When AI becomes the advertising claim

On 14 October 2026, the ASA upheld complaints concerning Finelo and HRY Games, trading as AI Scorecast. In both cases, the advertising presented AI as a reason consumers could expect financial success. The rulings concern those promises and the risks omitted from them; they impose no general prohibition on advertising AI.

“AI-powered” may describe a product accurately while helping to sell something it cannot deliver. A demonstration in which the software handles a difficult task and the customer enjoys the result can suggest that experience and caution are no longer necessary. Confirming that the product uses AI leaves that implied promise unexamined.

AI was presented as doing the difficult part

Finelo’s ads presented AI as taking care of investing with little experience or effort. The service actually provided education and simulated trading rather than trading consumers’ real funds. The ASA upheld three issues, including misleading impressions about the service, investment risk and the ease of obtaining returns. It applied section 14 of the CAP Code to the financial marketing communications, despite the advertiser’s argument that its product was educational.

AI Scorecast’s ad claimed 97% predictive accuracy and connected its service with betting success. The ASA concluded that the impression of likely profit, without information about betting risk, was misleading and irresponsible. The advertiser did not respond to its enquiries. The claimed accuracy was an advertising assertion, not a result the ASA independently established.

In each case, the customer was invited to rely on the technology for judgement they might otherwise have expected to exercise themselves. Following the system could seem sufficient, even though investing and betting still involved uncertainty.

The same issue could arise in advertising an AI campaign optimiser. A demonstration might show that it produces recommendations, while the sales message promises better returns. Evidence for the first claim would leave the second unresolved.

The stronger claim need never appear in a sentence. Showing a novice starting the tool, doing very little and enjoying an impressive result may convey effortless success through the sequence alone. A reviewer needs to examine that sequence as customers will encounter it.

What the performance evidence needs to cover

A supplier may have strong results from general model tests without having tested the business outcome its advertisement promises. That gap becomes especially important when the copy invites customers to rely on the application for decisions.

Imagine the campaign optimiser being advertised as increasing the return on every campaign. Evidence that it generates plausible recommendations would leave the central claim unresolved. Commercial results also depend on the offer, audience, spending and execution. The advertiser would need evidence relevant to the promised outcome, including a clear account of the circumstances in which it had been assessed.

The distinction between model capability and application performance matters to the seller as much as the buyer. An advertised capability needs evidence that addresses the actual proposition, population and conditions of use.

The burden depends on what is claimed. A statement that a function is available may be straightforward to verify; reliable results across an entire business activity demand much more. Recording the expected consumer interpretation alongside the supporting evidence helps expose where a technically accurate description has been stretched into a broader assurance.

Risk cannot be reviewed in isolation from the creative

If the central narrative suggests certainty or effortless returns, a separate qualification may do little to convey the uncertainty involved. The presence of risk wording cannot settle the assessment without considering how prominently it appears and whether it contradicts the message around it.

This is especially relevant to video, creator content and demonstrations. A script can use cautious language while the visuals suggest complete automation. A presenter can describe a product as educational while showing an apparent real-world result. Individual components may point in different directions, leaving the audience with a proposition the reviewer never explicitly approved.

In the Midnite ruling on AI-generated advertising, the apparent age of a generated character caused the breach. The Finelo and Scorecast cases concern beliefs about a service. Both kinds of problem can escape a review confined to the written script.

A reviewer therefore needs to know whether the service executes actions, provides suggestions or merely simulates a process. The script, demonstration and destination page should agree on that point. A product described precisely in the brief can become something quite different by the time a video has been edited.

What an earlier approval actually covers

An approved claims library can help, provided it preserves the limits of the approval. A capability statement supported for one feature, task or customer group should not become a general permission to describe the whole service as reliable. Approval needs to remain connected to the evidence and the circumstances in which the statement was justified.

For example, a team might approve a description of a tool summarising specified documents. A later campaign could quietly expand that into a promise that the tool understands every obligation relevant to the user’s business. The second statement introduces questions about coverage, interpretation and organisational context that the original evidence never answered.

Approved claims depend on current product information. As features change, the team needs to check whether the evidence still supports the wording in use. A sign-off for document summarisation says little about a later promise to understand every obligation affecting a business.

Overstating autonomy may also attract customers who expect work the product cannot perform. Support teams then have to explain limits the advertisement appeared to remove, with an avoidable cost to both parties.

An advertisement that presents AI as doing the difficult part also tells the customer something about the role they are expected to play. It may imply that expertise, effort or caution are no longer necessary. That implied transfer of responsibility deserves as much scrutiny as an explicit performance figure.