You used AI somewhere in your workflow — a script, a B-roll clip, a thumbnail — and now you're trying to work out what you're supposed to declare. So you read five articles and got five different answers, at least two of which claimed that ticking the disclosure box will quietly kill your monetisation.

That last one is wrong, and it's the most repeated claim on the subject. Everything below was read from the platforms' own policy pages rather than from other people's summaries of them.

What you actually have to disclose

Every major platform now requires disclosure for the same narrow thing: AI-generated or altered content that looks real. A real person appearing to say something they didn't, an event that never happened, a place that doesn't look like that. Using AI for scripts, thumbnails, captions, editing, upscaling or idea generation is not disclosable anywhere. And on YouTube — the platform creators worry about most — disclosing does not affect monetisation at all; the help page says so in as many words. What risks your money is a completely separate policy about generic, mass-produced content. Those two things get conflated constantly, and the confusion is doing more damage than the rules.

Verified against primary sources in August 2026. Policies move fast — check the linked pages before making a decision that matters. This is information, not legal advice.

The confusion is two rules wearing one coat

Nearly every piece of bad advice on this topic comes from collapsing two separate policies into one.

Rule one is disclosure. It's about deception. Does this content look real when it isn't? If yes, label it. It has nothing to do with earnings.

Rule two is monetisation quality. It's about effort. YouTube's July 2025 policy update renamed "repetitious content" to "inauthentic content," and the operative section — Generic or Repetitive Content — names as non-monetisable: "AI-generated content made with generic or unoriginal templates giving the impression of mass production without adding the creator's original, authentic insights or perspective."

Read that again. The trigger isn't AI. It's templates and mass production. A channel publishing forty near-identical AI-narrated videos a week is at risk whether or not it discloses. A creator using AI thoughtfully on one considered video is not.

There are two further sections most coverage misses entirely: Unsatisfying or Off-putting Content, and AI Personas Related to Sensitive Topics — AI "doctors" or AI financial and legal advice personas, which aren't monetisable at all.

Sources: YouTube AI disclosure and YouTube monetisation policies

Who applies the label — you or the machine

The second thing worth understanding is that self-disclosure is only half the system. Platforms increasingly detect and label without you.

The mechanism is C2PA Content Credentials — provenance metadata embedded by the tool that generated the content. YouTube, TikTok and Meta all read it. When they find it, they apply a label automatically, and on YouTube and TikTok you cannot remove it.

Three practical consequences:

  • Your generation tool decides for you. If it writes C2PA metadata, the label is coming regardless of what you tick.
  • Stripping metadata is a bad idea. It reads as evasion, and detection isn't only metadata-based — platforms run their own classifiers too.
  • Disclose anyway. If the label is going to appear, appearing to have volunteered it is strictly better than appearing to have been caught.

Platform by platform

YouTube. Disclose realistic synthetic content via the "AI use" setting under Attributes in Studio. Note the name — older articles and screenshots still say "Altered content," and the UI has moved on. Labels appear in the expanded description, and on the player itself for photorealistic content. Exempt: scripts, thumbnails, captions, upscaling, idea generation, and cloning your own voice for voiceover or dubbing. Automatic labels apply to YouTube GenAI tools, C2PA-tagged uploads, and internal detection.

TikTok. Labelling is required for realistic AI images, audio or video, and encouraged for anything substantially AI-made. Auto-labels fire on TikTok AI effects and C2PA credentials, and can't be removed. Some AI content is banned even when labelled: fake crisis events, fake authoritative sources, false endorsements, likenesses of under-18s, and private adults without permission. Source

Meta (Instagram, Facebook, Threads). The "AI info" label fires on detected C2PA/IPTC metadata and invisible markers, or on self-disclosure. The written obligation is narrower than most reporting suggests: Meta requires disclosure for organic content with photorealistic video or realistic-sounding audio. Still images aren't covered by that stated requirement — they get labelled by detection instead. Source

Substack. "Scan for AI text" launched 21 July 2026, powered by Pangram. It's reader-initiated from the three-dot menu, not an automatic badge, and returns an estimated human-versus-AI percentage. It covers posts from that date onward, doesn't work on audio, video, email or custom domains, and can be disabled manually per post — there's no publication-wide switch. You can also add a "How I make this" statement in Settings. Source

Etsy. Under the Creativity Standards, seller-prompted AI creations qualify as "Designed by a seller" and are allowed — but you must disclose AI use in the listing description. Worth knowing if you sell digital products: AI prompt bundles are explicitly prohibited, in the same clause that bans collections of someone else's work. Standards were last updated June 2025 and applied to existing listings. Source

EU AI Act. Article 50 applies from 2 August 2026 — in force now. Providers must mark AI output in machine-readable form; deployers must disclose deepfakes and label AI-generated text published to inform the public on matters of public interest, with an exemption where a human reviewed it and holds editorial responsibility. The part that catches creators: the Commission's FAQ puts personal-capacity posting outside scope, but says you become a deployer once you derive economic benefit on a regular basis. Monetised creators are in scope. Reaching EU users is what matters, not where you live. Penalties run to €15m or 3% of worldwide turnover. Commission FAQ

The money question: can you own it

Separate from disclosure, and settled.

The D.C. Circuit affirmed in March 2025 that the Copyright Act requires a work be authored by a human being. The Supreme Court denied certiorari on 2 March 2026, so anything you read describing Thaler as ongoing or headed for the Supreme Court is out of date.

What survived matters: the court was explicit that this "does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence." The Copyright Office's position is that AI-assisted work is registrable where a human contributed sufficient expressive elements — but prompts alone don't confer authorship. Prompt, and you own nothing. Prompt, then meaningfully select, arrange and edit, and the human contribution is protectable.

The quick reference

Platform Disclose what Auto-labelled? Real risk
YouTube Realistic synthetic people, events, places Yes — GenAI tools, C2PA, detection Not disclosure. Template mass-production under Generic/Repetitive
TikTok Realistic AI image, audio or video Yes — TikTok effects, C2PA; can't remove Banned categories stay banned even when labelled
Meta Photorealistic video, realistic audio Yes — C2PA/IPTC, invisible markers Unspecified penalties for non-disclosure
Substack Nothing mandatory No — reader-initiated scan False positives on your own writing
Etsy AI use, in the listing description No Prompt bundles prohibited outright
EU AI Act Deepfakes; public-interest AI text N/A Applies to monetised creators reaching EU users

Three prompts worth keeping

1. The disclosure statement writer

Write a short, plain-English statement describing how I use AI in my work,
for my [channel about page / newsletter / shop listing]. Be specific about
what AI does and what I do. No defensiveness, no hedging, no apologising
for using tools. Under 80 words.

What I use AI for: [be honest and specific]
What I always do myself: [the human parts]

Why it's built this way: the "no apologising" constraint matters — most AI disclosure statements read like confessions, which invites the suspicion they're trying to defuse. Specificity reads as confidence. Swap in: your real workflow. Vagueness here is worse than nothing.

2. The back-catalogue audit

I'm going to describe how I made a piece of content. Tell me, for each of
YouTube, TikTok, Meta and Etsy, whether it needs disclosure under their
current rules — and where a rule is ambiguous, say so rather than guessing.
Cite which rule you're applying.

How it was made: [describe it]
Where it's published: [platforms]

Why it's built this way: the "say so rather than guessing" clause is the whole prompt — without it models produce confident answers about policies that changed last month. Treat the output as a shortlist to verify, never as an answer. Swap in: one piece at a time; batching makes it sloppy.

3. The mass-production check

Read the content below and tell me honestly: would this pass as
original work with a specific point of view, or does it read as
templated output? Point to the specific sentences that make it feel
mass-produced. Be blunt.

[paste your content]

Why it's built this way: this tests against the criterion YouTube actually applies — original insight versus template — instead of against a detector. Swap in: nothing. Run it before publishing anything you made quickly.

What to actually do

Disclose the realistic stuff and stop worrying about it — on YouTube it costs you nothing, and the label is likely coming anyway via C2PA. Then put your effort where the real risk is: not being template output. Every platform rule, read closely, points the same direction. Nobody is penalising the tool. They're penalising the absence of a person.

Related guides

→ AI Prompts for Content Creators — the working set, ready to copy.