containsSyntheticMedia: The AI Disclosure Field Your Uploader Never Sends

The field exists. The upload path around it does not.

When a person uploads through YouTube Studio, they are walked past an Attributes screen with an AI use question and a Yes / No choice. It is unavoidable. Answer it and the disclosure is recorded.

When software uploads through videos.insert, that screen never exists. Google added status.containsSyntheticMedia to the API on 30 October 2024 precisely so programmatic uploads could carry the same declaration — it is settable on videos.insert and videos.update, and it is returned in the video resource once set. But it is optional in the request. Omit it and the API does not complain, does not warn, and does not prompt anyone.

That asymmetry is the whole problem. A compliance obligation that is enforced by a form in one interface becomes an unenforced optional field in the other, and product teams building on the API discover the gap only when a label appears on a client’s video.

What YouTube actually requires you to declare

Google’s own wording is narrower than most people assume, and the narrowness is what makes this hard to automate. Disclosure is required for content that is generated or meaningfully altered with AI and appears realistic. YouTube gives three defining cases:

• Content that makes a real person appear to say or do something they did not do
• Content that alters footage of a real event or place
• Content that generates a realistic scene that did not actually occur

Read those two columns as a product specification and the difficulty becomes obvious: the boundary is a judgement about realism and about whether a viewer could be misled. No automated pipeline can infer it from the file. Someone has to answer the question per video, and your software has to give them somewhere to answer it.

Where programmatic uploaders actually break

1. The question is never surfaced

   The scheduler’s upload form mirrors title, description, tags, category, privacy and publish time — the fields everyone copied from the sample code years ago. The AI-use question was added to the platform later and never made it into the interface, so the operator is never asked and the field is never populated.

2. There is no per-video place to store the answer

   Even when a team decides to support it, the declaration is per-video and cannot be a global account setting. That means a schema change, a migration, a queue format change and a back-fill decision for everything already scheduled.

3. Bulk and CSV import paths bypass it entirely

   Agencies and SMM panels move video in batches. A bulk importer that was written before the field existed will keep uploading without it long after the single-video form has been updated.

4. The correction path is a second, different call

   Fixing an already-uploaded video means a videos.update against the same status part, with the part semantics handled correctly so nothing else on the video is wiped. Teams that treat update as a copy of insert routinely clear other status fields by accident.

5. Nobody owns the policy decision

   The developer builds the field, the agency operator uploads the file, and the channel owner carries the penalty. In multi-tenant products those are three different parties, and the obligation quietly falls into the gap between them.

The label can arrive whether you declare or not

This is the part that turns a documentation detail into a commercial risk. YouTube does not rely only on your declaration. It states that a label may be applied automatically to content made with YouTube’s own generative AI tools, to content that carries C2PA Content Credentials metadata, and to content its internal systems detect as AI-generated or altered.

A creator who was labelled in error can normally change the answer in the Studio disclosure survey. There are three documented exceptions where they cannot: content made with YouTube’s AI tools, content containing C2PA metadata, and content labelled after manual review. In those cases the label is permanent, and no amount of arguing with the tool vendor removes it.

What non-disclosure costs

Google is explicit that disclosing does not shrink a video’s audience and does not affect its eligibility to earn money. The downside is entirely on the other side of the choice.

• A label applied manually by YouTube, which the creator cannot remove
• Content removal under the Community Guidelines, which apply to AI-generated material exactly as they do to anything else
• Suspension from the YouTube Partner Program for creators who consistently choose not to disclose
• For a SaaS or agency, a client channel damaged by a defect in your upload pipeline — with the paper trail pointing at your software

None of these is an API error. Your integration keeps returning HTTP 200 the entire time.

Where this compounds with the rest of your API footprint

The disclosure field rarely fails on its own. It sits on top of three constraints that already catch programmatic uploaders, and a review of one usually surfaces the others.

• The unverified-project restriction. Every video uploaded through videos.insert from an unverified API project created after 28 July 2020 is forced to private viewing — which is why a “successful” upload can be invisible. Covered in YouTube API uploads locked as private.
• The upload bucket. videos.insert sits in its own quota bucket at 100 calls per day, separate from the general unit pool. Raising it is a review, not a setting — see how the quota increase audit works.
• The compliance audit. Whether your client honours YouTube’s policy-required fields is exactly the kind of thing a Developer Policies review examines. The rules that quietly fail projects are set out in the YouTube API compliance audit guide.

Treated together, these are a product-design problem rather than a coding problem: what your interface asks the operator, what your database stores per video, how your bulk path behaves, and how you correct a video after the fact. That design work — and the audit preparation that follows it — is where most teams want help.