Short answer: a workable EU AI Act compliance strategy for a small or mid-sized website fits on one page. Name an owner, list where AI is used, decide what each use requires, put the disclosures in place, write down your decisions, and review them on a fixed schedule. You don’t need a governance framework built for banks. You do need something more deliberate than “we’ll deal with it if someone complains”, because the transparency rules have applied since 2 August 2026.
The strategy in six parts
1. Make one person responsible
Pick an owner: the site owner, marketing lead or web manager. They don’t need to be a lawyer. Their job is to keep the AI inventory current, approve new AI tools, and know when to ask for legal advice. Agencies should name an owner per client.
2. Map your AI use
List every AI feature, plugin, script and type of AI-generated content on the site, plus AI tools used internally to produce content. Our website AI audit guide has a checklist and template. Expect to find more than you thought.
3. Classify each use
For each item, decide:
- Role: deployer (you use it) or provider (you built it and run it under your name). See responsibility for third-party AI tools.
- Risk level: banned, high-risk, transparency, or minimal. Almost all website AI is transparency or minimal. Anything touching hiring, credit, insurance or education deserves a closer look at high-risk rules.
- Specific duty: chatbot notice, deepfake label, public-interest text label or review, or nothing.
4. Set your content policy
This is where most websites need an actual decision. Write down, in a few lines:
- Which AI tools staff may use, and what they must not paste into them (client data, personal data, confidential information).
- For AI-written text on public-interest topics: do you review and take editorial responsibility, or do you label? Who reviews, and how is it recorded?
- For AI images: no photoreal images of people, places or products that would mislead, or a visible label if you use them.
- Whether you add voluntary “AI-assisted” labels for trust, and the wording.
Our Article 50 content test explains the legal line.
5. Implement the disclosures
Chatbot greetings, labels on images, notices on articles, and the placement and wording that make them clear and accessible. See where to place AI disclosures and how disclosures affect user experience.
6. Document, train, review
- Document: keep a dated record of each decision, including “no label needed, because…”. This is your evidence if an authority, client or journalist asks.
- Train: Article 4 requires organisations to support the AI literacy of people using AI on their behalf. A one-page internal guide and a 30-minute session cover most small teams.
- Review: every quarter, and whenever you add an AI tool, change a workflow or the rules change.
A 90-day plan
| When | What |
|---|---|
| Week 1 | Name the owner. Run the AI audit. Fix any chatbot that doesn’t say it’s an AI. |
| Weeks 2–3 | Agree the content policy. Label or replace misleading AI images. Review or label public-interest AI articles, newest and most visited first. |
| Week 4 | Update the privacy notice for AI vendors. Check data processing agreements and switch off vendor training on your data where possible. |
| Month 2 | Write the internal AI guide and brief the team. Add “uses AI?” to the publishing checklist. |
| Month 3 | First review: re-check new content and tools, update the record. |
Key dates to plan around
- Since 2 February 2025: bans on prohibited AI and the AI literacy duty.
- Since 2 August 2026: Article 50 transparency rules for chatbots, deepfakes and AI text.
- 2 December 2026: AI tools already on the market must mark their output in a machine-readable way. That’s mainly a vendor task, but expect more metadata on AI images.
- 2 December 2027 and 2 August 2028: high-risk obligations.
Common mistakes
- Labelling everything “AI-generated” to be safe, which hurts trust and doesn’t replace actual review.
- Assuming the vendor handles everything. Deepfake and text duties are the deployer’s.
- Treating compliance as a one-off project. New plugins arrive constantly.
- Keeping no record, so a sound decision can’t be shown later.
On WordPress, the free EU AI Act Ready plugin covers steps 5 and 6 for content: a disclosure level on every post, notices only where needed, AI image labels, chatbot notices and an exportable record of your decisions. For a worked example of the whole strategy on a real-world site, read a practical example of EU AI Act compliance.
Frequently asked questions
Do I need a lawyer to build this strategy?
Not for a typical website. Get legal advice if you might be in high-risk territory, if you build your own AI systems, or if you publish a lot of public-interest content.
How much documentation is enough?
For a small site, one table listing each AI use, the decision and the date, plus your content policy. That’s usually enough to show you took the rules seriously.
What about fines?
Up to 15 million euros or 3% for Article 50 breaches, with the lower amount for SMEs, and always proportionate. See fines and penalties explained.