Skip to content

AI Image Generation and Design · Copyright, likeness and disclosure · lesson 15 of 18 · 8 min

Labeling, disclosure and provenance

Why disclosure matters

Audiences increasingly want to know when images are AI-generated or significantly altered, and regulators and platforms are responding. Clear disclosure protects trust, keeps you within platform rules, and in some jurisdictions is becoming a legal requirement for certain content.

Platform labeling (check current policies)

Major platforms have introduced AI labeling approaches. As of 2026, common patterns include:

  • Creator disclosure tools: platforms such as YouTube, TikTok, Instagram and Facebook provide ways for creators to indicate that content is AI-generated or materially altered, and require disclosure for realistic synthetic content that could mislead viewers.
  • Automatic labels: platforms may automatically label content when they detect industry-standard provenance signals (for example C2PA Content Credentials or invisible watermarks from AI tools).
  • Ads: ad platforms may require disclosure for AI-altered content in certain categories, especially political and social-issue ads.
  • Penalties: failure to disclose where required can lead to labels being added, content removal, reduced distribution or account penalties.

Policies change frequently — check each platform's current help center before campaigns.

Regulatory developments

  • European Union: the AI Act includes transparency obligations. Providers of generative AI systems must ensure outputs are marked in a machine-readable way, and those who deploy AI to create deepfakes (realistic content depicting people, places or events that could falsely appear authentic) must disclose that the content is artificially generated or manipulated, with some exceptions and lighter rules for evidently artistic or satirical works. These transparency obligations apply from August 2026.
  • Advertising regulators (such as the UK's ASA/CAP and the US FTC) apply existing rules: ads must not mislead. If AI imagery could mislead consumers about a product or endorsement, disclosure alone may not be enough.
  • Other jurisdictions — including China, several US states and countries in the Middle East and South Asia — have introduced or are developing rules on labeling synthetic media or deepfakes. Always check the markets you target.

When to disclose: a practical framework

| Content | Recommended approach | |---|---| | Realistic image of people, places or events that didn't happen | Disclose clearly (and consider whether to publish at all) | | Realistic product scene composite | Disclose if it could mislead; ensure product and claims are accurate | | Clearly stylized illustration or fantasy | Often not required by platforms, but voluntary transparency builds trust | | Minor edits (cleanup, color correction, background extension) | Usually not required; follow platform guidance | | AI avatars or synthetic voices of real people | Disclose; ensure consent | | Political, social-issue or health content | Follow strict platform/ad rules; disclose |

When in doubt, disclose.

How to disclose well

  • Use the platform's built-in tool where available.
  • Add a clear caption note, for example "Image created with AI" or "AI-generated illustration", near the start of the caption, not buried in hashtags.
  • On-image labels for ads or realistic content can be appropriate.
  • Be specific where useful: "Background generated with AI; product photographed in studio."
  • Keep provenance metadata (Content Credentials) intact.

Disclosure and advertising disclosure together

Creators often need two disclosures in the same post: a paid partnership disclosure (#ad, "Paid partnership") and an AI disclosure. Both must be clear. One does not replace the other.

Client and internal policies

Many brands now have AI policies specifying when AI imagery is allowed, approval requirements and disclosure wording. Ask clients for theirs; if they have none, offer a simple one:

Simple AI imagery policy (starter)
1. AI imagery allowed for: illustrations, backgrounds, concepts, edits
2. Not allowed for: real customers/testimonials, product results, news-like content
3. Real people only with written consent
4. Disclosure: platform tools + caption note for realistic AI content
5. Provenance metadata kept; prompt logs retained
6. Review: all AI assets checked by a second person before publishing

Worked example: a creator's sponsored post

A creator promotes a travel booking app with a dreamy AI-generated image of a destination at sunset. The image is realistic.

Correct approach: paid partnership label and "#ad" at the start of the caption; the platform's AI label turned on; caption note "Scene created with AI — real photos from my trip in the next slides"; include real trip photos to avoid misleading followers about the destination.

Common mistakes

  • Hiding AI disclosure in hashtags.
  • Stripping provenance metadata.
  • Assuming stylized content never needs disclosure.
  • Using AI disclosure in place of #ad or vice versa.

Content Credentials and C2PA in practice

Content Credentials are a way of attaching tamper-evident provenance information to a file, based on the open C2PA (Coalition for Content Provenance and Authenticity) standard. A credential can record which tool created or edited an asset, when, and whether AI was used. Current practice:

  • Generators: OpenAI says its image tools add C2PA metadata; Google says images from its latest Gemini image models in the Gemini app, Vertex AI and Google Ads carry C2PA metadata as well as its invisible SynthID watermark; Adobe Firefly attaches Content Credentials.
  • Editors: Photoshop and other Adobe apps can attach Content Credentials on export; some cameras and phones can sign photos at capture.
  • Platforms: Meta shows an "AI info" label when it detects industry signals (C2PA and IPTC metadata, invisible watermarks) or when a creator self-discloses; YouTube requires creators to disclose realistic altered or synthetic content under "Altered content" when uploading; TikTok and LinkedIn read or display Content Credentials. Behavior differs by platform and changes often.
  • Limits: metadata can be stripped by screenshots, some exports and some uploads. Credentials show history; they do not prove an image is "true", and their absence proves nothing.

Hands-on: check and keep provenance

  1. Export a finished AI-assisted asset from your editor with Content Credentials enabled (in Adobe apps this is an export option; other tools vary).
  2. Inspect it with a Content Credentials verification tool (for example the public Verify tool on the Content Credentials website). Note what it shows: tool, edits, AI use.
  3. Upload a test copy to a private or test account on each platform you use and check whether a label appears automatically.
  4. Record the result in your disclosure log so the team knows which platforms read credentials.
DISCLOSURE LOG — asset_id | platform | realistic? | AI elements |
credentials kept? | auto-label seen? | manual label switched on? |
caption wording | #ad needed? | approved by

Before/after caption wording

| Before | After | |---|---| | "Dreamy sunset vibes 🌅 #ai #travel #ad" (disclosures buried in hashtags) | "Ad — paid partnership with [brand]. Sunset scene created with AI; real trip photos in slides 2–5." + platform AI label on + paid-partnership tool on | | "New collection out now!" on a realistic AI model wearing the clothes | "New collection — model images are AI-generated; see real photos on our product pages." (or better: shoot the real garments) |

Summary

Follow platform labeling tools and policies, track regulations such as the EU AI Act's transparency rules, disclose realistic synthetic content clearly, keep provenance data, combine AI and advertising disclosures, and adopt a simple AI imagery policy.

Video lecture: Labeling, disclosure and provenance

Lecture coming soon · 11 chapters · about 9 minutes. Read the full transcript below.

  1. Labeling and disclosure
  2. Why disclose
  3. Content Credentials (C2PA)
  4. Platform labels (check current policies)
  5. The law
  6. When to disclose
  7. Worked example 1: a stylized nook illustration
  8. Worked example 2: a sponsored travel post
  9. Watch me do it: checking provenance
  10. Common mistakes
  11. Recap and try this now

Lecture transcript

Labeling and disclosure

Picture a travel creator posting a breathtaking sunset over a beach. Thousands of likes. Then comments start. That place doesn't look like that. Is this AI? Suddenly the conversation isn't about the destination or the sponsor. It's about trust. And the platform might have already slapped its own label on the post, because the image file carried provenance data the creator didn't know about. In this lecture, you'll learn why disclosure matters, how platform labels and Content Credentials work, what regulations now require, and exactly how to word a disclosure so it's clear without killing your post. By the end, you'll have a disclosure framework and a log you can use on every campaign.

Why disclose

Why does disclosure matter? Three reasons. Trust, because audiences increasingly want to know when images are generated or significantly altered, and discovering it later feels like being tricked. Rules, because platforms require labels for realistic synthetic content, and some laws now do too. And resilience, because honest disclosure protects you when a post goes viral. Think of disclosure like a nutrition label. Nobody refuses to eat a snack because it has a label. They get upset when they find out what was hidden. A clear label up front turns a potential scandal into a non-event.

Content Credentials (C2PA)

Let's understand Content Credentials, because they now drive a lot of automatic labeling. Content Credentials attach tamper-evident provenance information to a file, based on an open standard called C2PA. They can record which tool created or edited an image, when, and whether AI was involved. Major generators now add provenance. OpenAI says its image tools add C2PA metadata. Google says images from its latest Gemini image models carry C2PA metadata plus an invisible watermark called SynthID. Adobe Firefly attaches Content Credentials, and Photoshop can add them on export. Here's the key idea. Credentials show history. They don't prove an image is true, and missing credentials prove nothing, because metadata can be stripped by screenshots and some uploads.

Platform labels (check current policies)

Now the platforms, and remember, policies change often, so check each help center before a campaign. Meta shows an AI info label when it detects industry signals like C2PA or IPTC metadata and invisible watermarks, or when a creator self-discloses. YouTube requires creators to disclose realistic altered or synthetic content through the altered content setting at upload, and may add a label itself if creators don't. TikTok and LinkedIn read or display Content Credentials. Ad platforms have stricter rules, especially for political and social issue ads. And the penalties for not disclosing where required range from labels being added for you, to reduced distribution, removal, or account penalties.

The law

What about the law? In the European Union, the AI Act's transparency obligations apply from August twenty twenty-six. Providers of generative AI must mark outputs in a machine-readable way, and people who deploy AI to create deepfakes must disclose it, with lighter rules for evidently artistic or satirical work. The European Commission has been developing a code of practice on marking and labeling AI content to support this. Advertising regulators, like the ASA in the UK and the FTC in the US, apply existing rules: ads must not mislead, whatever tool made them. And other markets, including China, several US states, and countries in the Middle East and South Asia, have introduced or are developing synthetic media rules. Check every market you target.

When to disclose

Here's a practical framework for when to disclose. Realistic images of people, places or events that didn't happen: disclose clearly, and ask whether you should publish at all. Realistic product scene composites: disclose if they could mislead, and make sure the product and claims are accurate. Clearly stylized illustration or fantasy: often not required by platforms, but voluntary transparency builds trust. Minor edits like cleanup, color correction or background extension: usually not required, but follow platform guidance. AI avatars or synthetic voices of real people: always disclose, with consent. And political, social issue or health content: follow the strictest rules. When in doubt, disclose.

Worked example 1: a stylized nook illustration

Worked example one, simple. A home decor shop posts a clearly illustrated, watercolor-style image of a cozy reading nook to announce a new cushion collection. Is disclosure required? It's stylized, and nobody would think it's a photo, so platforms generally don't require a label. But the shop adds a short line in the caption, illustration created with AI, cushions photographed in our studio, because it helps customers and sets expectations. Then, for the product carousel, they use real photos. The illustration drew attention. The real photos drove sales. And the disclosure cost them nothing.

Worked example 2: a sponsored travel post

Worked example two, a creator's sponsored post, with illustrative details. Mariam is a travel creator in Dubai promoting a booking app. Her hero image is a dreamy, realistic AI-generated sunset over the destination. Before, her caption was sunset vibes, with hashtag AI and hashtag ad buried at the end. After, it starts with, Ad, paid partnership with the brand. Sunset scene created with AI, real trip photos in slides two to five. She switches on the platform's paid partnership tool and its AI label. And she keeps the image's Content Credentials. Two disclosures, one for the sponsorship and one for AI. Neither replaces the other, and both are at the start, where people actually read.

Watch me do it: checking provenance

Watch me do it. I'll check provenance on a finished asset. I export the final ad from my editor with Content Credentials switched on. Then I open a Content Credentials verification tool in my browser and drop the file in. It shows the tool, the edit history, and that generative AI was used for the background. Next, I upload a test copy to a private account on the platform we're using, and check whether an AI label appears automatically. It does on one platform, not on another. I record both results in the disclosure log, and I switch on the manual label for the platform that didn't detect it. Now the whole team knows what to expect.

Common mistakes

Common mistakes. Hiding AI disclosure in a pile of hashtags. Stripping provenance metadata, for example by screenshotting a file instead of exporting it. Assuming stylized content never needs disclosure, when context matters. Using an AI note in place of the ad disclosure, or the other way around. Relying only on automatic labels, which don't always trigger. And having no policy, so every designer decides differently. Offer your clients a simple AI imagery policy: what AI can be used for, what must be real, consent rules, disclosure approach, provenance and prompt logs, and a second-person review before publishing.

Recap and try this now

Let's recap. Disclose realistic synthetic content clearly, at the start of the caption and with the platform's own tools. Keep Content Credentials intact, but remember they show history, not truth, and can be stripped. Know the rules: platform policies, EU AI Act transparency from August twenty twenty-six, and advertising law that says ads must not mislead. Use two disclosures when needed, one for sponsorship and one for AI. Try this now. Pick three AI images you've made or plan to publish. For each, decide whether disclosure is needed using the framework, write the exact caption wording, check the file in a Content Credentials verification tool, and fill in a disclosure log row.

Video transcript

Here's a simple rule for AI imagery in 2026: if people could reasonably believe it's real, and it isn't, tell them. Disclosure matters for three reasons. Trust, because audiences feel deceived when they discover undisclosed synthetic content. Platform rules, because the major platforms ask creators to label realistic AI-generated or materially altered content, and they may label it automatically when they detect provenance signals like Content Credentials. And law, because regulations are catching up. In the European Union, the AI Act's transparency rules, including disclosure of deepfakes, apply from August 2026, and other countries are introducing their own rules on synthetic media. So how do you disclose well? Use the platform's built-in AI label when it exists. Add a clear note near the start of your caption, like "Image created with AI" or "Background generated with AI, product photographed in studio." For ads or very realistic content, an on-image label can help. And keep the provenance metadata that tools attach — don't strip it out. If you're a creator doing paid work, remember there are two separate disclosures: one for the paid partnership, one for AI. You may need both, and neither replaces the other. And here's the important caveat. Disclosure doesn't fix a misleading claim. If an AI image exaggerates what a product does, or shows a hotel room that doesn't exist, a label won't make it acceptable. Use real imagery for real claims. Finally, write it down. A simple AI imagery policy — what's allowed, what's not, how you disclose, and who reviews — keeps you and your clients consistent.

Key takeaways

  • Disclose realistic synthetic content clearly at the start of the caption and with each platform's built-in tools.
  • Content Credentials (C2PA) record provenance; major generators and editors add them, but metadata can be stripped and does not prove truth.
  • Platforms such as Meta, YouTube, TikTok and LinkedIn read signals or require self-disclosure — test uploads and log results.
  • EU AI Act transparency obligations apply from August 2026; advertising law still requires ads not to mislead.
  • Sponsorship disclosure and AI disclosure are separate; one never replaces the other.

Try it

Choose three AI images you have made or might publish. For each, decide whether disclosure is needed using the framework table, write the exact caption wording, and note any platform label you would switch on.