Short answer: yes, in most cases. When you add a third-party AI tool to your website, the vendor is the provider of that AI system, and you are its deployer. The EU AI Act gives each role its own duties. Under Article 50, several of the transparency duties that visitors actually notice, such as labelling deepfakes and unreviewed AI text, sit with the deployer. And even where a duty is formally the vendor’s, such as making a chatbot identify itself, the way you configure the tool often decides whether visitors are informed.
This guide explains where your responsibility starts and ends, with examples for the tools most websites use.
Provider vs deployer in plain terms
- Provider: develops an AI system, or has one developed, and places it on the market or puts it into service under its own name. OpenAI for ChatGPT, Tidio for its AI chatbot, a plugin developer for its AI writing feature.
- Deployer: uses an AI system under its own authority in a professional context. A business that embeds that chatbot or publishes that AI-generated content.
A purely personal, non-professional use falls outside the deployer definition. A business website, a freelancer’s portfolio or an agency’s client project does not.
Who is responsible for what
| Duty | Provider (vendor) | Deployer (you) |
|---|---|---|
| Design chatbots to disclose they’re AI (Art. 50(1)) | Yes | Configure it so the disclosure actually reaches visitors |
| Machine-readable marking of generated content (Art. 50(2)) | Yes | No, but don’t deliberately strip it |
| Inform people about emotion recognition or biometric categorisation (Art. 50(3)) | No | Yes |
| Disclose deepfakes (Art. 50(4)) | No | Yes |
| Disclose unreviewed AI text on public-interest topics (Art. 50(4)) | No | Yes |
| AI literacy of staff using it (Art. 4) | For its own staff | For yours |
| Personal data processed by the tool (GDPR) | Usually your processor | Usually the controller |
“The vendor said it’s compliant.” Is that enough?
Not on its own. A vendor’s compliance statement covers the vendor’s duties. It can’t cover choices you make: the bot’s name and greeting you wrote, the AI image you chose for your homepage, or the unreviewed article you published. The Commission’s guidance is explicit on one point that catches many site owners: a deployer can’t rely on the provider’s invisible watermark to meet its own deepfake disclosure duty, because visitors can’t see it.
Common third-party tools and what to check
Chatbot widgets
Most widgets allow you to rename the bot, add a human photo and edit the first message. Check that the greeting or header clearly says “AI assistant”, and that hand-offs to human agents are announced. Details in our chatbot compliance guide.
AI writing plugins and SEO assistants
The plugin is the provider. You decide what to publish. For text on matters of public interest, either have a qualified person review it and take editorial responsibility, or label it. Our AI content disclosure test walks through it.
Image generators
Midjourney, DALL-E, Firefly or a WordPress image plugin: the tool marks its output in metadata, and you decide whether the image you publish is a realistic depiction that needs a visible label.
Auto-posting and content automation plugins
These publish AI content without anyone looking at it, which means there’s no human review to rely on. Public-interest posts from such tools need a label. They also tend to create duplicate, thin pages that hurt search rankings, so review what they’ve published.
Recommendation, personalisation and search tools
Usually no Article 50 duty for you. Consumer law may require you to disclose personalised pricing, and marketplaces must explain ranking parameters. See our ecommerce guide.
Translation tools
Machine translation of your own pages is generally treated like editing support. If you translate public-interest content and publish it without review, treat it like AI-generated text.
When you become the provider
You move from deployer to provider if you develop an AI system yourself, or have it developed, and put it into service under your own name. A typical example is a custom assistant built on an AI model’s API and presented as “our assistant” on your site. The model maker remains the provider of the general-purpose model, but you’re the provider of the system your visitors use. That brings the provider duties, including designing the system so it discloses that it’s an AI. Agencies building these for clients should read what agencies must do under the EU AI Act.
How to manage third-party AI responsibly
- Inventory: list every AI tool on the site, including plugins and scripts. Use our website AI audit template.
- Vendor check: for each tool, note what it does with visitor data, whether it uses your data for training, and what AI disclosure options it offers.
- Configure: switch on built-in AI labels, write honest greetings, and turn off training on your data where you can.
- Add what the vendor can’t: visible labels for deepfakes and unreviewed public-interest text.
- Contracts and privacy: data processing agreement in place, privacy notice updated with the vendor’s name.
- Record: keep a dated note of each decision.
On WordPress, the free EU AI Act Ready plugin covers the deployer side for common tools: AI notices for 15 chat widgets, labels for AI images, per-post disclosure levels and an exportable compliance record. It runs on your own server and needs no account.
Frequently asked questions
If the chatbot vendor breaks the rules, can I be fined?
Fines attach to whoever breaches their own obligation. The vendor answers for its design duties. If your configuration hid the AI from visitors, or you failed a deployer duty, that’s on you. Article 50 breaches can be fined up to 15 million euros or 3% of turnover, the lower of the two for SMEs. See fines and penalties explained.
Does using an AI tool for internal work make me a deployer?
Yes, if it’s professional use. Article 50’s visitor-facing duties only matter when visitors interact with the AI or see its output, but Article 4 AI literacy covers internal use too.
Do I need the vendor’s documentation?
For ordinary, non-high-risk tools, there’s no formal requirement for deployers to hold it. It’s still useful to keep the vendor’s AI and privacy information with your audit record.