Around 86% of creators now use generative AI somewhere in their process. Declaring it is where people get caught out, because during 2026 disclosure stopped being a platform preference and became a legal obligation in the EU, with platform monetisation tied to it separately.
The short version: using AI as a tool is fine nearly everywhere and remains fully monetisable. Not saying so, where saying so is required, is what causes the damage.
The law: EU AI Act, in force since 2 August 2026#
The European Commission’s AI Office, together with national authorities, began enforcing the AI Act’s transparency obligations on 2 August 2026. They cover four situations: direct interaction with people, AI-generated content, emotion recognition and biometric categorisation and deepfakes plus AI-generated text on matters of public interest.
What that means concretely:
- Chatbots and interactive AI systems must tell people they are not human.
- Deepfakes must be labelled — images, video or audio generated or edited with AI.
- AI-generated or altered content must carry machine-readable marks so it can be detected automatically.
- Clear, visible labelling is required specifically for deepfakes, and for AI-generated text published to inform the public on matters of public interest.
Penalties reach €15 million or 3% of worldwide annual turnover. The Commission also published a voluntary Code of Practice on Marking and Labelling of AI-Generated Content, including a set of icons creators and publishers may use.
Sources, all official: European Commission on enforcement, Commission guidelines on transparency, Commission announcement. For a readable walkthrough of the relevant article, see this guide to Article 50.
Does it apply to you?#
Here we have to be careful, because the answer depends on your role and your audience and it is not the same for everyone.
The obligations attach to providers and deployers of AI systems. A creator using a generative tool to make content for publication can fall within scope as a deployer, particularly for deepfakes and for AI-generated text on matters of public interest. The framework reaches organisations outside the EU where the output is used inside it, so a UK creator with European viewers is not automatically outside it.
That is as far as we will go. Whether a specific obligation applies to your specific output is a legal question about your circumstances, and we are not lawyers. If you generate realistic depictions of real people, or publish AI-written material on public-interest subjects to a European audience, that is worth paying someone to answer properly.
The general position we would take: labelling clearly costs almost nothing, and the downside of not labelling where you should have is disproportionate. When in doubt, label.
Does this need a label? A working decision table#
Not legal advice and a reasonable rule of thumb built from what the platforms and the Commission have published.
| What you made | Label it? |
|---|---|
| AI helped research or draft; you rewrote it | No requirement. Mention it if you like. |
| AI did the editing, captions, translation, cleanup | No. |
| AI-generated illustration, obviously stylised | Not required by platform rules. Good practice to say so. |
| AI-generated photorealistic scene that never happened | Yes. Platform disclosure plus, potentially, the EU marking obligation. |
| Real footage altered so a person appears to say or do something they did not | Yes, without exception. This is the core deepfake case. |
| Synthetic voice reading your own script | Disclose. Cheap to do and it is what audiences expect. |
| Someone else’s voice or likeness, synthesised | Do not publish it without their permission. Labelling does not fix this. |
| AI-written text on a public-interest subject, EU audience | Very likely yes. Take advice. |
What each platform requires#
YouTube#
You must declare when AI has been used to generate or meaningfully alter photorealistic content: making a real person appear to say or do something they did not, altering real footage, or generating realistic scenes that never happened. The mechanism is the “AI use” toggle in Studio at upload, not a line in the description, which does not satisfy the requirement on its own.
What does not require the toggle: script drafts, editing, translation, colour work, or effects nobody would mistake for real footage. The test is realism plus the potential to mislead about something that actually happened. YouTube also states that the label does not reduce a video’s reach or its eligibility to earn; making video with AI goes through the paid uses that sit inside these rules.
Separately, YouTube’s Inauthentic Content policy since July 2026 excludes from monetisation content built from generic templates and mass-produced at scale, and AI personas giving advice on finance, law, health or medicine. Disclosure does not rescue content in those categories, which is a point people get wrong in both directions. See what gets you demonetised.
Music platforms#
Spotify is developing a disclosure standard with DDEX, letting artists state which elements involved AI, vocals, instrumentation, post-production. It added a Verified badge for human artists in April 2026 and removed more than 75 million tracks it classed as spam over the year. AI-assisted music is permitted; voice clones, mass uploads and misleading metadata are the enforcement risks.
Brand and advertising work#
If AI is involved in producing content for a brand, agree in writing who is responsible for disclosure. This gets overlooked routinely and the party that ends up liable may not be the one who generated the material. Ask before delivering, not after publication. Contract clauses that cost you money has wording for it.
How to word it#
Specificity is the whole thing. A generic “this content may contain AI-generated elements” reads as legal cover and buys you nothing with an audience. Saying what the tool did and what you did reads as confidence.
Some versions that work:
“Voiced with a synthetic version of my voice. Script, research and the opinions are mine.”
“The three landscape shots at 4:10 are generated. Everything with a person in it was filmed.”
“Drafted with AI, then rewritten, fact-checked and cut to about half its length by me. The errors that remain are mine rather than the model’s.”
“Illustration generated with AI. The charts are drawn by hand from the source data, because a generated image cannot be relied on to render a number correctly.”
That last pattern, declaring the AI use and in the same breath naming what was deliberately not left to a model, is the most useful shape available, because it turns the disclosure into evidence of judgement rather than an admission.
Put it where a viewer will see it: on screen or in the first lines, plus the platform’s own toggle. A disclosure nobody encounters is not a disclosure.
The commercial case for disclosing#
There is a widespread belief that labelling AI use costs you audience. On the evidence, concealing it costs more.
Audience appetite for AI-generated creator content fell from 60% in 2023 to around 26%. Mentions of “AI slop” rose ninefold in a year, with negative sentiment peaking at 54%. iHeartMedia’s internal research found 90% of its listeners want media made by humans — including listeners who use AI tools themselves, which is the detail that matters. People are not against the tools. They are against being deceived about what they are consuming.
The market response has been to turn human authorship into a claim worth making. The Authors Guild launched a Human Authored certification, described as an organic label for books, and had over 3,000 authors certifying more than 5,000 titles within a year. Not By AI runs a badge scheme for digital content. Spotify’s verified human artist badge is the same idea from the other direction.
So the stance is simple: disclose the AI you use and be specific about what you did yourself. It is more persuasive than silence and considerably better than being found out, because the cost of being caught is not a policy strike. It is the audience deciding they cannot tell what is real on your channel, which is not recoverable.
Sources: CNN on the AI backlash, ABC News on Spotify.
Common mistakes#
Disclosing in the description only. Platforms check their own toggle. Regulators care about whether a viewer would notice. A description line satisfies neither reliably.
Blanket disclaimers on everything. A site-wide “we may use AI” notice tells nobody anything and does not meet a per-content labelling requirement. It also devalues the disclosure on the pieces where it matters.
Assuming disclosure fixes a format problem. Templated mass-produced content is excluded from monetisation whether or not you declare the tooling.
Forgetting the machine-readable part. The EU obligation includes marks that can be detected automatically, not only text a human can read. If you are in scope, a visible caption alone may not be enough, this is one of the details worth checking properly.
Leaving it unsettled in brand work. Two parties each assuming the other is handling it is how nobody does.
What to do#
- Label anything photorealistic that could mislead someone about a real person or a real event. This is the clearest requirement and the one with the worst consequences if ignored.
- Use the platform’s own mechanism, such as YouTube’s AI use toggle, in addition to saying it in the content.
- Say what the tool did and what you did. Specificity reads as honesty; boilerplate reads as cover.
- Do not rely on disclosure to rescue a templated format. That is a separate problem with a separate fix.
- Settle responsibility for labelling in brand contracts, in writing, before delivery.
- If you produce realistic depictions of real people, or publish AI-written material on public-interest subjects to a European audience, get proper legal advice. That is the category where the penalties are real and this page stops being sufficient.
This describes rules as at that date and is not legal advice. The EU framework is new and guidance is still developing, check the Commission’s own pages, linked above, for the current position, and take advice if anything here might apply to you.