Tools

Do I have to label this?

Six questions on chatbots, images, texts and emotion recognition. The result classifies under Art. 50 of the AI Act — with a source under every statement. No sign-up, nothing is stored.

The problem

Do I have to say that a text or image comes from an AI?

A chatbot on the website, a product photo from the image generator, a newsletter text from the assistant: Article 50 of the AI Act governs when deployers disclose or label. The rule distinguishes by type of content, by audience and by whether someone has reviewed and taken responsibility for the content.

What the check does: The check asks six questions about chatbot, images, texts and emotion recognition and classifies under Article 50, every statement with reference and source status.

What it does not do: It names no authority and no penalty for Austria, because neither is officially documented as of the source status, and it decides no individual case.

Start the check · 3 min · free

Does this also apply to internal texts?

The check asks about it: what stays in the business and is not published, it places differently from what goes out.

What about images that were only edited?

Generated or altered, the check asks about both; the placement names the place in Article 50 that belongs to it.

How current is the legal status?

The source status is on the result page with a date: wording read against EUR-Lex on 18 September 2026, guidelines from an extract.

Question 1 of 6How do you use AI tools?Art. 50 distinguishes providers (who develop an AI system or offer it under their own name) and deployers (who use a third-party system professionally).

In the business, with third-party toolsThen you are a deployer. Paragraphs 3 and 4 of Art. 50 apply to you.

Source: Art. 3(4) (deployer). Employees are not deployers in their own right; the business is (guidelines para. 12).

Only privatelyThe Regulation does not apply to purely private use.

Source: Art. 2(10) AI Act; guidelines para. 18.

We developed an AI system ourselves or offer one under our own nameThen you may be a provider; paragraphs 1 and 2 apply in addition, which this check does not examine.

Source: Art. 3(3) (provider); guidelines paras. 10 and 11. Whoever substantially modifies a third-party system may become a provider (Art. 25, not examined here).

Question 2 of 6Does an AI system (chat, voice, avatar) respond directly to customers, prospects or other persons?This means a system that replies without a person in between — such as a chatbot on the website or a voice assistant on the phone.

Yes, directlyThe person must be clearly told, at the latest at the first interaction, that they are interacting with AI.

Source: Art. 50(1) and (5); guidelines paras. 34 to 40. The notice must not appear only in terms and conditions or a footer (para. 38); the “obvious” exception is narrow (paras. 42 to 45).

No, a person reviews and sends every replyThen there is no direct interaction; paragraph 1 does not apply.

Source: guidelines para. 30 (“directly”).

No, we use AI only internallyNo obligation under paragraph 1 towards customers; internally the “obvious” exception may apply.

Source: guidelines para. 45 (trained, AI-literate staff).

Question 3 of 6Do you generate or alter images, audio or video with AI that you publish?Your own avatar or your own voice in an explainer video count as well.

NoNo obligation under paragraph 4, first subparagraph.

Yes, but only graphics, illustration or fantasy that nobody would take for realNot a deepfake within the meaning of the Regulation; no obligation under paragraph 4, first subparagraph.

Source: Art. 3(60) (deepfake: “existing” persons, objects, places, events); guidelines para. 113.

Yes, with realistic persons, places, objects or eventsThat is a deepfake: disclose that the content has been artificially generated or manipulated.

Source: Art. 50(4) first subparagraph and (5); recital 134; guidelines paras. 113, 117, 143. In advertising the artistic exception usually does not apply (para. 122).

Yes, a real product in front of an AI backgroundUsually not a deepfake, as long as the product is not shown differently from how it is — case by case.

Source: guidelines para. 116.

Question 4 of 6Do you publish AI-generated or AI-altered text publicly?Website, blog, social media or a newsletter to an open list. Internal texts and individual correspondence are not “published”.

No, only internally or in individual correspondenceNot “published”; no obligation under paragraph 4, second subparagraph.

Source: guidelines para. 131 (i).

Yes, but only advertising, offers and product descriptions without statements on health, safety or sustainabilityNot covered according to the guidelines; mixed texts remain uncertain.

Source: guidelines para. 131 (i) to (iii).

Yes, on topics such as law, health, safety, environment, politics, administration or the economyThen the next question decides: review and editorial responsibility.

Source: Art. 50(4) second subparagraph; guidelines para. 131 (iii).

Question 5 of 6Is this text reviewed for content by a competent person before publication, and does a named person or your company publicly hold editorial responsibility?Only relevant if you chose topics of public interest in the previous question. A spell check alone is not a review of content; an AI change after the review removes the exception.

Both yes, and the AI changes nothing after the reviewThen the exception applies; name the responsible party with contact details publicly, for instance in the legal notice.

Source: Art. 50(4) second subparagraph, second sentence; guidelines paras. 133 to 138.

One of these is missingThen you must disclose that the text has been artificially generated or manipulated.

Source: Art. 50(4) second subparagraph; guidelines paras. 135 and 136.

Does not apply to us (no such text)Then this question changes nothing.

Question 6 of 6Do you use systems that recognise emotions or assign persons to categories on the basis of biometric data?For example a camera that detects the mood in the salesroom.

NoNo obligation under paragraph 3.

YesThe persons concerned must be informed about the operation of the system; at the workplace, prohibitions apply that this check does not examine.

Source: Art. 50(3); guidelines paras. 100 to 110. Prohibitions: Art. 5(1)(f) AI Act; plus the GDPR — both outside this check.

The tool classifies, it does not decide. Whether labelling is required in a given case depends on the content, the audience and questions only an individual review can settle. It is not legal advice. Source status: wording retrieved 14 September 2026 and read against EUR-Lex on 18 September 2026 (research/ki-vo/01, addendum); the guidelines from an extract. Status to be verified.