AI Label Checker

Free · No sign-up · Runs in your browser — your file never uploads.We read the machine-readable evidence platforms read, then predict — honestly — per platform.

We scan for C2PA Content Credentials, IPTC AI source declarations and known generator markers.

What decides the label — the short version

What this tool cannot tell you

  • It cannot read pixel watermarks (SynthID and similar) — those are checkable only by their owners.
  • It cannot see platform-side classifiers. "No evidence found" is a statement about the file, not a promise about the platforms.
  • It is not legal advice. Disclosure duties depend on the content and your jurisdiction.

The same file under different rulebooks

A note on "AI detectors": independent testing of AI-content detectors consistently finds meaningful false-positive rates — real human work flagged as AI. Where audits publish numbers, they are sobering: in the only published multi-tool audit (n=15 tools), detectors collectively mislabelled 13.33% of authentic photos as AI-generated, and the worst single tool reached 40%. On lightly edited real photos, error rates reached up to 93%. And recompression alone can collapse detection: in one benchmark a detector fell from 95.3% to 0.2% accuracy after JPEG quality-50 recompression — and re-encoding is what every platform does to every upload. That is why this site reports the evidence in your file and never an "AI probability score": a confident percentage from a fragile method would be a worse answer than an honest one.

What to do if a platform labels your real video as AI

It happens: a fully human-made video picks up an "AI" label because something in the production chain left markers behind — an AI-generated thumbnail, a stock clip from a generative library, an upscaler or editor that embedded provenance data. Before appealing anywhere, run the file through this checker and find out which evidence triggered it. If the markers are real and describe part of your video accurately, the label is doing its job for that part, and removing the segment or replacing the element is the honest fix. If the evidence is wrong or describes a tool that only touched your file in passing, you have something concrete to appeal with.

Whichever platform it is, appeals are won with records, not arguments. Keep the original camera files, the editor project/timeline, and any prompts, licences or receipts for stock and AI elements — and keep them after publishing, not just before. Practise provenance hygiene going forward: keep one untouched master, export deliverables from it, and check the export before upload. Don't run files through random "metadata remover" tools to dodge a label — that destroys your own evidence, and the disclosure duty applies to the content, not to the metadata.

Frequently asked questions

How do platforms decide to label a video as AI?

First, from machine-readable evidence in the file: C2PA Content Credentials and IPTC source-type fields that declare AI generation. YouTube reads these at upload and applies its 'AI' disclosure automatically; TikTok reads C2PA the same way and labels partner-tool content; Instagram/Meta reads C2PA and IPTC too. Second, from their own detection systems, which are not public and cannot be checked from outside.

So this checker gives a prediction, not a verdict?

Exactly — an evidence-based prediction. If your file carries AI declarations, platforms that read them will very likely label it: that's near-mechanical. If it carries none, no honest tool can promise you won't be labeled, because platform-side detection exists beyond the file's metadata.

I made it with AI but the file shows no markers. Am I safe?

From automatic metadata labeling, possibly — many editors strip metadata on export. But 'no label applied' is not 'no duty': YouTube requires you to disclose realistic AI content yourself, and the EU AI Act adds transparency duties (Article 50; deployer duties phase in to December 2026). Undisclosed realistic AI content risks demonetization or removal when detected later.

Does an AI label hurt my video's reach?

YouTube states that disclosing AI content does not by itself limit a video's audience or monetization eligibility. The label is a transparency signal, not a penalty flag. What hurts is being caught NOT disclosing.

What about SynthID?

Google's SynthID is a pixel-level watermark, not metadata — no outside tool can read it, and Google offers no public detection API. If your content came from Google AI tools, assume it may be identifiable through Google's own systems regardless of what any metadata scan says.

Do you upload my file to check it?

No. The scan reads your file's bytes in this tab only. Close the tab and everything is gone.

Related tools

CreatorChecks is not affiliated with YouTube, Google, TikTok or Meta.