Plenty of us use AI to draft posts, summarize sources or freshen up old articles. And a lot of people now worry that each of those posts needs a big “Written by AI” banner on top. Relax. Under the EU AI Act, the rule for text is narrow, and there’s a built-in exception for content a human has properly checked.
Here’s which text the rule actually covers, how the human-review exception works day to day, and how to label a post when you have to.
A tool, not legal advice. This guide explains public documents in plain language. Whether a particular post is in scope depends on its topic and how it was made. If it matters for your business, ask a lawyer.
The rule, in one sentence
Article 50(4) says that if you use AI to generate or edit text that’s published to inform the public on matters of public interest, you have to disclose it. Unless a human reviewed it or it went through editorial control, and a person or organization takes editorial responsibility.
That’s three conditions. All three have to be true before you need a label.
1. It’s published
The text has to be out there for the public, like on a web page. The Commission gives a handy contrast: a news summary a chatbot writes just for the person who asked isn’t “published”. Post that same summary on your blog, and it is.
2. It’s meant to inform people
The text should be trying to share knowledge, opinions or facts. Short bits that don’t really do that, like a button label or a one-line slogan, don’t count. An explainer or a news-style post does.
3. The topic is a matter of public interest
This is the condition that keeps most business content out. The guidelines describe public-interest topics as things that matter to society at large and deserve public debate or scrutiny. Their list:
- politics and elections
- public administration and services
- justice and law enforcement
- fundamental rights
- public security
- public health
- the environment
- consumer safety
- economic, financial, political, scientific or cultural developments that people might debate
They also point out that what counts can shift over time and from one context to another.
Straight from the guidelines
In scope (label it, unless it’s been reviewed by a person):
- An AI summary of a human-written news story about a town council decision.
- AI-edited parts of a lifestyle article comparing how different diets affect a disease in middle-aged women.
- AI-edited investor info in a listed company’s reports.
- An AI-written storm warning posted by a weather service.
Out of scope:
- AI-written fantasy novels.
- AI-edited ads or product descriptions, unless they make health, consumer-safety or sustainability claims.
- A news summary a chatbot shows only to the person who asked for it.
- AI-edited advice a consultant sends a client.
Pay attention to that diet example. It’s from a lifestyle site, not a newspaper. A wellness blog that talks about medical conditions can be in scope even if the owner never thought of it as journalism.
What this looks like on a normal blog
| Your post | In scope? |
|---|---|
| “10 ways to style a small balcony” | Unlikely. Lifestyle tips. |
| “How the new EU battery rules affect e-bike owners” | Likely. Public administration, consumer safety. |
| “Our spring collection is here” | No. That’s advertising. |
| “Does intermittent fasting help with type 2 diabetes?” | Likely. Public health. |
| “Weekly market recap: interest rates and your mortgage” | Likely. Economic developments. |
| “Recipe: lemon pasta in 15 minutes” | Unlikely. |
These are judgment calls, not hard rules. If a post sits on the fence, a label costs you next to nothing.
The human-review exception
Even if a post is in scope, you can skip the public label when both of these hold.
Someone actually reviewed it
The guidelines describe human review as a person with the right knowledge going through the substance of the text on purpose. Fact-checking is the bare minimum. Editorial control means an editor who can approve, change or reject the content on substance.
They’re also clear about what doesn’t count:
- spell-check or grammar fixes
- just having an editorial policy somewhere
- automated review
- a quick “looks good” without really engaging
And watch out for this one: if you run the text through AI again and change it substantively after the review, the exception’s gone. The version you publish has to be the version that was reviewed. Let AI “punch up” the intro after sign-off, and you’re back to needing a label.
Someone’s name is on it
A person or organization has to hold ultimate responsibility for publishing the content. The guidelines add that, for accountability, the name and contact details of whoever holds that responsibility should be easy to find on your site, say on your About page or in your legal info.
So what does that mean for you?
If you run a small blog and personally fact-check every AI draft before it goes live, keep a simple record of that and make sure your contact details are on the site. If you’re publishing AI drafts after a skim and a spell-check, label them. Simple as that.
Where the label goes
The Act wants the disclosure in place no later than the moment someone starts reading. For text, the Code of Practice suggests putting it above the text, near the headline or at the very beginning. A note at the bottom of a long post doesn’t help anyone. If AI only wrote part of a post, the Code says you can label just that part.
Something plain works best:
This article was written with the help of AI and has not been reviewed by an editor.
Older posts
The guidelines say in-scope text that was AI-generated and published before 2 August 2026 doesn’t need a label after the fact. But if you wrote a draft with AI before that date and hit Publish on or after it, it needs a label like any new post.
What AuthorLabel does (and what it won’t do)
AuthorLabel adds an “AI transparency” panel to both the block editor and the classic editor. For each post, you pick one of four statuses:
- AI-generated: shows a notice at the top of the post: “This text was generated with AI and has not been reviewed by an editor.”
- AI-assisted: shows a notice at the top of the post: “This text was written with help from AI.” You can add which parts used AI, for example “summary and FAQ”.
- Reviewed by a human editor: you name the editor and AuthorLabel records the review. No public label, since the exception may apply.
- Not AI: no label.
You can also note which AI tool you used. It shows up on your generated “How we use AI” page, so visitors can see what you rely on.
Two things AuthorLabel deliberately won’t do:
- Guess from writing style. “AI detectors” that judge text by how it reads are unreliable and throw false positives all the time. AuthorLabel only records AI use when you mark it, or when a supported AI writing tool reports it (automatic labeling for several popular AI plugins comes with AuthorLabel Pro).
- Touch your post content. The label is added when the page is displayed. Switch the plugin off and your text is exactly as you wrote it.
For older posts, bulk labeling lets you filter by post type, date, author or tool and set a status for up to 200 posts in one go. Every bulk action can be undone, and the action log shows who labeled what and when.
The short version
- Only text published to inform people on matters of public interest is in scope.
- Ads and ordinary product descriptions are out, unless they make health, safety or sustainability claims.
- Real human review with fact-checking, plus a named responsible person, means no public label.
- Spell-check isn’t review, and AI edits after sign-off cancel it.
- Put the label at the top.
Sources
- Regulation (EU) 2024/1689 (AI Act), Article 50(4) and (5).
- European Commission, Guidelines on the transparency obligations under Article 50 of the AI Act, C(2026) 7702 final, Sections 6.2 and 8.4. digital-strategy.ec.europa.eu
- Code of Practice on Transparency of AI-Generated Content, deployer section, Sub-measure 1.2.2. digital-strategy.ec.europa.eu
The Commission’s guidelines aren’t legally binding, and the Code of Practice is voluntary.
Last reviewed: November 2026. AuthorLabel is a tool, not legal advice.