Do You Have to Label AI Social Media Posts? The EU AI Act for Small Businesses
If you used ChatGPT to tidy up last week’s LinkedIn post, or asked an image tool for a picture to go with it, you have probably seen a headline telling you that AI-generated content must now be labelled in the EU. The rule it refers to is real. Article 50 of the EU AI Act has applied since 2 August 2026, and it does deal with labelling AI generated content.
What the headlines leave out is how narrow it is. For most of what a small business posts on social media, the AI Act does not require a label. There are three situations where it does matter, and one simple house rule that keeps you on the right side of all of them. This article walks through both, in plain language, using the European Commission’s own explanation rather than the press summary of it.
What changed on 2 August 2026
The AI Act is being switched on in stages. The part about transparency, Article 50, started to apply on 2 August 2026. It puts duties on two kinds of people: the companies that build AI tools (the Act calls them providers) and the people who use those tools in their work (deployers). If your business uses an AI tool to make content, you are a deployer.
Most of the heavy lifting sits with the tool makers. They have to make sure what their tools produce is marked in a machine-readable way, so that it can be detected as AI-generated. Tools that were already on the market before 2 August have until 2 December 2026 to do that. None of this is your job as a user. It is the reason some platforms may start labelling your AI images on their own, which we come back to below.
Your duties as a deployer are narrower, and they are about two things: realistic fakes, and text written to inform the public about matters of public interest.
The three cases that actually matter
1. Images, audio or video that could pass for real
The Act uses the word deepfake, which makes it sound like something only politicians and celebrities need to worry about. The legal definition is broader than that. It covers AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic.
Read that list again with your own feed in mind. Objects and places are on it, not only people. A photorealistic AI picture of “our team at work” that never happened, a generated image of your shop that looks like a photo of your shop, a product shown in a real-looking setting it has never been in, or your own voice cloned for a video you did not record: each of those could fall under the definition. If you publish one, the Commission says you have to disclose it clearly, at the latest when someone first sees it.
There is a lighter rule for work that is evidently artistic, creative, satirical or fictional. You still disclose, but in a way that does not spoil the piece. An obviously illustrated cartoon mascot is a different thing from a picture pretending to be a photograph.
2. Text published to inform the public on matters of public interest
The second duty covers AI-generated text published to inform the public on matters of public interest. The Commission’s examples are topics such as politics, public health and environmental protection. A caption announcing your autumn opening hours is not what this is about.
The important part is the exception. Text that has been through human review or editorial control, where a person or company takes editorial responsibility for publishing it, does not need a label. The Commission describes human review as a deliberate check by someone with the relevant knowledge. Glancing at a draft and pressing publish is not that.
3. Tools that talk to your customers
If you put an AI chatbot in your website chat or your Messenger inbox, the transparency rules also expect people to be told they are talking to a machine, unless it is already obvious. That duty is designed into the tool by its provider, but it is your inbox and your customers. Check that the tool you use does it.
What that means for a normal week of posts
Put together, here is how the rules land on the kind of content a small business actually publishes:
- A caption you drafted with ChatGPT, then rewrote and approved yourself: no legal label needed. You reviewed it and you are responsible for it.
- A clearly illustrated or stylised image that nobody would mistake for a photograph: generally not a deepfake. Use your judgement, and disclose if in doubt.
- A photorealistic image of people, a place or an event that never happened, presented as real: this is the one to label, every time.
- A video with an AI-cloned voice or an AI presenter who looks like a real person: label it.
- A post giving public health, safety or political information, written with AI and published without a proper review: either review it properly or label it. Better still, do the first.
The platform’s label is a different thing
Separately from the law, the big platforms have their own policies on AI content, and some of them add a label automatically when they detect that an image or video was made with AI. That detection often relies on the machine-readable marking the tool makers now have to add. So you may find a label appearing on a post you did not label yourself.
That is the platform applying its own rules, not a finding that you broke the law. The practical consequence is simpler: assume that a realistic AI image will eventually be recognised as one. If your post only works as long as nobody notices, it was a bad post before the AI Act existed.
The penalties, kept in proportion
The Commission’s FAQ puts the ceiling for breaching these transparency rules at up to €15 million or 3% of worldwide annual turnover. For small and medium-sized businesses, the Act applies whichever of the two is lower, and fines are meant to be proportionate.
For most small businesses, though, the fine is not the risk worth planning around. The more realistic one already existed before the AI Act: a customer spotting that the “photo” of your team or your workshop was generated, and wondering what else on your page is not real. For a small business that trades on being trusted locally, that costs more than most fines would.
A house rule you can adopt this week
You do not need a policy document. Four lines, agreed with whoever makes your content, cover almost everything above:
- A named person reads every AI-assisted post before it goes out, and is allowed to change it. That review is what keeps ordinary text outside the labelling rules, and it is what keeps your feed sounding like you.
- No photorealistic AI images of people, places or events presented as real. Use real photos of your real business, or use illustrations that are obviously illustrations.
- If you ever do publish a realistic AI image or voice, say so in the post itself, in plain words: “Image generated with AI.”
- Anyone making content for you tells you which posts used AI for images, audio or video. You should not have to guess.
The first rule is the one that matters most, and it is also the one that most improves the content. We made a similar argument in an earlier piece about what AI tools genuinely do well for a small business and where they fall short.
The AI hype in marketing: what actually helps a small business
If someone else makes your content
If you work with an agency, a freelancer or an outsourced team, these are fair questions to put to them, and to us: which tools do you use for images and video, who reviews the text before I see it, and will you tell me when something is AI-generated? A provider who cannot answer those three in a sentence each has not thought about it yet.
With us, every piece of content sits in a shared calendar for you to review and approve before it is published, and nothing goes out unseen. That approval step was built for quality reasons, not for the AI Act, but it is exactly the kind of human review the rules have in mind. If you want to know how to make that review quick rather than a chore, we wrote about batch approval and the other habits that make outsourcing work.
5 effective ways to work with an outsourced social media manager
The short version
For most small business posts, the EU AI Act does not require a label. Label realistic AI images, audio and video of people, places, objects or events that could pass for real. Review AI-written text properly, especially anything on a public interest topic, and it needs no label. Keep a real person in the loop, and the rules take care of themselves.
The two official documents behind this article are short and readable, and worth a look if you want the source rather than our summary:
European Commission: FAQ on the transparency obligations under Article 50 of the AI Act
European Commission: Code of Practice on marking and labelling AI-generated content
And if you would rather hand the whole question to a team that reviews every post before you see it, start small. Our €1 a day offer (€365 a year for one post a day on one platform) lets you see how we work before you commit to anything bigger.
Comments
Loading comments…