Telling Clients Which Parts of Their Video Used AI, and Why

A client of mine — a mid-size outdoor apparel brand — got a pointed question from their legal team last year: for a hero video going into a paid campaign, could we confirm nothing in it was AI-generated in a way that might trip a platform's synthetic-media disclosure rules or a licensing issue? I could answer in about four minutes, because I'd kept a short note of exactly where AI touched the project. The editor they'd used before me could not, and that gap is a large part of why I have the account.

This is the part of the job nobody trained us for. We're used to disclosing licensed music and stock footage. Disclosing which of our tools touched the frames is newer, and in 2026 it's shifting from a nice-to-have to something clients increasingly expect — sometimes because a platform requires it, sometimes because their own brand-safety people ask, and sometimes just because trust is worth money.

Why bother telling them at all

Three reasons, in rough order of how often they bite.

• Platform and disclosure rules. Ad platforms and some broadcasters have synthetic-media labeling requirements, especially for anything depicting real people saying or doing things they didn't. If you used a generative tool to alter a person's face, voice, or words, your client may be legally or contractually obligated to disclose it downstream. They can't comply with a rule they don't know applies. Your note is what lets them.
• Rights and provenance. Generative b-roll, AI-extended shots, AI-cloned voiceover, and music from generative tools all carry licensing and ownership questions that differ from a normal stock clip. If your client ever needs to prove they own or licensed everything in a video — for a bigger campaign, a resale, a dispute — a clear record of what was generated versus filmed versus licensed is gold.
• Trust and the relationship. Some clients care, on principle or for their audience, whether a voice is a real human or a clone, whether a testimonial face is real. Finding out later that something was AI when they assumed it wasn't is how you lose an account. Telling them up front is how you keep one.

Draw the line at "materially changes what the viewer believes"

You do not need to disclose that you used auto-sync, or that your NLE's scene-detection chopped the rushes, or that a denoiser cleaned the audio. Nobody wants a disclosure that reads like a software license. The useful test I apply: would a reasonable viewer or the client feel misled if they learned this after the fact? That splits the tools cleanly into two buckets.

Worth disclosing

• Generative video or b-roll that wasn't filmed (text-to-video shots, AI-extended or outpainted frames, generated backgrounds).
• Voice cloning or AI voiceover standing in for a real person, or any synthetic voice at all.
• Face or lip alteration — AI dubbing that reshapes mouths, face swaps, de-aging, expression edits.
• AI-generated music or sound that carries its own licensing story.
• Anything that changes what a real person appears to say or do.
• Generative fill or object removal when it fabricates content that wasn't there, as opposed to cleaning a boom mic out of a corner.

Not worth disclosing

• Transcript-based editing, silence removal, auto-cut rough assembly.
• Audio cleanup, denoise, voice isolation, leveling.
• Auto-captions and transcription (though you should still proof them).
• Upscaling, stabilization, color-matching, standard restoration.

That middle grey zone — generative object removal — is where I lean toward telling them anyway if I'm unsure. It costs one sentence and it never comes back to bite you.

What the disclosure actually looks like

Keep it plain. I include a short block in the delivery email and a matching line in the project's delivery notes. Something like:

"AI tools used in this project: generative b-roll for the three aerial-style transition shots (0:42, 1:15, 2:03); the rest is filmed footage. Audio was cleaned with AI denoise. Captions were auto-generated and human-proofed. No voices were cloned or synthesized; all on-camera people are real and unaltered."

Notice it does two jobs: it flags what was AI, and it affirms what wasn't. That second half — "no voices were cloned, all faces are real" — is often the part the client actually needed, because it's the assurance their legal or brand team is looking for. State the negatives on purpose.

Put it in the contract, once

Rather than reinventing this per project, I added a short clause to my standard agreement. In plain terms it says: I may use AI-assisted tools in production; I'll disclose in delivery notes any generative or synthetic elements that materially alter content or depict people; and the client is responsible for any downstream platform disclosures their distribution requires. That last clause matters — it puts the obligation to actually label the ad on the party who controls the distribution, while making sure they have the information to do it. I'm not a lawyer and this isn't legal advice; if your work regularly touches regulated categories or political content, have an actual attorney look at your clause. But having a clause beats having the conversation improvised over email after something's gone out.

Ask before you generate, not after

The cleanest version of all this happens at the start. When a project might call for generative elements — a shot they can't afford to film, a voiceover in a language the client doesn't speak — I raise it before I build it. "I can create that establishing shot with a generative tool rather than a location day; are you comfortable with an AI-generated shot there, and does your distribution have any labeling requirements I should design around?" Most say yes and appreciate being asked. The ones who say no just saved us both a revision cycle.

The honest trade-off

Some editors worry disclosure makes the work look cheaper or less impressive — that admitting the aerials were generated undercuts the magic. I've found the opposite. Clients read the note as a sign that I'm organized, that I understand their risk, and that I won't surprise them. In a market where anyone can generate a slick shot, being the editor who tells you exactly what you're getting is a differentiator, not a confession. The magic was never in hiding the tools. It's in the judgment about when to use them and the discipline to say so.