YouTube AI content disclosure: creator checklist

This is a YouTube upload and review checklist, checked against official help on October 22, 2026. It covers the native disclosure decision, the words that explain a synthetic segment, and the record you keep afterward. The worked project below is hypothetical; it is not a published upload or a tested Magic Hour output. Platform wording can change, so use the current linked instructions when preparing a real upload.

What the current rule asks you to identify

YouTube’s current GenAI guidance covers realistic, meaningful alterations. The familiar categories are a real person seeming to say or do something they did not, changed footage of a real place or event, and a realistic scene that never happened. YouTube’s disclosure announcement explains that viewer confusion about authenticity is the concern.

Its current examples also include AI-generated music and distinguish cloning your own voice for voiceovers or dubs from synthetic speech that attributes words to somebody else. Inspect sound as well as images. AI-assisted scripts, captions, or minor cleanup alone are different from a realistic synthetic segment; review the complete export rather than treating the editing tool’s name as the answer.

Build a timestamped inventory of the final export

Watch once for meaning, then review the relevant segments with the source files beside you. Do not start from an old prompt list: a generated insert may have been removed, or a realistic extension may have been added during editing. Use the actual file that will be uploaded, including its final voiceover and music mix.

Copy this inventory into your production notes. One row should identify one meaningful alteration or one element that needs a closer look. For an unchanged recording, write that it is source footage rather than leaving a blank that another reviewer has to interpret. The last column records a reason, not merely a yes/no tick.

For a team handoff, add the final filename, export version, reviewer, review date, and unresolved question above the table. Preserve a reference to the original clip or recording so a reviewer can understand what changed. You do not need to share private prompts or personal source media publicly; keep the internal record accessible to the person responsible for the upload.

Worked example: a city explainer with one generated insert

Imagine a two-minute explanation of drainage design. It contains the creator’s recorded narration, filmed street footage, and a six-second photorealistic insert showing water flowing through a named city street. The insert was generated to illustrate a scenario; it is not footage of an actual flood. The timestamps here are made up for the example.

The creator should not title the upload “Flood caught on camera” or describe the insert as eyewitness evidence. A specific on-screen caption at the insert could read: “AI-generated illustration; not footage of an actual event.” The description can identify the range: “The scene at 00:25–00:31 is a generated illustration of the scenario discussed.” Those are proposed words for this hypothetical brief, not a guarantee that every use or video is permitted.

This review produces an upload decision for the finished video and context for the segment. It does not mean the other 114 seconds were generated. Keep the distinction visible in the record so a teammate does not overstate how much of the video is synthetic. If the insert is replaced with a clearly stylized diagram before export, review the new file and update the inventory instead of carrying the old decision forward automatically.

Write context that identifies the alteration

Use a sentence that tells the viewer what changed and what not to infer. “AI was used” may be too broad to explain which part is illustrative. “Everything is real” contradicts a realistic generated insert even if the underlying topic is factual. Avoid a disclaimer that sounds like a confession unrelated to the scene; describe the actual production choice.

Reusable description template: “At [time range], [image/audio/scene] was [generated or meaningfully altered] to [illustrative purpose]. It is not [a real recording, an actual quotation, or evidence of the depicted event]. [Source/context for the factual explanation].” Replace every bracket before upload. Use only the clauses that fit; do not imply a real person consented unless that approval exists.

For a fictional portrait scene, a suitable production note might be: “The presenter is a synthetic character; the spoken text is scripted brand narration.” For an illustration of a historical setting, explain which visuals are reconstructed and cite the sources for the historical statements separately. Put context close enough to the relevant segment that someone watching only an excerpt can understand it. This is an editorial recommendation alongside the native platform setting.

Use the current YouTube Studio upload control

Open YouTube Studio and start the normal upload flow. Current official desktop instructions put “AI use” in Attributes: select Yes for qualifying content or No when it does not qualify. Continue the other video details. This replaces older walkthrough wording that directs you only to an “Altered content” field. Follow the live interface and desktop help instructions if its layout differs.

Prepare the inventory before entering the upload flow, then keep it open while answering the question. Record the answer selected and the final export it belongs to. Add the specific description or on-screen context you planned. Text in a description is useful context; it should not be your operational substitute for answering the native question when disclosure is required.

Before scheduling, compare the uploaded title, description, thumbnail, and spoken framing with the export. A neutral explanation can become misleading if its thumbnail depicts a synthetic scene as a real incident. If the upload was prepared by someone else, have the final approver check the actual fields rather than assuming the editor’s message “AI disclosed” means the setting was saved.

A reusable preflight checklist

1. Final file identified: the filename and version match the upload, including the actual soundtrack. A preview of an earlier cut does not close this check.

2. Inventory complete: generated or meaningfully altered inserts, faces, speech, and music have been reviewed. Each uncertain element has a source reference and a named owner for the decision.

3. Framing accurate: title, thumbnail, captions, and narration do not present a synthetic illustration as an authentic recording or quotation.

4. Native answer recorded: the AI-use answer belongs to this upload and reflects the final inventory. Check again after a meaningful edit or replacement of a segment.

5. Context specific: any extra wording names the relevant alteration and avoids claiming that permission, factual accuracy, or platform approval follows from the label.

6. Other approvals separate: source and likeness permissions, music rights, commercial-use terms, and any paid-promotion review have their own record and owner.

7. Watch-page follow-up assigned: one person will inspect the resulting player and expanded description, save the URL, and record what the viewer can see. If the video is scheduled, this check remains pending until the relevant watch page can be inspected.

Check the viewer-facing label, including automatic labels

YouTube’s “How this content was made” help describes disclosures in the player or expanded description. Manual disclosure, YouTube’s AI tools, and valid Content Credentials (C2PA 2.1 or higher) indicating the entire video was made with AI can produce a “Made with AI” disclosure. Do not require one fixed label position to consider the review complete; inspect the player and expanded description.

Save the watch URL, upload version, disclosure decision, review date, and a note or screenshot of the visible treatment. Check that your explanatory text remains present as well. Record “verified” only after inspecting the page; a scheduled upload or a saved setting is a different state. If you cannot inspect the viewer-facing page yet, leave that item pending and name who will finish it.

Current YouTube help also describes labels from internal detection and Studio notifications. Correction options are limited: some detection errors can be corrected through the AI disclosure survey, while labels tied to YouTube AI tools, C2PA, or manual review cannot be adjusted. Read the current notice and linked instructions rather than promising that every automatic label is removable.

Do not confuse camera provenance with AI disclosure

The “Captured with a camera” disclosure is a separate provenance feature. YouTube’s help says it depends on compatible capture technology and metadata; its absence does not prove a video was synthetically generated. A clean-looking image or missing label is not an authenticity test. Keep your own original-media references even when the platform carries provenance information.

For this checklist, the practical question is whether the upload decision and context match what you actually made. Do not strip or alter metadata to evade labeling. If a client asks what a provenance mark means, point them to the specific help page and your production record instead of treating a camera badge as permission to use a face, voice, or music track.

Disclosure is one gate in a broader release review

YouTube’s Community Guidelines still apply. A label does not make impersonation or misleading content acceptable. Use the commercial-use rights checklist for permissions, and the AI video QA checklist for the finished file. The general disclosure guide covers cross-platform decisions.

If you create supporting shots in Magic Hour, keep their origin with the project and review the new export before uploading. Use AI Video Generator for that production step; finish the upload disclosure and viewer-label review in YouTube Studio.

Original YouTube references

YouTube: AI content disclosure rules

YouTube: camera provenance labels