AI in citizen contact: transparent, accessible and accountable
Short answer
If you deploy an AI assistant for residents’ questions, they must know they are talking to AI, and as a public body you also have to meet digital accessibility requirements. Let the assistant answer from approved information, with a source reference and a visible route to an employee. For applications, complaints and personal situations, a human stays necessary.
Residents must know it is AI
Article 50 of the AI Act requires that people interacting directly with an AI system know it is AI, unless that is already obvious to a reasonably well-informed person. That duty has applied since 2 August 2026 and covers every chat or voice assistant on your website.
In practice: state it at the start of the conversation, do not give the assistant a human name and photo, and make it visible how someone reaches an employee. See An AI chatbot for customer service.
For public bodies, accessibility is not an extra
Public sector websites and apps fall under the European Web Accessibility Directive, implemented in national law and built on the EN 301 549 standard. A chat window is part of the website, so it falls under those rules and belongs in your accessibility statement.
- operable with a keyboard, without a mouse;
- readable by a screen reader, with clear labels and status messages;
- sufficient contrast and scalable text;
- no time limit that puts people under pressure;
- an equivalent alternative: phone, desk or email.
Answers that hold
| Type of question | Approach |
|---|---|
| Opening hours, fees, procedures | Answer from approved information, with a source reference |
| Status of an application | Only after reliable identification, otherwise refer on |
| Personal situation, debt, care | Straight to an employee; no automated advice |
| Complaint or objection | Register and pass on, do not handle |
| Emergencies and safety | Clear referral to the right number |
Let the assistant draw on your own current pages and by-laws. See AI on your own documents.
Measuring and adjusting
Do not measure the number of conversations, but whether people were helped: how often a phone call follows anyway, how often a conversation ends without an answer, and which subjects produce the most corrections. Review that list monthly with the contact centre; that is your improvement backlog.
With werqly the assistant runs on a node of your own organisation. Residents’ questions stay in your own infrastructure, the audit log shows which source belonged to which answer, and you set how long conversations are kept. See AI for local government and Generative AI in local government.
Frequently asked questions
Questions on this topic
Do we have to disclose that our chatbot is AI?
Yes. Since 2 August 2026, people interacting directly with an AI system must know it is AI, unless that is already obvious. State it at the start of the conversation and do not give the assistant a human name or photo.
Does digital accessibility apply to a chatbot?
Yes. A chat window on a public sector website falls under the accessibility rules implementing the European directive. It must be usable with a keyboard and a screen reader, and be mentioned in your accessibility statement.
May the assistant say anything about a pending application?
Only after reliable identification and with a connection to the case system using that person’s permissions. Without that, referring to an employee is the safe route.
What do we measure to know whether it works?
How often people call or visit anyway, how often a conversation ends without an answer, and which subjects produce the most corrections. That says more than the number of conversations.
Sources
- Directive (EU) 2016/2102 on the accessibility of public sector websites and mobile apps
- Artificialintelligenceact.eu — Article 50: Transparency obligations
This article is general information, not legal advice. Rules and guidance change; check the sources listed or ask your lawyer or DPO if in doubt.