Every Big Platform Now Labels AI Video — Here's the Split We'd Build Instead

If AI makes your marketing videos end to end, the platforms will eventually say so on your behalf — so the cheap move this month is to decide which parts of your pipeline are synthetic before a detector decides for you.

What the platforms actually require (as of July 2026)

This is the part that's written down, with links. Everything after it is our judgement, and we'll say so.

• YouTube requires creators to disclose realistic altered or synthetic content at upload, and says it may apply a label itself on content with the potential to confuse or mislead viewers.
• TikTok requires you to label realistic AI-generated content and auto-labels content that arrives carrying C2PA provenance metadata.
• Meta announced in February 2024 that it would label AI-generated images across Facebook and Instagram using industry provenance signals.
• Google Search doesn't penalise AI as such. Its spam policies name scaled content abuse — mass-producing pages to game rankings — regardless of how the pages were made, and Google has said it rewards high-quality content however it's produced. The disqualifier is low-value volume, not the tool.

So: disclosure obligations on the two biggest short-video platforms, automated labelling machinery on all of them, and a search engine that's explicitly agnostic about the tool but not about the volume.

What we can't tell you

We have not found a platform statement quantifying what an AI label does to your reach, your payouts, or your conversion rate. Anyone who gives you a percentage on that is guessing.

Our read — clearly labelled as opinion — is that the machinery being installed first (labels, provenance metadata, mandatory upload flags) is the machinery you'd need if you later wanted to sort content by how it was made. We plan around that possibility because the plan is cheap. That's a bet on direction, not a reported fact.

The fix is a split, not a retreat

Don't stop using AI. Split the pipeline in two and be deliberate about which side each task sits on.

Keep the human in: anything a customer would recognise as you. On-camera presence, the voice in a testimonial, before/after photos of your actual work, the answer to a technical question in your field.

Keep the AI on: scripts and outlines, captions and subtitles, descriptions and chapter markers, repurposing one shoot into six clips, alt text, first-draft blog posts, comment triage, title and hashtag variants, transcript cleanup.

That split is deliberately boring. It survives the next policy change, because the synthetic parts are the parts no platform is policing for authenticity.

Worked example: 12 short videos a month

Every number below is hypothetical. These are our assumptions about a typical small-shop workload — not measured client data, not a benchmark. Replace each one with your own times before you trust the total.

Assume a $60/hour loaded rate for owner time (your rate will differ).

Option A — fully synthetic. AI script, AI avatar presenter, AI voice, AI b-roll. Assume 35 minutes per clip once templates are dialled in, mostly spent re-rolling takes that land wrong.

• 12 × 35 min = 7.0 hours → $420/month (assumed)
• Disclosure required on YouTube and TikTok: yes, per the policies linked above

Option B — AI draft, human on camera. AI writes twelve scripts. You batch-film all twelve in one sitting on a phone. AI handles captions, cuts, titles and descriptions.

• Scripts and shot notes: 12 × 10 min = 2.0 hours
• One batch filming session: 1.5 hours
• Edit, caption, publish (AI-assisted): 12 × 15 min = 3.0 hours
• Total 6.5 hours → $390/month (assumed)

On these assumptions the human version is slightly cheaper, because batching the filming beats re-rolling twelve synthetic takes. Run it with your own numbers and it may come out the other way — say the hybrid runs an hour longer, $60 more. That's a small premium for keeping the labelling decision in your hands.

The expensive version isn't AI or human. It's producing one video at a time, on a Tuesday, with no template.

The disclosure habit

Disclosure is cheap as a checkbox in your workflow and expensive as a retraction. Three rules we run on our own content:

1. Flag it at upload, every time. YouTube's altered-or-synthetic toggle goes on for anything with a synthetic presenter or voice. It costs nothing, and it's required for realistic synthetic content anyway.
2. Put the disclosure where a human reads it. One line in the description or at the foot of the post: what was AI-assisted, what wasn't. Our tutorial videos state on screen that the presenter is an AI avatar; our posts carry the "tbdb.ai studio" byline for the same reason.
3. Never let AI assert a fact you haven't checked. At your scale the failure mode is an AI-written post quoting a price, a spec or a regulation that's wrong — and a customer holding you to it. Every claim gets a source or it gets cut.

What we'd actually build

For a shop publishing 8–16 short videos a month, the build is small and it's mostly plumbing, not models:

• A script generator wired to your real inputs — recent jobs, FAQ emails, review text — so the twelve scripts are about your work rather than generic advice. This is the piece owners skip, and it's the one that makes the videos worth filming.
• A batch shot list delivered the day before your filming block: twelve prompts, one line each, in filming order.
• A post-production pass that turns raw clips into captions, three title options, a description with the disclosure line pre-written, and platform-specific crops.
• A publishing checklist with the synthetic-content flag as a required field, so nobody has to remember.

Typical scope for us as of July 2026: a one-week build, then you own the templates. The recurring cost is transcription and model usage; at this volume it's a small line item against the six-plus hours a month you're already spending. We're not quoting a figure here because it depends on your clip length and tooling — ask and we'll price your actual volume.

The one-line version

AI is the cheapest part of your content operation and the part platforms are building the most machinery around. Keep it on the drafting, the captions, the repurposing and the admin. Keep your face on the camera. Tick the disclosure box. That's the whole strategy, and it doesn't need revisiting the next time a platform changes its mind.

---

Want this mapped to your actual channels? Book a 30-minute strategy call. We'll look at what you're publishing now, mark which pieces would need a synthetic-content disclosure, and sketch the AI-draft/human-on-camera split for your next month of content. No deck, no pitch — you leave with the shot list either way.