AI Content Labelling Rules: When You Must Disclose the Machine

The short answer: label anything that could mislead

If a reasonable viewer could believe your AI-generated image, video or voice clip shows something real that never happened, you need a label. That is the common thread running through Meta's, YouTube's and TikTok's policies, the EU AI Act's transparency articles, and long-standing UK advertising law. Everything else is nuance, but it is nuance worth knowing, because the penalties range from reduced reach and removed posts to formal ASA rulings and, for firms trading into the EU, regulatory fines.

For UK small businesses the confusion usually comes from mixing up three separate layers: platform rules (contractual, enforced by the platform), advertising rules (the CAP Code, enforced by the ASA), and statutory rules (the EU AI Act if you sell into the EU, plus UK consumer protection law everywhere). A piece of content can be fine under one layer and non-compliant under another, so it pays to check all three.

What the major platforms require in 2026

Each big platform now has an AI disclosure mechanism, and they are not interchangeable. The broad positions:

• Meta (Facebook and Instagram): applies "AI info" labels to content it detects as AI-generated or edited, and expects advertisers to disclose digitally created or altered media in ads about social issues, elections and politics. Photorealistic AI imagery in organic posts may be auto-labelled whether you like it or not.
• YouTube: requires creators to tick a disclosure in Creator Studio when realistic content is meaningfully synthetic, including altered footage of real events, synthetic voices of real people and realistic scenes that never occurred. Obvious animation and beauty filters are exempt.
• TikTok: requires an AI-generated content label for realistic synthetic media and auto-labels content carrying C2PA Content Credentials metadata. Unlabelled realistic AI content risks removal.
• LinkedIn and X: lighter-touch, but both prohibit deceptive synthetic media, and LinkedIn surfaces Content Credentials where present.
• Google Ads: demands transparency for synthetic content in political ads and penalises misleading imagery in Shopping and Performance Max, such as AI product photos that misrepresent the actual item.

Note the pattern: the trigger is realism, not the mere use of AI. A cartoon mascot generated in Midjourney rarely needs a label. A photorealistic image of your product in a customer's home that was never photographed usually does.

UK rules: the ASA, the CMA and consumer law

The UK has not passed a dedicated AI labelling statute, but do not read that as a free pass. The existing framework already catches most problem cases.

The CAP Code requires that marketing communications must not materially mislead. The ASA has made clear this applies regardless of how content was produced: an AI-generated "customer photo", a synthetic testimonial voice, or an AI-enhanced before-and-after image can all breach the Code if they exaggerate what your product or service actually does. AI does not create a new offence; it creates new ways to commit an old one.

The Digital Markets, Competition and Consumers Act gives the CMA power to fine businesses directly for unfair commercial practices, including fake reviews, without going to court first. AI-written reviews presented as genuine customer feedback sit squarely in that danger zone. If you use AI to draft review responses or summaries, fine; if you use it to manufacture reviews, that is a serious legal risk, not a labelling question.

For text content generally, such as blog posts and product descriptions, there is no UK obligation to declare AI assistance, provided the content is accurate and not deceptive about authorship where authorship matters (for example, claiming a named expert wrote something they never reviewed).

Selling into the EU? The AI Act's transparency duties apply

If your business markets to EU customers, the EU AI Act's transparency obligations are relevant even though you are UK-based, because the Act applies to outputs used in the EU. The key duties for marketers sit in the Act's transparency provisions:

• Deepfakes: AI-generated or manipulated image, audio or video content that resembles real people, places or events must be disclosed as artificially generated or manipulated.
• AI-generated text published to inform the public on matters of public interest must be disclosed, unless it has undergone human editorial review with a person taking responsibility.
• Chatbots: people must be told they are interacting with an AI system where it is not obvious from context.
• Machine-readable marking: providers of generative AI systems must ensure outputs are marked as AI-generated, which is why C2PA Content Credentials increasingly travel with files from tools like Adobe Firefly and DALL·E.

The practical consequence: strip-and-repost workflows that remove provenance metadata are becoming both technically harder and legally riskier. Keep the Content Credentials attached where your tools embed them.

A simple labelling decision table

Run each asset through these questions in order. The first "yes" tells you what to do.

• Does it depict a real person saying or doing something they did not? Label it prominently, get consent, or better still do not publish it. Highest-risk category everywhere.
• Is it a photorealistic image or video of an event, place or product scene that never existed? Label it (platform AI toggle plus a caption note like "Image created with AI").
• Is it a synthetic voice-over or AI presenter in video content? Disclose on YouTube and TikTok via their tools; add a caption credit elsewhere.
• Is it an AI chatbot talking to customers? Identify it as AI at the start of the conversation.
• Is it AI-assisted text (blogs, emails, product copy) that a human has reviewed and stands behind? No label required in the UK; ensure accuracy.
• Is it obviously stylised or artistic (illustration, cartoon, abstract graphics)? No label required, though voluntary credits cost nothing.
• Are AI product images used in ecommerce listings? Only safe if they accurately represent the physical item; Amazon and Google both penalise misrepresentation.

Build disclosure into your workflow, not your apologies

The businesses that get caught out are rarely acting in bad faith; they simply have no process. A lightweight one looks like this:

• 1. Keep an asset register noting which published images, videos and audio were AI-generated or AI-edited, and with which tool.
• 2. Preserve Content Credentials metadata rather than flattening files through screenshot-and-crop.
• 3. Add the platform's native AI disclosure at upload time; retrofitting labels after a complaint looks far worse.
• 4. Give whoever signs off ads a one-page checklist covering the decision table above.
• 5. Review quarterly: platform policies moved several times between 2024 and 2026 and will keep moving.

Honest labelling costs you very little. Most audiences now expect some AI in marketing content, and a small "created with AI" note is far less damaging than a removed ad, an ASA ruling or a viral callout. If you want a second pair of eyes on your content workflow and disclosure policy, our team can help you set one up.