Why provenance is now a marketing task
AI-made creative is moving from experiment to default. The IAB's State of Data report projects that AI-generated video will make up around 40 percent of all video ads, and it found that 86 percent of digital video ad buyers are using or planning to use generative AI for creative. When that much content is synthetic, audiences and platforms start asking where an asset came from.
For a small business the practical questions are ordinary ones: was this image edited, is that person real, do we still have the original, and who approved this claim. Answering them takes a little structure, and this guide covers three pieces: the technical standard, the platform labels, and your own policy. None of it is legal advice, and rules differ by country and platform, so check current policies before you rely on anything here.
What Content Credentials and C2PA are
C2PA is the Coalition for Content Provenance and Authenticity, an industry group that publishes an open technical standard for attaching a signed history to a media file. Content Credentials is the name used for the user-facing version of that idea. Think of it as a tamper-evident label that travels inside the file.
What a credential can record
- Which tool or camera created the asset.
- Which editing steps were applied afterwards, including whether generative AI was involved.
- Who signed the record, so a viewer can decide how much to trust it.
What it cannot do
A credential does not prove that an image is true. It shows a history and who stands behind it. It is also fragile in practice: a screenshot, a re-export or an upload that strips metadata can remove it, and many platforms do not yet display it. That is why the standard helps but should not be your only record. It also depends on the tools in your chain: a credential is only written if the software that created or edited the file supports it, so a mixed workflow can leave gaps in the history. Keep your own files and notes as well, as described later in this article.
What platforms label on their own
Platforms are adding their own layers, and they change often. One concrete example: ByteDance's Seedance 2.0 is integrated into TikTok's Symphony Creative Studio, and content made there gets an automatic AI-disclosure label. In that case the platform handles the labeling for you, which is convenient but only covers assets made inside that tool.
In general terms, major social and ad platforms offer creators a way to mark content as AI-generated or AI-altered, some ask for that label on realistic content, and some read embedded provenance data where it exists. The details vary and move quickly. Read the current policy page for each platform where you publish, and revisit it each quarter.
When to disclose
A useful test is to imagine a reasonable viewer learning exactly how the asset was made. If they would feel misled, disclose. The list below is a starting point for your own rules.
Disclose by default
- Realistic people, including AI avatars and synthetic voices presented as if they were real customers or staff.
- Altered product results: before and after images, texture, size, performance or anything a buyer would use to decide.
- Realistic scenes of events that did not happen, or places you did not visit.
- Testimonials or reviews that are generated rather than quoted from a real person.
Usually not needed
Routine touch-ups such as color correction, cropping, background cleanup or resizing rarely change what a viewer believes. Clearly stylized illustration and abstract graphics are also low risk. If you are unsure which side an asset falls on, disclose. The cost is a short line of text.
Avatar and UGC-style ads deserve the most care, because the format invites viewers to assume a real person is speaking. Our article on AI avatars and UGC for brand marketing looks at how to use them without misleading anyone.
Write an internal disclosure policy
A policy can fit on one page. Its main job is to remove case-by-case debate. Here is a version for a fictional bakery, Hearth and Crumb, with three staff and a single social manager.
- Scope: applies to every image, video, voiceover and caption published under the bakery's accounts, including paid ads.
- Always label: any AI-generated person, any generated voice, and any image where the product shown was created or materially altered by AI.
- Never do: present a generated person as a real customer; show a cake, size or decoration we cannot make; imitate a real individual or a competitor's brand.
- Wording: use the platform's built-in AI label where available. Otherwise add a plain line such as "Image created with AI, product photographed in our kitchen."
- Approval: the owner signs off on every AI-made ad before spend begins, and records the date.
- Records: save the source photo, the prompt, the final file and the approval note for each campaign for at least a year.
- Review: reread platform policies and this page every quarter, and update it when a rule changes.
Nothing in this list needs software. A shared folder and a spreadsheet are enough to start.
Write the label in plain words
Disclosure text works best when it is short, specific and placed where people will see it, next to the asset or in the platform's own label field. Avoid technical language and avoid apologizing. A label is information, not a confession.
- For a generated background: "Background created with AI. Product photographed by us."
- For an avatar presenter: "AI-generated presenter. Not a real customer."
- For a fully generated visual: "Image created with AI."
Match the label to what actually happened. Saying "enhanced with AI" on a fully generated scene understates it, and saying "AI-generated" on a lightly retouched photo can confuse people about the product. Precision builds more trust than caution does.
Keep source files and edit history
If a customer, platform reviewer or partner asks how an ad was made, you want the answer in one place. For each campaign, store four things together: the original photo or footage, the generation settings and prompts, the exported final, and a short note of who approved it and when.
Name files so the story is readable without opening them: spring-loaf-original.jpg, spring-loaf-ai-background-v2.png, spring-loaf-final-1080.png. Keep the layered or project file too, since it shows what was added on top of the generation. Be careful with what you publish, though. Prompts and private customer details belong in your internal record, not in public metadata.
Mistakes that cause trouble later
Most problems come from a few repeat patterns rather than from bad intent.
- Deleting the original photo after generating a variation, then having nothing to show when asked.
- Letting a freelancer publish AI-made assets without telling the account owner.
- Reusing one generated face across many ads until it reads as a fake customer base.
- Assuming the platform's label covers assets made in a different tool.
- Writing the policy once and never opening it again.
Each of these is cheap to prevent with the one-page policy above and a quarterly fifteen-minute review.
Where this fits in your workflow
The habit that makes all of this stick is doing it at the point of creation, not after publishing. In the AI video generator and image tools in SEENALYZE AI, drafts, brand settings and edits live in the same workspace as the final post, so the history of an asset stays close to the asset. Our guide to keeping AI visuals on-brand shows the editing side of the same discipline.
Frequently asked questions
Are Content Credentials required?
They are a voluntary standard, not a universal requirement. Some jurisdictions and platforms are introducing their own labeling rules, so check current guidance for the places you sell and advertise.
Do I need to label AI-edited product photos?
If the edit changes what the product looks like or does, yes. Cropping, color balance and background cleanup usually do not need a label, but when in doubt, add one.
Will a label hurt performance?
We do not have reliable data on that, so test it on your own audience. Honest disclosure protects trust, and a misleading ad costs far more than a label does.
What if a platform strips the credential?
It often will. That is why you keep your own source files, prompts and approval notes as the durable record.
Keep every creative traceable
Draft, edit and approve AI-made posts in one workspace so source files and decisions stay attached to the final asset.
