The Provenance Layer: Designing AI Disclosure That Travels With the Cut

AI-generated frames are no longer a novelty in UK brand motion. They turn up in crowd scenes, product packshots, background plates, voice tracks and the odd impossible camera move. The awkward part is not making them. It is telling people they are there without turning a 15-second spot into a compliance leaflet.

That tension is now a design problem. Legal teams want a visible notice. Creatives want a clean frame. Platforms want metadata. Viewers want to know what they are looking at, but they will not read a paragraph in the corner of a vertical video. The answer is not to pick a side. It is to treat disclosure as part of the motion system, with rules, variants and a behaviour of its own.

Start with a disclosure tier, not a sticker

Not every use of AI needs the same treatment. A studio that uses AI to remove a cable from a plate is in a different place from a brand that generates a synthetic presenter. A sensible starting point is a three-tier approach decided at the brief, not at delivery.

Metadata-only makes sense where AI is an invisible assist and the final frame is not synthetic in any meaningful sense. A visible label belongs on work where a reasonable viewer could be misled about what they are seeing: generated people, generated scenes presented as real, synthetic voices, or images that imply a real event. Both is the safer call for high-realism work in paid media, especially where the brand is making a claim about a product or a place.

The wording matters as much as the tier. AI-generated is plain English. AI-assisted is useful when the tool helped but did not create the whole frame. Synthetic media sounds like a seminar. Keep it short, keep it honest and keep it consistent across the campaign so viewers learn what the label means.

The first stable frame is the real estate

The end card is too late for most social work. If the label only appears after the product shot, the viewer has already made up their mind about whether the scene was real. On vertical platforms, the first two seconds are the practical window. On broadcast and connected TV, there may be more room to place a lower-third after the opening title, but the label still needs a stable hold rather than a flicker.

A useful rule is to place the notice in the first stable frame where the viewer can actually read it. That might be a corner chip during the opening beat, a lower-third that sits with the brand mark, or a line in the end card for longer-form work. What it should not be is a two-frame flash that satisfies a checklist and nobody else. Hold it long enough to register, then let it leave before the call to action so it does not compete with the thing you want people to do.

Build a disclosure chip into the brand system

This is where motion designers earn their keep. A disclosure label should look like it belongs to the brand without pretending to be decoration. That means a small set of tokens: a type size that survives 320px-wide screens, a contrast ratio that meets accessibility guidance, a corner radius that matches the existing graphic language, and a motion behaviour that is calm rather than attention-seeking.

A simple fade or short slide is usually enough. Bounces, sparkles and glitch effects make the label feel like part of the spectacle, which is exactly the wrong signal. The chip should also have format variants: 16:9 for broadcast and YouTube, 9:16 for social, 1:1 for retail screens, and a static bug for live or stadium environments where movement is a distraction. Lower-thirds need a collision check against subtitles and platform UI. There is no point designing a beautiful notice that sits underneath the comment bar.

Metadata is the quiet layer

Visible labels are only half the job. C2PA-style Content Credentials, IPTC fields and sidecar manifests are the provenance layer that travels with the file. They are not a substitute for a visible notice, and a visible notice is not a substitute for metadata. The two do different jobs.

The practical work is in preservation. Credentials need to survive transcodes, re-edits, colour passes and the various exports that happen between the studio and the media agency. Ask early whether the platforms and ad servers in the plan actually read the metadata. If they strip it, say so in the delivery notes and lean harder on the visible label. Keep an asset register that records which shots used AI, which model or tool was involved where appropriate, and what the approved disclosure wording is. That register is also useful when a client asks a perfectly reasonable question three weeks after launch.

Rules that survive client review

Disclosure goes wrong when it is treated as a post-production chore. Put the decision in the treatment. Give one person ownership of the label across the campaign. Use plain English. Do not stack three notices because three departments asked for one. Test the label at phone size, in dark mode and against the busiest frame in the cut. Make sure it works with sound off, because most of the audience will meet it that way.

It also helps to separate disclosure from watermarking. A watermark is a rights and preview tool. A disclosure is a statement about how the image was made. They can sit near each other, but they should not be confused. If the brand wants a visible AI notice on every social cut, design it as a recurring element with a consistent position and timing. If the work is broadcast, check the relevant UK advertising and broadcasting expectations before the edit is locked, not after the clock starts.

The on-brand bit is not a contradiction. A disclosure can be neat, legible and consistent with the rest of the motion system. What it cannot be is hidden. The studios that get this right will not just keep legal happy. They will make provenance feel like part of the craft, which is a better look than a sticker sheet every time.