Who: The Advertising Standards Authority (ASA)
Where: United Kingdom
When: 9 October 2025
Law stated as at: 25 October 2025
What happened:
The ASA has published the results of a trial using artificial intelligence (AI) to monitor compliance with alcohol advertising rules at scale. The trial analysed almost 6,000 online paid-for ads shown to UK consumers in early 2025 across search, display and social media platforms, covering alcoholic drinks, alcohol-free alternatives and related promotions.
While the ASA has previously reported effective use of its AI-based Active Ad Monitoring System to identify non-compliant ads, this marks the first time that the system has been deployed at scale across an entire section of the UK Code of Non-broadcast Advertising and Direct and Promotional Marketing (CAP Code). The ASA sought to see whether AI could effectively flag a wide range of potential compliance issues, including those not previously on the regulator’s radar.
The Active Ad Monitoring System assessed each ad’s text and imagery against the rules in section 18 of the CAP Code. Additionally, the ASA fed the system with illustrative examples drawn from previous relevant ASA rulings and additional regulatory principles used by ASA experts when assessing similar ads. The large language model (LLM) was asked to analyse each ad and flag those that it categorised as more likely to be in breach of the rules, along with a rationale as to why. Ads flagged by the AI as potentially non-compliant were then reviewed by human experts to confirm whether a breach had occurred.
The results showed that around 96% of alcohol ads reviewed likely complied with the alcohol-specific rules in the CAP Code, while 1-3% appeared to breach them, and the remaining 1% required further review.
The most frequent issues identified across non-compliant ads included:
- Unauthorised health or nutritional claims around calorie content, sugar levels or weight control, which are prohibited for alcoholic drinks.
- Promotion of irresponsible drinking.
- Use of content likely to have particular appeal to children, for example, cartoon characters or sweets-themed alcoholic products.
Ads in the alcohol-free sector showed significantly higher rates of potential non-compliance. Around 48% were flagged for potential breaches, with almost all relating to unclear or missing ABV (alcohol by volume) labelling.
As a result of this monitoring, the ASA will be contacting advertisers responsible for the identified breaches and taking action. It also plans to publish an insight article for the alcohol-free sector, focussing on ABV labelling requirements.
The trial has given the ASA technical insights, which will shape how it applies AI in future work across other sectors. In particular, the ASA will explore:
- Integrating human feedback into LLM prompting, so that assessments more closely reflect expert interpretation of the CAP Code.
- Incorporating additional ASA guidance, such as AdviceOnline articles, into the context, helping the LLM better understand how rules are applied in practice.
- Improving the model’s understanding of the ASA’s enforcement priorities, such as weighting results based on likely harm, reach or impact.
- Linking ads from the same campaign, so that the ASA can review messaging in context and avoid duplication of effort.
- Experimenting with newer models and more advanced prompting methods to improve accuracy and reduce false positives.
Why this matters:
The ASA said that the trial provided a valuable test case for how AI can contribute to complex regulatory work. The ASA concluded that AI can be used effectively to assess sector-specific ads at scale and help it spot potential issues in instances where there are no complaints, but where there could still be an impact on consumers or vulnerable groups.
However, the ASA also flagged some limitations around the use of AI. This trial focused on identifying relatively straightforward rule breaches based on the content and imagery of the ads. AI does not, however, possess the same depth of judgement or contextual reasoning as human experts. Therefore, more complex compliance issues, particularly those requiring substantiation of claims may prove more challenging for AI assessment.
With systematic AI monitoring now in operation, proactive compliance of online ads across all sectors becomes even more important.
