AI labels on Instagram, TikTok and YouTube: what C2PA means for your product posts
Find out which of your product pictures and videos need an AI label, who adds it, and what C2PA has to do with it. You get a decision table and a five-step routine to use before every post.
By SemPost EditorialWritten with AI, read by an editor
Published 10 min read

In this article
Why a label is a product question, not only a platform rule
An AI label is easy to read as a penalty. For a shop it is better read as a statement about the product: this scene was made, the product in it is real. Seen that way, the question is not "how do I avoid the tag?" but "does this picture tell the buyer the truth?"
A label reaches a post in two ways:
- You switch it on yourself when you publish, using the network's disclosure setting.
- The network adds it after reading signals in the file, such as embedded records or metadata written by the tool that made or edited it.
The second way is the one that surprises people. You can end up with a tag you did not choose, or without one you expected, because of what is inside the file.
For a small shop, the real risk is not the tag. It is a buyer feeling that a picture promised something the product does not look like. A label on a clearly AI-made scene is usually harmless. We know of no data showing that a label changes reach or sales either way, and nobody can promise you one. A missing label on a picture that misleads is the actual problem, because it costs you trust and returns.
The rest of this article gives you a way to sort your pictures, understand the technical side, and build a routine so the decision is made on purpose.
What C2PA and Content Credentials are, and what they are not
C2PA (opens in a new tab) is an open standard for attaching a signed record to a picture or video. The record can say who made the file, with which tool, and what was edited. Content Credentials (opens in a new tab) is the name and the small icon people see when that record is shown.
Think of it as a sealed note stapled to the file. If someone changes the file in a way the note does not cover, the seal no longer matches.
Networks can read this record, or similar metadata such as the fields defined by IPTC (opens in a new tab), and use it to add a label automatically. This is why a file straight from an AI tool may be tagged without you doing anything.
The limits you should know
- The record can be lost. Re-exporting a file, taking a screenshot or heavy compression can remove it.
- A missing record proves nothing. A file without credentials is not necessarily real, and a file with them is not necessarily honest in every respect. The record says what the tool reported, not whether the picture is a good representation of your product.
- Tools write signals differently. Some editing tools may mark a file as AI-touched even after a small automated fix. That is a common reason a lightly retouched photo gets tagged.
You can check any file yourself with the Content Credentials verify tool (opens in a new tab). Nothing in this article says a small shop must build C2PA itself. It is a record your tools may write, and your job is mostly to avoid destroying it and to know what it says.
What Instagram, TikTok and YouTube ask of you (as of November 2026)
Each network publishes its own rules, and they change often. The table summarizes what the linked pages describe as of November 2026. Re-read the linked page before relying on it. We have deliberately left out numbers, limits and exact menu paths, because those are the parts that move.
| Network | What you are asked to disclose | Automatic labels | Where to look |
|---|---|---|---|
| Instagram (Meta) | Meta describes an "AI info" label and asks people to disclose realistic AI-made content when they share it. | Yes. Meta says it reads industry-standard signals in files. | The posting flow in the app. See Meta's announcement on labeling AI images (opens in a new tab) and its approach to labeling AI content (opens in a new tab). |
| TikTok | TikTok asks creators to label realistic AI-generated content and offers a disclosure setting. | Yes. TikTok says it uses Content Credentials to apply labels. | The post settings before publishing. See TikTok's AI-generated content help (opens in a new tab) and its newsroom post on AI transparency (opens in a new tab). |
| YouTube | YouTube asks creators to disclose realistic altered or synthetic content. | See the help page for what YouTube applies on its own. | The upload details in YouTube Studio. See YouTube's help page on disclosing altered or synthetic content (opens in a new tab). |
Two practical notes:
- The wording on each page matters. "Realistic" is the key word in all three: the networks care most about content a viewer could take for a real event, place or person.
- Pinterest, LinkedIn and other networks have their own settings. If you post there, check their help pages (Pinterest (opens in a new tab), LinkedIn (opens in a new tab)) instead of assuming they match the table.
Which of your product pictures need a label: a decision table
Here is a method that takes a minute per picture. Ask one question: what did the AI change? Then read the action from the table.
| What the AI changed | Example | Action |
|---|---|---|
| Nothing | A photo as shot, cropped and resized | No label needed |
| Light retouch | Dust removal, exposure, color balance | No label needed. Label to be safe if the tool tagged the file or you are unsure. |
| Plain background swap | Studio backdrop replaced with another plain one | Label to be safe |
| A generated scene or setting | Your mug on a generated kitchen counter | Always label |
| A generated person, or motion made from a still | A model wearing your scarf, a still turned into a short clip | Always label |
| The product itself | Changed shape, color, size, texture or details | Do not publish |
The rule of thumb: if a buyer could think a generated scene or person is a real photo, label it. If the product itself was changed, do not publish the picture at all. No label fixes a product that looks different from what you ship.
Three cases walked through
- A candle photographed on a real shelf, dust spots removed and brightness raised. This is light retouch. No label is needed, but if the editing tool marked the file, accept the tag rather than fight it.
- A ceramic mug placed in a generated sunlit kitchen. The scene is made, and a buyer could take it for a real photo. Always label. Also check that the mug's glaze color in the picture matches the real one.
- A scarf on a generated model, where the AI also "fixed" the stitching. Two problems. The generated person needs a label, and the changed stitching alters the product. Regenerate with the product kept faithful, or use the real photo.
A pre-publish routine: five steps you can use today
Do these in order for every AI-assisted picture or clip.
- Record the tool and the prompt for each picture. A line in your content sheet is enough.
- Keep the original product photo. It is your proof of what the product really looks like.
- Export without stripping metadata. If your export dialog has a "remove metadata" option, leave it off for AI-assisted files, and avoid screenshots as a shortcut.
- Check the file in the Content Credentials verify tool (opens in a new tab) to see what it carries.
- Switch on the network's label whenever the decision table says "label" or "always label."
Here is a copyable checklist:
- Tool and prompt noted for this picture
- Original product photo saved
- Exported with metadata kept
- File checked in the verify tool
- Decision table consulted, action noted
- Network label switched on if required
- Product in the picture matches the product we ship
Leave a note for your teammate
Labels get undone when someone else schedules the post and does not know why a setting was on. Save a one-line note next to each post in your content sheet:
AI scene, label onOr, for a plain case:
Real photo, light retouch, no labelIf a network labels a real photo as AI
It happens, usually because an editing tool wrote a signal into the file. Use the appeal or feedback route described in the network's help pages, and keep the original file as proof. Attach or describe what you did in the edit. Next time, check which tool touched the file and whether its AI features can be turned off for simple fixes.
What the EU and Turkish rules say, in general terms
Rules differ by country, and this section only describes the general picture.
The EU. The EU AI Act (opens in a new tab) includes transparency duties for AI-generated or manipulated content. The European Commission's overview of the AI regulatory framework (opens in a new tab) is a good place to see how the rules are organized and when parts apply. In general terms, people who publish such content to a public audience are the ones the transparency ideas are likely to touch. Whether and how a given shop is covered depends on its facts and on the timing of the rules, so check the sources above.
Advertising rules. Consumer-protection rules on misleading advertising apply whatever tool made the picture. A deceptive product image is a problem even if it came from a camera. For the general principle on deceptive ads, see the US Federal Trade Commission (opens in a new tab) and, for Turkey, the Ministry of Trade (opens in a new tab).
The practical takeaway matches the rest of this article: show the product as it is, and be open when a scene or person is generated.
Note
This is general information, not legal advice. For questions about your own shop, ask a qualified professional in your country.
How we do it at SemPost
SemPost marks files made with AI, adds Content Credentials to them, and gives uploads a Made with AI box. You can also mark files later, and each network gets its own label. The details are on the page AI labels on your pictures and videos (opens in a new tab).
Frequently asked questions
Do I have to label every AI-made product picture on Instagram, TikTok or YouTube?
Not necessarily every one, but the networks ask you to disclose realistic AI-made content, and a generated scene or person that could pass for a real photo should always be labeled. A plain photo with light retouch usually needs no label. Check each network's help page before relying on this, because the rules change.
What is C2PA, and does my shop need it?
C2PA is an open standard for attaching a signed record to a picture or video, saying who made it, with which tool and what was edited. Content Credentials is the name and icon for that record. A small shop does not need to build it, but the tools you use may write it, and you should avoid stripping it on export.
Why did my post get an AI label when I only retouched the photo?
Networks can add labels by reading signals in the file, and some editing tools may write an AI signal even for small automated fixes. Appeal through the network's help pages and keep the original photo as proof. Then check which tool touched the file and whether its AI features can be switched off.
Do Content Credentials survive when I upload to a social network?
Not always. The record can be lost when a file is re-exported, screenshotted or compressed, and a missing record proves nothing either way. Export without removing metadata, check the file in the verify tool, and set the network's label yourself when your decision table says so.
Can a label hurt my reach or sales?
We know of no data on this, and no one can promise an outcome. A label on a clearly AI-made scene is usually harmless, while a missing label on a misleading picture risks buyers' trust and returns. Showing the product faithfully matters more than avoiding the tag.