You’ve probably seen the headlines. The EU AI Act is here, it’s long, and everyone on LinkedIn seems to have an opinion about it. If you run a blog, a store or a company site, here’s the good news: most of the Act isn’t about you. It’s about high-risk stuff like AI in hiring, credit scoring or medical devices.
The part that does reach ordinary websites is short. It’s Article 50, and it boils down to one idea: people should know when they’re dealing with AI.
We read the Act, the European Commission’s guidelines on Article 50 (published November 2026) and the voluntary Code of Practice on AI-generated content, so you don’t have to. None of them mention WordPress, so we’ve translated them into the situations site owners actually run into.
A tool, not legal advice. This guide explains public documents in plain language. It can’t tell you whether your business complies with the law. If that question matters to you, talk to a lawyer.
Article 50 in two minutes
Article 50 has applied since 2 August 2026. It contains four separate duties:
- 50(1), chatbots and other interactive AI. People have to be told they’re talking to an AI, unless that’s obvious.
- 50(2), machine-readable marking. AI-generated images, audio, video and text need a technical marker, like metadata or a watermark, that software can pick up.
- 50(3), emotion recognition and biometric categorization. Anyone exposed to these systems has to be informed. You almost certainly don’t run one.
- 50(4), deepfakes and some published text. Realistic AI images, audio or video (deepfakes) have to be disclosed. So does AI text published to inform the public on matters of public interest, unless a human editor reviewed it and takes responsibility.
The law splits the work between two roles. The provider builds the AI system and puts it on the market: the company behind the image generator or the chat widget. The deployer uses that system for work: say, a shop owner who posts AI product shots.
Most website owners are deployers. Keep that in mind, because the two roles don’t get the same homework.
Which parts land on your desk
Chatbots: mostly the vendor’s job
Article 50(1) is aimed at providers. If you drop a vendor’s chat widget onto your site, the vendor is usually the provider and has to make sure visitors know they’re chatting with AI. There’s a catch, though. If you build your own bot, or run an AI system under your own name, the Commission says you can become the provider yourself.
Even when it’s technically the vendor’s duty, it’s still your site. Vendor notices can be tiny, buried in a menu or switched off in settings. And the guidelines are blunt: a line in your terms and conditions doesn’t cut it on its own, and something vague like “Services on this website use AI” is too generic. More on that in Chatbots on your website.
Machine-readable marking: the tools handle it
Article 50(2) is another provider duty. The image generator or writing tool is supposed to embed the marker. Your job is simpler: don’t strip it out. Some image optimization plugins wipe metadata on upload, so it’s worth a quick look at your settings.
Deepfakes: this one’s yours
Publish a realistic AI image, video or audio clip of a real-looking person, product, place or event that could pass as genuine, and the disclosure is on you. This is where shops and publishers need to pay attention.
That said, not every AI image is a deepfake. A cartoon, an obvious fantasy scene, or a real product on an AI-generated background usually isn’t. A product photo “improved” so the item looks better than it really is can be. See Do you need to label AI images?.
Public-interest text: yours too, with an escape hatch
If you publish AI-written text that informs people about things like health, politics, the environment, consumer safety, or economic and scientific developments, you need to say so. Unless a real person reviewed the substance, facts included, and someone named takes editorial responsibility. Running spell-check doesn’t count. Most marketing copy and everyday product descriptions are outside this rule anyway. See AI-written blog posts and articles.
Personal projects are off the hook
Deployer duties don’t apply to purely personal, non-professional use. A family travel blog with no ads? Probably out of scope. Once a site earns money on a regular basis or is run for a business, the guidelines treat it as professional.
When and where to tell people
The Act wants the disclosure to be clear and easy to spot, no later than the first time someone sees the content or starts the interaction. In real life that looks like this:
- Images and video: on the media itself or right next to it. The Code of Practice uses the top-right corner of an image as its example.
- Text: at the start, above the text or near the headline. A note at the bottom of a 2,000-word post comes way too late.
- Chatbots: before or at the very start of the chat, close to the chat window. A first message like “You are chatting with an AI assistant” plus a small badge near the input field are both cited as good practice.
What about stuff you published before August 2026?
According to the guidelines, images, audio, video and deepfakes generated before 2 August 2026 don’t need labels after the fact. Same goes for in-scope text that was generated and published before that date.
Here’s the gotcha: text you generated before 2 August 2026 but publish on or after that date needs a label like anything new.
You may also hear about 2 December 2026. A separate amendment gives providers of generative AI tools that already existed before August until that date to add machine-readable marking. That extra time covers marking only. It doesn’t push back chatbot notices or the deepfake and text disclosures.
What happens if you ignore it
Fines for breaking Article 50 go up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. For small and medium-sized businesses, it’s whichever is lower. Authorities are also told to weigh things like the size of the business, how serious the breach was and how cooperative you’ve been. Bottom line: a clear, good-faith labeling setup puts you in a much better spot than doing nothing.
Your first hour
You don’t need to slap a label on everything. Start by figuring out what actually applies:
- List every place AI touches your site. Chat widgets, AI writing plugins, image generators, AI product descriptions, machine-translated pages, avatar videos.
- Match each one to a duty. Chatbot (50(1))? Realistic media (deepfake under 50(4))? Public-interest text (text under 50(4))?
- Decide who reviews text. If someone on your team fact-checks AI drafts before they go live, write that down and name the person responsible.
- Add visible labels where they’re needed. On the image, at the top of the text, at the start of the chat.
- Explain it in one place. A short “How we use AI” page helps visitors and shows good faith.
- Keep a record of what you labeled and when.
Where AuthorLabel fits in
AuthorLabel is a free WordPress plugin built around exactly this order. Activate it and a first-run check asks six quick questions and scans your plugins and scripts for known chat widgets. You get each Article 50 duty marked as applies, does not apply or your call, with a one-line reason.
It’s also smart about what’s actually live. If Tidio is installed but not connected to an account, the check tells you visitors can’t see it yet, instead of nagging you to label a chatbot nobody can use.
From there you can turn on a chatbot notice, label posts, products and media right in the editor, generate an editable “How we use AI” page, and see everything you’ve done in an action log. The plugin makes no external requests and never touches your post content.
Sources
- Regulation (EU) 2024/1689 (AI Act), Article 50 and Article 113.
- European Commission, Guidelines on the transparency obligations for providers and deployers of AI systems under Article 50 of the AI Act, C(2026) 7702 final. digital-strategy.ec.europa.eu
- Code of Practice on Transparency of AI-Generated Content. digital-strategy.ec.europa.eu
The Commission’s guidelines aren’t legally binding. Only the Court of Justice of the EU can give the final word on how the AI Act is read.
Last reviewed: November 2026. AuthorLabel is a tool, not legal advice.