The EU AI Act and human review of AI-written text

• The law applies to you if you use AI to write or change text that you publish to inform the public about matters of public interest.
• If it applies to you, you must say that the text was generated or changed by AI.
• You don’t have to say the text was generated by AI if a person has reviewed it or it has been through editorial control, and a person or organization takes editorial responsibility for publishing it.

The European Commission’s guidance adds that the review has to look at what the text says, including whether the facts are right. A spelling or grammar check on its own is not enough. Read the guidance.

Yes. This rule is in force now, and it has been since August 2, 2026.

The AI Act does not define the phrase. The European Commission explains it in its guidance on Article 50. It says matters of public interest are topics that matter to society at large, whether locally, nationally, across the EU or worldwide, and that deserve public debate or scrutiny.

The guidance names politics and elections, public administration and public services, the courts and law enforcement, fundamental rights, public security, public health, the environment and consumer safety. It also names any economic, financial, political, scientific or cultural development that people may debate in public. It says what counts can change over time and from one place to another.

It gives examples the law covers: an AI summary of a news story about a town council decision, a lifestyle article that compares diets for a disease, a public company’s reports for investors, and a weather service’s post warning people about a storm.

It also gives examples the law does not cover: fantasy novels, ads and product descriptions that make no claims about health, consumer safety or sustainability, a news summary that a chatbot shows only to the person who asked for it, and a consultant’s advice to a client. Private emails and a company’s internal messages are outside the rule too, because they are not published.

Source: European Commission, guidelines on the transparency obligations under Article 50 of the AI Act, July 20, 2026, section 6.2.1. Read the guidelines.

All Proofed editors are highly trained. They read your text and the AI’s changes, and they use their own judgment. They catch what the AI missed or got wrong, and they check that the meaning is right. Then they send you a final file.

You keep the editor’s file and the date they finished it. That is your record that a person reviewed the text before you published it.

So before you publish, a person has reviewed your text, not only a machine. You still decide what to publish, and you or your organization keep editorial responsibility for it.

Turn on Human review when you upload. See how human review works.

You run a news site and use AI to draft stories. News is about matters of public interest, so the law applies to you. You can label each story as AI-written. Or you can have a person review it before it goes out, and you or your company take editorial responsibility for it.

Your company’s shares trade on a stock exchange, and you use AI to help write the reports you post for investors. The Commission’s guidance uses this very case as an example the law covers. Have a person review each report before you post it, or say that AI helped write it.

You write health articles for a lifestyle blog and use AI to tidy them up. Health is a matter of public interest, even on a lifestyle site. The guidance gives an article comparing diets for a disease as an example.

You use AI to write the product descriptions in your online store. The guidance says product descriptions are usually outside this rule. That changes if they make claims about health, safety or sustainability.