Label AI-generated content

Why label AI-generated content?

Your audience has a right to know when the content they are reading was created or significantly shaped by AI. This is the principle behind Article 50 of the EU AI Act, which introduces transparency obligations for organizations publishing AI-generated or AI-modified content for public audiences. Before we go any further, two ideas are worth separating here:

• Marking is machine-readable: watermarks or metadata that travel with the content so software can detect that AI was involved. It applies primarily to the organizations that build AI systems.
• Labeling is what a person sees: a visible note on the published page. It applies primarily to the organizations that publish content. This lesson focuses on labeling.

The Act provides an exemption for content that has undergone human editorial review, specifically when an individual assumes editorial responsibility for the publication. It is up to your organization and legal team to assess whether this exemption applies to your content and operations. That said, establishing a labeling practice now is sensible regardless. It gives your team visibility into how widely AI is used across your content operations, and it prepares you for a regulatory landscape that is likely to get more specific, not less.

Model labeling as structured content

Kontent.ai stores your AI label as structured content and delivers it through the Delivery API. Displaying it to your audience, the exact wording shown, the placement, whether you use the EU Commission’s icons, is implemented in your front end, where you control how each channel presents it. That separation is deliberate. Your website, your mobile app, or a partner who republishes your content may each need to present the label differently, so keeping the label as structured data rather than a fixed presentation means you only need to model it once. This same separation gives you room for scope questions that aren’t settled yet, such as: whether disclosure belongs to an individual element, a content item, a page, or the whole site. Modeling your labeling as clean, structured content means you can adjust how it’s presented as needed.

Choose your approach

Kontent.ai’s flexible content model lets you implement AI content labeling in a way that fits your project. Two options work well here: a multiple choice element or a taxonomy group.

Multiple choice element

With multiple choice element, you define the options directly in your content type. It’s best suited for projects with a single content type that requires AI labeling, or for situations where you don’t need to filter or report on AI-generated content regularly. Advantages:

• Quick to set up – Define the options once in your content type, and you’re done.
• Immediately visible – When Aiko Agent works with the content type, it will see all available options without needing to reference anything external.
• More predictable behavior – Aiko tends to select an appropriate value more consistently than when relying on a referenced taxonomy.

Considerations:

• No filtering in content inventory – You can’t filter your content items by multiple choice values in the UI, making it difficult to find labeled content for auditing or reporting purposes.
• Not reusable – If you need the same AI disclosure categories for other content types, you’ll need to recreate the multiple choice element for each one.
• No centralized updates – If labeling requirements change and you need to add a new category, you need to update each content type individually.

Taxonomy group

A taxonomy group defines a reusable set of terms that can be referenced by multiple content types. Best suited for organizations that need to track, filter, and report on AI-generated content across multiple content types, especially when compliance auditing is important. Advantages:

• Filtering and reporting – You can easily filter content items in the content inventory by taxonomy terms, making auditing and compliance reporting straightforward.
• Reusable across content types – Create the taxonomy once and use it in Articles, Blog posts, Case studies, or any other content type that needs AI labeling.
• Centralized management – Update the taxonomy in one place, and changes apply everywhere it’s used.
• Hierarchical categorization – If your disclosure requirements become more nuanced, such as “AI-assisted > with human review” vs. “AI-assisted > without human review”, taxonomies support nested structures.

Considerations:

• Requires explicit instruction for Aiko to work – Since the content type only stores a reference to the taxonomy group, Aiko needs to fetch the taxonomy before it can see the available terms. This only happens when Aiko’s specifically instructed to work with the element. If you choose taxonomy, include the labeling step explicitly in your prompt and check the results.

Make labeling work reliably

Regardless of which element type you choose, proper labeling requires more than adding an element to your content type. These reliability principles apply whether you’re labeling for compliance, internal content tracking, or other organizational needs.

1. Make the element required

The most fundamental safeguard is making your AI labeling element required. This prevents any content from being published without a categorization set. In Kontent.ai, you configure this when setting up your element. A required element ensures that:

• Content items cannot move to the Published workflow step without a value.
• Aiko cannot complete content creation and publish the item without addressing this element.
• Human contributors need to make a conscious decision about their involvement from the editorial and review perspective.

A required element guarantees that something is set, but doesn’t guarantee the correct value is chosen. That’s where the other mechanisms come in.

2. Add clear guidelines

Use a Guidelines element in your content type to explain to both human contributors and Aiko how to make the right call. Place this near your AI labeling element for maximum visibility. Example guidelines: “EU AI Act compliance – AI disclosure required?

• Yes - Select if AI generated substantial parts of the content and a human did not fully validate, edit, and take responsibility for the final output.
• No - Select only if a human has meaningfully reviewed, validated, and assumes full editorial responsibility for the final output.”

Clear criteria reduce ambiguity and help ensure consistent categorization across your team.

3. Establish AI instructions

The Guidelines element provides reference material that both humans and Aiko can see, but it doesn’t automatically enforce behavior. To ensure Aiko actively applies these guidelines, you need to give it explicit instructions during your conversation. For example: “When creating or modifying Article content items, refer to the AI disclosure guidelines and set the appropriate value based on the criteria provided. Always confirm your decision with me before publishing.” This transforms Aiko from a passive tool that waits for complete instructions into an active participant that helps ensure compliance.

4. Add a workflow step with human review

If your workflow includes a human review step, make AI disclosure part of the reviewer’s checklist. Before approving content for publication, the reviewer can verify that the label is accurate and reflects how the content was actually created. This keeps your workflow lean while still ensuring a human has signed off on the labeling. For organizations where AI-generated content requires more focused attention, consider adding a dedicated workflow step for reviewing AI-generated content as a whole. This gives reviewers a dedicated space to assess quality, accuracy, and compliance, including verifying the disclosure label, before content moves forward. If you’re relying on the human review exemption, it helps to record who actually reviewed the content, not just that the workflow moved forward. Add an editorial owner element naming that person: a text element or a linked item element if your project already models authors as content items.

5. Combine multiple safeguards

The most reliable approach uses layered safeguards:

• Required element prevents publishing without a value
• Guidelines explain how to choose correctly
• Conversational instructions ensure Aiko proactively addresses labeling
• Human review ensures quality, accuracy, and correct labeling before publication

No single mechanism is perfect, but together they give you a solid foundation to build on as your team’s AI usage grows and regulatory requirements continue to take shape.

Know what Kontent.ai can’t detect