February and July 2026. One examined supplement claims concerning ADHD and autism. The other addressed prostate supplements and PSA home testing kits.

A regulator can use automated screening to find patterns that would be laborious to identify manually. A business reviewing a sample of approved assets may consequently know less about its live advertising than the regulator investigating it. The reports describe how detected ads were assessed and what enforcement followed. They show how much work remains once a system raises a flag.

Captured, flagged and confirmed are different counts

For the ADHD and autism work, the Active Ad Monitoring system captured 72,400 unique paid supplement ads that appeared relevant. Screening flagged 1,595 ads from 17 advertisers. Subsequent assessment found that 1,383 ads broke the rules. Those are three distinct stages, rather than interchangeable counts of non-compliant advertising.

The full report describes claims about preventing, treating or curing traits of ADHD or autism and other problematic health claims. It reports that advertising volume fell after contact, but compliance among the remaining monitored ads did not improve sustainably.

The prostate report covers a different selected population: 2,730 unique ads from advertisers contacted during the enforcement work. Of those, 2,580 were flagged as containing potentially non-compliant claims. That figure should not be relabelled as confirmed breaches or extrapolated into a prevalence estimate for all health advertising.

Neither report establishes how many problematic ads the system missed or the compliance rate of the entire market. The capture figures cannot tell us how completely a market was observed, while results from a targeted group should not be attributed to the average advertiser.

For a business building its own monitoring, those limitations suggest what to establish first: channel and format coverage, the meaning of a flag and the assessment needed to confirm an issue. Without that information, management cannot tell whether a changing count reflects behaviour, detection or the population being observed.

Find the versions customers can see

An approval system records what a team intended to release. Once agencies and distribution partners have adapted that material, the record may describe only part of what customers can see.

Imagine a hypothetical supplement brand that approves a narrow health claim for a product page. An agency creates social variants, an affiliate adds a testimonial and an older version remains active on another account. The original approval can be perfectly documented while providing an incomplete view of the marketing in circulation.

Before buying a detection tool, the brand needs to establish where its communications appear and who can change them. A system that only sees centrally managed accounts could work accurately while missing the affiliate whose testimonial has turned an acceptable claim into a problematic one.

The customer journey can carry a proposition across several assets. Monitoring should preserve enough context to assess what the audience is being told, including the destination and relevant surrounding claims.

The brand need not review every page with the same intensity. It could prioritise material changes or high-risk claims, provided management understands what has been left outside the check. A sample becomes misleading when it is reported as evidence about advertising the system never saw.

Getting a flagged advert changed

AI-assisted screening can create a new queue more quickly than a business can resolve it. If every ambiguous phrase becomes an urgent escalation, specialist attention is consumed by low-value objections. If the tool is tuned to reduce the queue without understanding what it misses, important issues may become less visible.

The person assessing a flag needs the creative, relevant product information and the basis for the objection. Once they confirm an issue, somebody must be able to correct or withdraw the ad. That may be straightforward in a centrally controlled account and considerably harder when a distribution partner controls publication.

The review record should preserve why the issue was confirmed or dismissed. A dismissal can reveal a false positive, a missing piece of evidence or an ambiguous policy interpretation. Treating all dismissals as the same event loses information that could improve both the monitoring system and the organisation’s guidance.

The product information behind an approved claim also needs to be current. A system could locate every instance of a sentence and still assess it incorrectly against an obsolete product version. Increasing coverage would then multiply the same mistaken finding.

Closing the ticket should prompt a check of the live advertising. An old version might remain active in another account, or an affiliate might have copied it before the correction. Without that check, the business has evidence that someone requested a change rather than evidence that customers stopped seeing the claim.

What repeated findings reveal about production

The commercial value of monitoring should extend beyond removing individual ads. Repeated findings may reveal that a briefing template encourages unsupported claims, that local teams lack usable guidance or that an agency repeatedly works from outdated evidence. Those patterns point towards changes earlier in the process.

Incident counts need to be read alongside changes in monitoring coverage and advertising volume. A rise could follow the addition of affiliate accounts; a fall could reflect fewer campaigns or a loss of access to a channel. Either could be mistaken for a change in compliance if the report omitted those circumstances.

Speed measures have the same weakness if teams can close findings without checking the live communication. A faster queue is a poor result when the advertisement remains unchanged.

The ASA reports offer no universal prescription for monitoring frequency or technology. They do show why the original approval file is an incomplete account of a campaign. Knowing which versions remain in circulation and who can change them gives a business a better chance of correcting a problem before it is repeated in the next round of creative.