EU AI Act Compliance Checklist for Content Transparency
An EU AI Act transparency checklist covers five things: knowing your role for each system, training the people who operate them, agreeing accurate disclosure wording, applying machine-readable marking at publication, and keeping independently checkable records of every declaration.
1–2. Roles and literacy
Record, per system, whether you act as provider or deployer — the duties differ and most organisations are both somewhere. Then close the AI literacy obligation: the staff operating AI systems need a level of understanding proportionate to their role. Both are documentation exercises, and both are already applicable.
3–5. Inventory, wording, pipeline
Inventory every channel where AI-generated or AI-modified content reaches the public — including the ones marketing runs outside the CMS. Agree standard disclosure wording per channel that is accurate about what AI did. Then move disclosure into the publishing pipeline. A manual step that depends on someone remembering is a control that will fail on a busy day.
6–7. Marking and durability
Apply machine-readable marking to published assets, and assume it will be stripped somewhere downstream. Pair embedded credentials with a mechanism that survives re-encoding and an external record that can be located afterwards. Test the durability yourself: publish, re-share through the channels you actually use, and check what remains readable at the other end.
8–9. Evidence and rehearsal
Keep records of each declaration outside the system that produced the content, time-anchored so the timing can be shown rather than asserted. Then rehearse the audit: pick a published asset at random and confirm an outsider could verify its declaration end to end without your help. If that rehearsal needs an internal screenshot, the evidence is not audit-ready.
10. Review the claims you make
Last step, and the one most often skipped: read your public statements about AI and authenticity. Any claim that content is 'verified true', 'human-made' or 'guaranteed authentic' goes beyond what a technical mechanism can support. Accurate, narrow claims are both better compliance and better defence.
EU AI Act transparency readiness checklist
- 1 Map roles per system Record whether you are provider or deployer for each AI system in use.
Map roles per system
Record whether you are provider or deployer for each AI system in use.
- 2 Deliver AI literacy training Train staff who operate AI systems, proportionate to their role and context.
Deliver AI literacy training
Train staff who operate AI systems, proportionate to their role and context.
- 3 Inventory AI-assisted channels List every channel where AI-generated or modified content reaches the public.
Inventory AI-assisted channels
List every channel where AI-generated or modified content reaches the public.
- 4 Agree disclosure wording Define accurate, standard disclosure language for each channel and content type.
Agree disclosure wording
Define accurate, standard disclosure language for each channel and content type.
- 5 Automate disclosure Move disclosure into the publishing pipeline rather than relying on manual steps.
Automate disclosure
Move disclosure into the publishing pipeline rather than relying on manual steps.
- 6 Apply machine-readable marking Mark published assets so software can detect the declaration without human interpretation.
Apply machine-readable marking
Mark published assets so software can detect the declaration without human interpretation.
- 7 Test durability Re-share through your real distribution channels and check what marking survives.
Test durability
Re-share through your real distribution channels and check what marking survives.
- 8 Register evidence externally Keep time-anchored records outside the system that produced the content.
Register evidence externally
Keep time-anchored records outside the system that produced the content.
- 9 Rehearse an audit Verify a random published asset end to end as an outsider would.
Rehearse an audit
Verify a random published asset end to end as an outsider would.
- 10 Review public claims Remove authenticity claims that no technical mechanism can support.
Review public claims
Remove authenticity claims that no technical mechanism can support.
Frequently asked questions
Do small companies have to do all of this?
The transparency obligations apply regardless of size, though penalty caps are lower for SMEs and start-ups. The checklist scales down: the steps stay, the effort per step shrinks.
Is a policy document enough to show compliance?
A policy shows intent. Supervisors ask for evidence that disclosures were actually made at publication and can be checked independently.
What if we only use AI internally?
Transparency duties attach to output that reaches people. Purely internal use still triggers the AI literacy obligation for staff operating the systems.
How often should the checklist be reviewed?
At least whenever a new publishing channel or generative tool is introduced, and ahead of each phased application date.
Test your evidence trail
Verify a published declaration the way a supervisory authority would.