YouTube AI disclosure: a clip of someone giving advice they never gave

Yes, disclose it. YouTube's page on disclosing GenAI content (read 2026-10-08) lists, as a case that needs disclosure, making it appear as if someone gave advice that they did not actually give. The test is whether realistic content could be mistaken for something real. Editing steps that do not change what a person appears to say, such as trimming or burning captions, are not what the page is about.

What the page lists

The page says to disclose when you meaningfully alter or generate photorealistic content. It lists examples that need disclosure and examples that do not, and says where the label appears. Photorealistic AI content gets a label in the video player. Non-photorealistic or animated content gets it in the expanded description. YouTube may also apply a label itself to content made with its own GenAI tools or carrying C2PA metadata.

Where Sume tools fall

Video trim cuts a range from a clip and does not change what is said. Video captions burn text that Sume aligns to the speech, and with script_text the burned wording follows your script while the timings come from speech-to-text. That is the one place to be careful: captions that put words in a speaker's mouth that were not spoken are an editing choice you own. Keep script_text equal to what is said.

Generation is different. If you create a realistic person speaking, or clone a real person's voice, the examples on the left apply no matter which tool made it. A fictional character that looks realistic is not named in the page's examples either way, so read the page and decide; the label is a toggle you set when you publish on YouTube, and no Sume call sets it for you.

A practical rule

Write down, per clip, who appears, who speaks and whether they said it. If a real person is shown or heard and did not say it, disclose. If you are not sure, disclose: the label sits in the player or the description, and nothing on the page says an accurate label counts against you. Keep a note of the date you read the page, because the list has changed before.

Keeping a record

Keep a short log per upload: which clips are real footage, which are generated, and who is shown. If you cut a talk into a short with Sume trim, the spoken words are the speaker's own, and the log can say so. If you replace the audio, the log should say that too.

Where Sume produces avatar or generated video, the output is generated by definition, and the disclosure question moves to what it depicts. A presenter that looks like a real person and gives advice is the situation the page's first example describes. The decision is yours, and YouTube's page is the reference.