Do you have to label AI voiceovers? The 2026 rules

Almost every rule in this article turns on one idea: does the content pretend to be something real that it is not. The EU AI Act writes it down precisely. Article 3(60) defines a deep fake as “AI generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”.

Read that slowly and a normal text to speech voiceover falls outside it. A stock AI narrator does not resemble an existing person. It reads your words, which you wrote and stand behind. Nobody listening is fooled into believing something false happened. It is closer to a font than to a forgery.

Three examples make the line clear.

• A faceless history channel narrated by an AI voice. Not a deep fake. No real person is imitated and nothing false is presented as real.
• Your own voice, cloned, reading your own script. Still not a deceptive deep fake, and YouTube lists cloning your own voice for voiceovers or dubs as needing no disclosure.
• A cloned voice of a real politician, actor or influencer saying words they never said. That is exactly what the definition describes. It needs a label under EU law, and every platform in this article asks for one.

The grey zone is a voice built to sound very close to a specific famous person without naming them. If a reasonable listener would think it is that person, treat it as imitation and label it. Beyond the labelling rules, imitating a real voice without permission raises separate personality and consumer protection issues in many countries, which is a good reason not to do it at all.

Paragraph 2 is for the companies that make AI tools. Providers of AI systems that generate synthetic audio, image, video or text must make sure the output is marked in a machine readable format and detectable as artificially generated or manipulated, using solutions that are effective, interoperable, robust and reliable as far as technically feasible. There are exceptions for tools that only perform standard editing or do not substantially alter the input. This is the watermark and metadata duty, and it sits with the tool maker, not with you.

Paragraph 4 is for you, the person publishing. The law calls you a deployer. If you use an AI system to generate or manipulate content that is a deep fake, you must disclose that the content has been artificially generated or manipulated. When the deep fake is part of an evidently artistic, creative, satirical, fictional or similar work, the duty is lighter: you disclose that generated content exists, in a way that does not spoil the work. A credit line or a note in the description is the usual answer.

The same paragraph adds a separate rule for AI generated text published to inform the public on matters of public interest. That text must be disclosed as AI generated unless it went through human review or editorial control and someone holds editorial responsibility. If you publish news style scripts that an AI wrote and nobody edited, that part applies too.

Paragraph 5 says when and how. The information has to be given in a clear and distinguishable way at the latest at the time of the first interaction or exposure, and it must meet accessibility requirements. A disclosure buried in a terms page that nobody reads does not meet that.

The dates, after the Digital Omnibus

The AI Act applies generally from 2 August 2026 under Article 113, and Article 50 follows that date. In 2026 the EU passed a simplification package, the Digital Omnibus on AI, which entered into force on 27 July 2026. It moved the heavy high risk rules to 2 December 2027 and, for AI built into regulated products, to 2 August 2028. It did not move the transparency rules. The only change for them is a short window for the machine readable marking in paragraph 2: AI systems that were already on the market before 2 August 2026 must comply by 2 December 2026. Everything else in Article 50 has applied since 2 August 2026.

Who it reaches

Article 2(1), point c extends the rules to providers and deployers based outside the EU where the output produced by the AI system is used in the Union. A channel run from the United States with a large European audience is in scope for the videos those viewers watch. Article 2(10) excludes people using AI in a purely personal, non professional activity, so a voice note for your family is not the target.

YouTube

YouTube asks creators to disclose content that is meaningfully altered or synthetically generated when it seems realistic. Its examples of what needs disclosure are content that makes a real person appear to say or do something they did not, altered footage of real events and places, and realistic scenes that did not happen. Its examples of what does not need disclosure include cloning one’s own voice to create voice overs or dubs, production help such as scripts, titles and thumbnails, and caption creation. Unrealistic content, such as animation, needs nothing.

Two lines on that page matter most to creators. Disclosing AI content won’t limit a video’s audience or impact its eligibility to earn money. And creators who consistently choose not to disclose may get a label applied manually, or face penalties including removal of content or suspension from the YouTube Partner Program.

TikTok

TikTok has required creators to label realistic AI generated content since 2023 and applies its own label to content made with its AI effects. Since 9 May 2024 it also reads Content Credentials, the C2PA metadata that some AI tools attach, and labels those uploads automatically. If a clip sounds or looks like something real that is not, turn on the AI generated content label when you post.

Facebook, Instagram and Threads

Meta’s wording is the broadest and the one most creators miss. Since February 2024 it has required people to use its disclosure tool when they post organic content with a photorealistic video or realistic sounding audio that was digitally created or altered, and it says it may apply penalties if they do not. It can also add a more prominent label when content carries a high risk of materially deceiving people on an important matter. A lifelike AI narrator is digitally created audio, so on Meta’s apps the careful move is to label it, even when YouTube would not require it.

Spotify

For music, Spotify announced on 25 September 2025 that vocal impersonation is only allowed when the impersonated artist has authorized it, and that it is backing an industry standard for credits so artists can show where and how AI played a role in a track. In the twelve months before that announcement it said it had removed more than 75 million spammy tracks.

Audible, through ACX

ACX, the route most independent authors use to reach Audible, is the strictest. Its audio submission requirements, dated 15 April 2026, say your audiobook must be narrated by a human unless otherwise authorized, and they list unauthorized text to speech, AI and automated recordings as not allowed. If you plan an AI narrated audiobook, check the rules of the specific store first. Our audiobook guide covers the options in detail.

China

China’s Measures for Labeling AI Generated Synthetic Content and a mandatory national standard took effect on 1 September 2025. They require explicit labels that people can see or hear, which for audio can be text, voice or rhythm cues, plus implicit labels in the file’s metadata. People publishing AI generated content on platforms must declare it and use the platform’s labelling tools.

United States, phone calls

There is no single US federal labelling law for AI voiceovers in videos. For calls there is a clear rule. On 8 February 2024 the FCC ruled unanimously that AI generated voices are artificial under the Telephone Consumer Protection Act, so calls that use them follow the rules for artificial or prerecorded voice calls, which means prior express consent. Several states have their own laws on voice likeness and election deepfakes, so check your state before using any voice that resembles a real person.