The Results are in: AI on the Front Line of Alcohol Advertising Regulation

The UK’s Advertising Standards Authority (ASA) has published the results from its most recent trial, which used artificial intelligence (AI) to assess alcohol-related ads’ compliance with the rules. The trial marks the ASA’s largest-scale deployment of AI-assisted monitoring to date, underscoring its commitment to innovation with AI described as a “core part of the organisation’s regulatory toolkit”. By deploying large language models (“the AI-agent”) to analyse content against a full section of the CAP code, the ASA was able to identify potential breaches before any public complaints were received. Such a result allows the ASA to remain agile and keep pace as a regulator in an area where AI-generated content is the current reality (as highlighted in an article by my colleague Willemijn Paul here). This approach is not merely informative/for greater insights; it directly informs enforcement actions. The recent ruling against Au Vodka Ltd, in which the ASA Council upheld a ban on certain ads identified for investigation by the AI-agent, illustrates this. That said, the ASA is adopting a hybrid approach with AI’s use in advertising regulation stating it is “not here to replace human judgment, but to extend its reach”

The ASA Trial: Key Findings 

The AI-agent analysed the text and imagery of nearly 6,000 online paid-for UK ads against section 18 of the CAP code in minutes. To enable the AI-agent to make such assessments it had been given prior contextual guidance, including illustrative examples taken from prior ASA rulings on alcohol. Alcoholic and alcohol-free products (under 0.5%) were assessed separately due to the differing applicable subsections of the CAP code. Any potentially noncompliant content identified by the AI-agent was flagged for expert review by the ASA. This hybrid method combines the speed and scale inherent to AI with the nuanced contextual reasoning that human judgment provides.

Key findings include: 

• Overall compliance of alcohol advertising was high. 96% of the reviewed ads were deemed likely to comply with the CAP Code.
• Alcohol-free alternatives were less compliant. 48% of the reviewed ads were assessed to have potential compliance issues.
• False positives present. 40% of ads flagged as a concern by the AI-agent were held to contain possible breaches upon ASA expert review.
• False negatives present (recall rate). The AI-agent was able to flag 65% of content with potential issues. 

Notably, the AI-agent identified non-compliant ads that had not generated public complaints highlighting the broad reach. 

Unclear or missing ABV information was the prevailing issue among the alcohol-free ads. Rule 18.19 of the Code requires marketing communications for alcohol alternatives to include a prominent statement of their ABV levels.

Looking Ahead 

The ASA’s use of AI-assisted regulation is seemingly here to stay. The trial demonstrates that AI can be deployed in tandem with human experts to assess sector-specific advertising compliance at scale. That said, it remains a complex balancing act: leveraging advances in AI whilst continuously testing and validating AI tools to identify and address inaccuracies. 

Providers in the alcohol advertisement space should expect the following:

• Increase in sector-specific guidance. As the ASA are able to efficiently identify themes/ problematic area, further targeted guidance is likely. The ASA have already released an article in response to this trial addressing the common breaches identified for the no/low alcoholic drink content, aimed at marketers to clarify the distinct regulatory requirements within rules 18.18 - 18.24 of CAP Code.
• Shift from reactive complaints-led regulation to continuous compliance. Brands should proceed on the assumption that their advertising footprint is being continuously monitored, with “market sweeps” across sensitive categories, like the alcohol industry, now a viable option for the ASA.   

Shorter publication-to-enforcement cycles. As the ASA’s reach expands and compliance issues are flagged before a complaint is received, it will be increasingly  important for advertisers to familiarise themselves with the relevant CAP code in the knowledge that this is the content the AI-agent is trained on.