Short answer: Since 2 August 2026, EU rules require businesses to disclose AI-generated or AI-altered images, video and audio that resemble existing people, places or events and could pass as authentic (Unite.AI, 2026). Some AI-generated text on matters of public interest is covered too. Most everyday brand content falls outside these rules, but every brand using AI tools should have a written policy, because audiences already check for AI and trust it less. This guide is general information, so check specific cases with a lawyer.
What does the EU AI Act require from brands?
The rules sit in Article 50 of the AI Act. For brands publishing content, the main points are:
- Deepfake-style content needs a label. If you publish AI-generated or AI-edited images, audio or video that resemble existing people, places or events and would falsely appear authentic, you have to disclose it.
- Text is covered in narrower cases. AI-generated text needs a label when it’s published to inform the public on matters of public interest. There are exceptions, for example when a person has reviewed it and someone holds editorial responsibility. The European Commission published draft guidelines on this on 8 May 2026 (Lausen, 2026).
- The label has to be clear at first sight. For images, that means a caption or an icon on the image itself. For social media thumbnails, Lausen recommends placing the label on the image so it stays visible when the post is shared. For audio, a spoken notice at the start.
- The EU has published free icons. There’s a basic AI mark, a variant for fully AI-generated content and one for photos partly altered with AI (European Commission).
- Older content is exempt. Content created before 2 August 2026 doesn’t have to be labelled retroactively, though the Commission encourages it where possible.
The companies that make AI tools have separate obligations to mark their output in a machine-readable way. As a brand, your responsibility is what you publish.
Which social media content needs a label?
Likely needs a label
- A realistic AI-generated person presenting or using your product.
- A video where an employee’s face or voice has been changed or generated with AI.
- A photo of your shop, venue or product edited with AI to show something that didn’t happen.
- AI-generated customer clips or testimonials, which damage trust with or without the law.
Usually lower risk
- Using AI to brainstorm ideas or draft captions that a person then edits and approves.
- Standard editing help such as colour correction, noise removal or automatic subtitles.
- Clearly illustrated or stylised visuals that nobody would mistake for a photo.
Check carefully
- AI-written posts on topics such as health, safety, finance or public affairs. Fanpage Karma advises editorial review and keeping an internal record of how AI was used for anything that could influence public opinion (Fanpage Karma, 2026).
Instagram and TikTok apply their own AI labels to some content. Treat those as a bonus and apply your own label wherever the rules call for one.
Why is labelling also a brand decision?
Audiences are already looking. CNET’s research found that 60% of people look for clear signs of AI in images and videos, and 30% look for labels or disclosures. Trust is low once they find it: CivicScience data shows only 22% of people view AI in ads positively, and only 12% are more likely to buy from a brand when they know AI was used (eMarketer).
Organic social media runs on trust in the people and places you show. A brand caught passing off AI content as authentic pays for it in comments and lost shares, and shares are one of the signals that decide reach. A clear label costs very little in comparison.
How do you set up an AI policy for social media?
- Map where AI is used. List every tool used in planning, writing, design, editing and publishing, and what each one produces. Lausen recommends starting with this inventory.
- Set labelling rules per content type. Decide which formats always get a label, which never need one and who decides the grey areas. Use the EU icons so labels look consistent.
- Keep a person responsible for every post. The EU’s code of practice recommends that AI-generated or AI-edited content is reviewed by a human before publication, with a person holding responsibility for what goes out (Harper Macleod, 2026).
- Brief everyone who makes content for you. Agencies, freelancers and creators should follow the same policy, so a creator’s AI edit doesn’t end up on your account unlabelled.
- Publish a short statement. A few lines on your website about how you use AI makes the policy visible to customers and partners.
Frequently asked questions
Does every AI-assisted social media post need a label?
No. The labelling duty for brands focuses on realistic AI images, video and audio that could pass as authentic, plus some AI text on matters of public interest. Captions drafted with AI and approved by a person, or standard edits like colour correction, usually don’t need one. Check the Commission’s guidelines or ask a lawyer about borderline cases.
Do posts published before 2 August 2026 need labels?
No. The rules don’t apply retroactively, although the European Commission encourages labelling older content where it’s feasible.
Where should the label go on Instagram or TikTok?
On the content itself, visible the first time someone sees it. For images, a caption or an icon on the image works, and for thumbnails the label should sit on the image so it stays visible when the post is shared.
Does the UK have the same rules?
Not yet. The UK government’s March 2026 report on copyright and AI recognised the importance of labelling AI-generated content, but there’s no clear plan yet for how it will work.
Want help setting AI rules for your social media? Tell us how your team uses AI today and we’ll reply within 48 hours with first thoughts on where we’d start.
Written by the Meridian team in Helsinki. This guide is general information and not legal advice. Last updated 9 October 2026.
