Do You Need to Label AI-Written Posts Under the EU AI Act?

Usually you don't need to label AI-written posts. Article 50 of the EU AI Act, which has applied since August 2, 2026, puts most of the work on the companies that make AI tools: they have to mark their output invisibly so it can be detected. The duty to label a post you publish is much narrower. It covers professional accounts publishing AI-generated text to inform the public on matters of public interest, and it doesn't apply when a person reviewed the text properly.

That's the short version. Below is what the law says in its own words, what the European Commission's July 2026 guidelines add, and what it means for an ordinary creator, founder or brand account. It's our reading of public documents, not legal advice. If you publish on public-interest topics at scale, talk to a lawyer.

Two duties, for two kinds of people

Article 50 splits the work between providers, the companies that build and offer AI systems, and deployers, the people and organisations that use them professionally.

Providers have to mark the output. Article 50(2) says providers of AI systems "generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated." Text is named explicitly. For text, that means an invisible statistical watermark built into the model's word choices; we explain how those work in invisible characters, watermarks and AI text.

Deployers have to label in a few cases. For text, the rule is the second part of Article 50(4):

"Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated."

The same sentence carries the exception that matters most for social media: the duty doesn't apply "where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content."

The deepfake rule in the first part of 50(4) only covers image, audio and video. Text is never a deepfake under the Act.

Does it apply to you?

Work through it in order. You can stop at the first "no".

1. Are you posting for work? Article 2(10) says the Act "does not apply to obligations of deployers who are natural persons using AI systems in the course of a purely personal non-professional activity." The Commission's guidelines say both words matter: if you earn from it regularly, run a business or freelance account, or post for your employer, it's professional. A hobby account isn't.

If you post for a company, the company is the deployer, not you personally.

2. Is the text AI-generated or AI-manipulated? Not if the AI only did standard editing. The guidelines list grammar correction, spellchecking, "minor stylistic polishing", translation and formatting as standard editing. Summaries and "paraphrasing or rewriting text that changes style, structure and meaning" go beyond it.

3. Is it published? The guidelines mean available to "an indeterminate, fairly large number of unrelated, potential readers". A public social account counts. A small private group chat or internal company network doesn't.

4. Does it inform the public on a matter of public interest? The guidelines list politics, public administration, justice, fundamental rights, public security, public health, the environment and consumer safety, plus "any economic, financial, political, scientific, or cultural development" relevant to public debate. Their own social media example is a weather institute's AI-written storm warning. They also say short texts that don't "materially communicate knowledge, opinions or facts" don't inform the public, and that advertising and product descriptions are out of scope unless they make health, safety or sustainability claims.

5. Did a person review it, and does someone hold editorial responsibility? If yes to both, no label is needed.

So a founder's AI-assisted post about their launch week is very likely out of scope: it's marketing, not public-interest information. A freelance health writer's AI-drafted thread on a new vaccine study is in scope, unless they reviewed it properly.

What "human review" actually means

This is the exception most professional accounts will rely on, and the guidelines set a real bar:

• Review is a "deliberate examination of the substance" of the text by someone with the knowledge to judge it.
• "Fact-checking the accuracy of the content is a minimum requirement."
• A spellcheck, "cursory editorial approval", an automated check or "the mere existence of an editorial policy" don't count.
• If the AI changes the text substantially after you've signed off, the review no longer covers it.
• Whoever holds editorial responsibility should have their identity and contact details publicly available.

That last point is clear for a news outlet and less clear for a solo creator. An identifiable profile with a way to reach you is probably the practical answer, but the documents don't settle it.

The practical habit is the same one we'd recommend anyway: read every AI draft as if a stranger wrote it, check every fact, number and name, and make it yours before it goes out. Our guide to writing posts with AI in your own voice has a workflow for this.

If you do need a label

Article 50(5) says it has to be "clear and distinguishable" and shown "at the latest at the time of the first interaction or exposure", and it has to meet accessibility rules. The Commission's FAQ adds that you can't rely on the tool's invisible watermark to cover it: the label has to be one people can see.

The EU's code of practice suggests putting it at the top of the text, near the headline, and offers an optional EU icon in "Fully AI-Generated" and "Partially AI-Modified" versions. On a social post, a plain line at the start such as "Written with AI, reviewed by me" is the simplest equivalent. The icon is optional; the label, when it's required, isn't.

The dates

The Omnibus didn't delay Article 50 in general. It added a four-month grace period, for providers only and for the marking duty only. Content made before August 2, 2026 doesn't need to be labelled retroactively.

The code of practice

The Commission's code of practice on transparency of AI-generated content was written by independent experts with input from industry and civil society between November 2025 and June 2026. The Commission found it adequate in July 2026, and about 190 organisations had signed by the end of that month, including OpenAI, Google, Anthropic, Meta and Microsoft.

Two details in it matter for social posts:

• Short text isn't watermarked. The code exempts text shorter than 200 tokens from watermarking. A single post on X, Bluesky or Threads is usually under that, so a short post from a signatory's model may carry no watermark at all.
• Text detectors may stay private. Providers may limit access to their text watermark detectors to vetted experts such as regulators, researchers and fact-checkers, because detection on text is less reliable.

Signing is voluntary. The legal duties in Article 50 aren't.

Outside the EU

Article 2(1) covers deployers outside the EU "where the output produced by the AI system is used in the Union". The guidelines read that as foreseeing that your content will reach the EU, and give posting on the globally accessible internet as an example. A public account with EU readers is therefore arguably in scope. How national authorities would enforce this against an individual outside the EU is an open question.

Platform rules are separate

The AI Act isn't the only rule you post under. As of October 2026:

• Meta, X and LinkedIn label AI images, video and audio, mostly from embedded C2PA or IPTC metadata. We found no rule on any of the three that requires labelling AI-written text.
• Bluesky's community guidelines don't mention AI disclosure.
• Mastodon rules are set per server. mastodon.social requires that "use of generative AI must be disclosed" and bans accounts that only post AI-generated content, and that rule isn't limited to media.

Hosting platforms aren't deployers under the Act just for carrying your posts, so the duty sits with whoever publishes the text.

What this means for AI writing tools

If you use an AI feature inside a writing or scheduling tool, the tool's maker carries the provider duties. The Commission's guidelines say a company offering an AI application "under its own name or trademark" is a provider, and a provider building on another company's model can rely on that model's watermark while staying responsible for showing it complies.

For your day-to-day posting, the split between standard editing and generation is the useful part:

• Standard editing, outside the marking duty: fix spelling and grammar, light polishing that keeps your meaning and style, translation, formatting.
• Generation, inside it: draft a post from a prompt, summarise an article, rewrite in a new tone, expand a post, turn notes into a thread in new words.

And whatever the tool marks, the labelling question above is still yours.

Where Shaflex fits

Shaflex's AI menu does both kinds of work. Fix grammar is standard editing. Improve writing, Rephrase and Fit to platform limit change your wording. Every result goes into your draft in the Composer, and nothing is scheduled until you pick a time, so there's always a point where you can check it. Whether that check counts as human review depends on what you do with it: fact-check it, don't just skim it. Shaflex doesn't add a label to your posts for you: whether one is needed depends on what you're publishing and why, which only you know.

Sources

Checked on October 20, 2026. Article wording quoted from the AI Act Explorer mirror; the official texts are on EUR-Lex.

• EUR-Lex: Regulation (EU) 2024/1689, the AI Act and Regulation (EU) 2026/1744, the Digital Omnibus on AI
• AI Act Explorer: Article 50, Article 2, Article 3, Article 99, Recital 134
• Council of the EU: Digital Omnibus on AI, final signed text (PE-CONS 30/26)
• European Commission: AI Omnibus enters into force
• European Commission: Guidelines on transparency obligations under Article 50, C(2026) 5054, and FAQ on Article 50
• European Commission: Code of practice on transparency of AI-generated content, the Commission's adequacy opinion, signatories and EU icons for labelling
• European Parliament Research Service: Digital Omnibus on AI briefing, June 2026
• Meta: Labeling AI-generated content
• LinkedIn Help: Content Credentials
• Bluesky: Community Guidelines
• mastodon.social: server rules