AI disclosure rules belong in your CMS, not a spreadsheet

Illustrated infographic summarizing: AI disclosure rules belong in the CMS, not a spreadsheet

By Greg Nowak. Updated 22 August 2026.

AI disclosure is now an operational publishing issue. The EU AI Act’s Article 50 transparency obligations have applied since 2 August 2026, while the associated Code of Practice offers a voluntary route for demonstrating compliance with its marking and labelling requirements.

The important detail for publishers is that the rules are not a blanket requirement to label every sentence touched by AI. Among other cases, deployers must clearly label deepfakes and AI-generated or manipulated text that informs the public about matters of public interest when it has not received qualifying human review or editorial control. For text, substantive human review combined with identifiable editorial responsibility can therefore be decisive.

That is precisely why a spreadsheet is the wrong long-term system. A spreadsheet can record what a team intended to do. It cannot reliably connect the disclosure decision, reviewer, evidence and final content revision at the moment someone clicks Publish.

Treat disclosure as a publishing decision

The useful question is not simply “Was AI used?” AI might have suggested headings, translated an approved source, produced most of a draft or generated a realistic image. Those uses have different editorial and transparency implications.

Your CMS should capture enough context to make a proportionate decision. At minimum, record the role AI played, the type and subject of the content, who performed the substantive review, who holds editorial responsibility, whether disclosure is required and which version was approved.

Do not turn this into an indiscriminate compliance form. Routine spelling help does not need the same path as an unreviewed public-interest article or synthetic video. Use conditional fields and escalation rules so low-risk work remains quick while higher-risk material cannot bypass review.

Publishing situation	Sensible default	Evidence to keep in the CMS
Ideas, structure or light copy-editing	Follow the internal policy; do not trigger a public label automatically	AI role and accountable editor
Substantially generated general website copy	Require an accuracy, originality and usefulness review	Sources, reviewer, material edits and disclosure decision
Public-interest text	Route through substantive human review and editorial responsibility, or apply the required label	Review record, responsible person and final decision
Realistic synthetic or manipulated media	Escalate for disclosure and legal review before publication	Asset provenance, intended context, label and approval
Large batch of similar SEO pages	Pause the batch until its distinct user value is demonstrated	Page purpose, overlap review, source basis and sample approval

A practical risk-based matrix for routing AI-assisted content through the right publishing checks.

Build a small set of fields that people will actually use

Start with an AI role field. Plain options such as none, ideation, drafting, transformation, translation and generated media are more useful than a single yes-or-no checkbox.

Add a review record containing the reviewer, review date and short source notes. A qualifying human review is more than grammar correction: the reviewer needs suitable knowledge and must examine the substance. Instead of asking editors to write an essay about every change, provide a compact checklist covering factual accuracy, source reliability, misleading omissions and whether the final page serves a real audience need.

Then capture the disclosure decision: not required, required and present, or needs specialist review. Store approved label text centrally where possible, so authors do not improvise inconsistent wording. For generated media, retain available provenance metadata when assets pass through resizing, optimisation or a digital asset manager.

Finally, assign an editorial owner. “Human reviewed” is weak evidence when nobody can say who approved the substance and accepted responsibility for publication.

Connect the fields to an enforceable workflow

A workable sequence is Draft → Editorial review → Disclosure decision → Ready to publish → Published. The CMS should prevent the final transition when a required field is missing, a high-risk answer remains unresolved or the content has changed materially since approval.

For WordPress, structured post metadata can hold the governance fields. Surface the important values in the editor and add columns and filters to the Posts screen, allowing an operations lead to find items awaiting review or missing a decision. Many teams can retain the familiar Pending Review status and add validation around publication; more complex organisations may need a dedicated editorial workflow implementation. Enforce critical checks on the server side rather than relying only on browser prompts.

Drupal provides a natural foundation through fields, revisions, Workflows and Content Moderation. Add states such as AI review or Disclosure check only where they improve routing. Transitions and permissions can determine who may approve sensitive content, while a filtered View can provide the editorial and audit queue. Revisions matter because the evidence should identify the version that was actually reviewed.

Include search quality in the same gate

Google does not treat AI assistance as an automatic search violation. Its current guidance focuses on accuracy, quality, relevance and context for readers. The clear SEO risk is scaled content produced mainly to manipulate rankings without adding user value.

Before approval, ask whether the page contributes original information or practical judgment, answers a distinct user need and deserves to exist separately from content already on the site. For batches, review representative samples and check proposed pages for substantial overlap before generating the remainder. A disclosure label cannot rescue thin or misleading content.

This is also useful commercially. Owners gain a visible queue rather than an unverifiable promise. Agencies gain a cleaner approval trail and handover. Editors know what needs attention without maintaining a second system that immediately drifts away from the live page.

Start with the decision model, not the plugin

Before configuring WordPress or Drupal, agree on which uses must be recorded, what counts as substantive review, who can accept editorial responsibility and which cases require specialist advice. The legal position will depend on the content, system and organisation, so the CMS design should implement an agreed policy rather than attempt to replace legal counsel.

Greg can help turn that policy into fields, permissions, review states, admin queues and publication safeguards that suit the way your team already works. If AI-assisted publishing is becoming difficult to oversee, a focused CMS workflow review is a practical place to begin.

Related on GrN.dk

Need help with this kind of work?

Plan a practical CMS workflow with Greg Get in touch with Greg.

Sources

Last modified

2026-08-22