Publication lifecycle
AI content review: the publication decision, with its evidence.
Marketing teams make and change content with AI. Before it goes out, somebody has to say what was generated or changed, which disclosure applies, what evidence is missing and who approved that exact version. cmpliance turns that decision into a workflow: agents prepare it, your reviewer makes it, and the record keeps the version, the evidence and the rule it rests on.
- Creation historyEvidence agentDeclared and compared
- Disclosure rulesClassification agentLaw, platform, policy
- Missing evidenceNarrative agentListed for review
Recommendation prepared; the missing evidence is listed for the reviewer.
- Three rulebooks apply to one asset: the law, the platform's rules and your own policy. They don't agree, and the record has to say which one drove the decision.
- A label is the visible part. The defensible part is the record: the version, the evidence, the rule and the person who signed off.
- We don't detect whether something was made by AI. We work from what was declared and what can be shown.
The review
What should a content review establish?
Before a campaign asset is published, the reviewer needs five answers.
- What was generated or changed, and what shows that.Evidence agentQuestion for your reviewerDone ·Generated imageEdited copydeclared and compared
- Which rules apply to this asset, this audience, this country and this channel.Classification agentQuestion for your reviewerDone ·LawPlatformPolicykept apart
- What evidence or permission is still missing.Narrative agentQuestion for your reviewerDone ·Likeness permissionlisted for review
- What disclosure is required, or why none is.Classification agentQuestion for your reviewerDone ·AI labelplatform rule applies
- Who approved this exact version, and whether the published version kept that approval.Your callM. Fischer
Manual review: 5 open questions for your reviewer.
cmpliance: 4 answered with evidence, ready for your reviewer’s decision.
The rules
Three rulebooks, one asset
Under Article 50 of the EU AI Act, in force since 2 August 2026, a company that publishes AI-generated or manipulated content must disclose deep fakes, and AI-written text on matters of public interest unless a person reviewed it and holds editorial responsibility.
Article 50 of the EU AI Act
The platforms add their own rules, and they differ.
Your policy may go further than both.
The review keeps the three apart and names the one that applies.
The record
A label is not the record
Content Credentials can show how a file was made and whether it changed since. They cannot show that a claim is true, or that you may use a person's likeness. A visible label can be right and still leave the question of who decided, on what evidence, unanswered. The record answers it.
A label is the visible part.
AI-generated- The versionv3 · sha256:7c2a…19fe
- The evidence4 sources linked
- The rulePlatform rule
- The person who signed offM. Fischer
How it works
Submit the asset. Prepare the assessment. Review the reasoning. Decide on the version. Record the result.
- 01SubmitAssetKey visual + headlineDeclaredGenerated image · AI-edited copyChannelPaid social · EU
Submit the asset and how it was made.
- 02PrepareAgents workingEvidence agentDeclared and comparedClassification agentLaw, platform, policyNarrative agentListed for review
Agents compare versions, find the evidence that is missing and propose an assessment against approved rules, with the law, the platform's rules and your policy kept apart.
- 03ReviewDisclosureAI labelRule that appliesPlatform ruleReviewerM. Fischer
The responsible reviewer sees the reasoning and the sources.
- 04DecideYour callApproveRequest evidenceApproved with AI label · v3Reviewer decision recorded · M. Fischer
The approval, the required disclosure and the evidence are bound to the exact version. A changed asset needs a new decision.
- 05RecordDecision recordRecordACR-2026-0318Versionv3 · sha256:7c2a…19fePublished2026-03-12 14:05 UTCSources4 linkedExport PDF
The published result is recorded.
Text and still images for marketing content distributed in the EU.
Common questions
Does cmpliance detect AI-generated content?
No. The review works from the declared creation history and the evidence that can be shown. Absence of metadata proves nothing on its own.
Does the tool approve content?
No. Agents prepare the assessment. A person accepts it, rejects it or asks for more evidence, and the reason is recorded.
Which content is covered?
Text and still images for marketing content distributed in the EU.
How do we start?
One brand, two channels, a set of your past assets, then a live batch. We agree the scope with you.