“Undisclosed AI” is fast becoming a consumer trust issue. With 2 December just two months away, the EU’s Article 50 transparency rules need attention now. Since 2 August, organisations have been required to label deepfakes. The December deadline is more specific: AI tools already on the market must be able to embed machine-readable markers in their outputs. For marketers, this is the most significant piece of gen AI governance yet.
The risk of isn’t just regulatory, it’s reputational: consumers, retailers and journalists are increasingly primed to react badly to brands that appear to “hide the machine”. The regulations were designed to protect consumers from deepfakes. Not to stifle creativity. Therefore the ‘stamp’ you now put on your content tells the consumer more than ‘it’s AI’ – it tells them they aren’t being misled.
This is a chance for brands to show their transparency, and creativity, in the use of highly crafted AI content. And stand out in doing so.
But the reverse is just as true. Consumers rarely object to responsible AI use: they object to feeling misled – but over-disclosure can be as damaging as under-disclosure. A practical challenge for CPG brands is deciding what counts as a deepfake. If every minor use of an AI tool triggers a clunky label and painful process, teams will either ignore the policy or flood consumers with meaningless notices. So where possible, define a clear policy and limit grey areas.
The challenge comes before the consumer sees your work. AI is already inside the creative process, but many have used it without a paper trail, or knowledge that a third party was using AI tools in delivery of the project.
Most will think about AI disclosure in the context of a big campaign asset (a hero film, a product visual). But the more common reality is fragmented AI use across agencies, studios, freelancers and martech vendors: retouched pack shots “finished” with genAI, synthetic backgrounds for e-commerce PDPs, auto-generated CRM copy variations, translated product descriptions via LLMs, and chatbot scripts rewritten by AI.
This blind spot is the problem: marketing teams often cannot answer basic questions quickly – what was AI-generated, by which tool, approved by whom, and where it was published.
Some brands anticipated the need for governance and have been tagging AI generated assets at the point of creation. They are already one step ahead.
Brands need to think about introducing an AI content register, making it as routine and rigorous as usage rights tracking, and enforce it with third-party suppliers too.
That covers content, but Article 50 goes further. It also covers customer interactions where people think they are interacting with a human. That makes customer service chat, WhatsApp commerce, live chat on brand sites, and in-app assistants a priority. In fmcg, these touchpoints often sit outside “marketing” – owned by CX, e-commerce or a contact centre vendor.
Invariably these come to the forefront during high-stress scenarios: recalls, shortages, promotions, live AI translation, and complaint handling, where tone and accuracy are critical. Here, brands may consider a standardised approach to catch all, rather than a case-by-case basis. A simple “you’re chatting to an AI assistant” with the option to easily escalate to a human, and guardrails that stop bots from improvising on regulated topics such as health, nutrition claims, babycare, alcohol and environmental claims.
“Undisclosed AI” is the next trust trigger. The fastest route to negative coverage isn’t using AI – it’s getting caught appearing to conceal it. For CPG brands built on familiarity and trust, transparency is now part of responsible marketing hygiene. The winners won’t be those who avoid AI: they’ll be the brands that can prove what they did, explain it simply, and keep a human accountable for the final message.
Angela Tangas is global CEO at Oliver
No comments yet