A digital citizens' office answers citizens' questions around the clock via chat. The AI chatbot provides information on required documents, the responsible office, opening hours, and direct access to online services. It only draws answers from sources approved by your municipality. Every answer includes its referenced source. Inquiries requiring a decision are forwarded to the responsible department. This gives your staff more time for matters that require personal interaction.
This article demonstrates how a digital citizens' office works. At the end, you will find a checklist for your requirements specification.
What tasks does a digital citizens' office handle?
Many requests in public services are repetitive. Citizens want to know what they need to bring, who is responsible, and when offices are open. The chatbot answers these questions independently. The table outlines typical use cases.
Request | What citizens ask | What the chatbot provides |
|---|---|---|
Services | How do I register my residence? | Description, required documents, fees, processing time, and link to the online service |
Jurisdiction | Who handles my request? | Responsible authority, department, and address |
Contacts | Who can I reach in the building authority? | Name, role, telephone number, email address, and office hours |
Forms | Where can I find the application? | The appropriate application form or online service |
Events | What is happening this weekend? | Date, location, contact person, and registration details |
The page for the public sector outlines how Mercury.ai supports municipalities and authorities. The article From Online Forms to Digital Dialogues contextualizes the role of chatbots in citizen services.
Why must a citizen chatbot only respond using approved sources?

Citizens rely on information provided by their municipality. A chatbot that supplements responses with general knowledge may state an incorrect deadline or direct users to the wrong department. For this reason, Mercury Intelligence retrieves answers exclusively from the sources approved by your municipality. General knowledge from the internet is excluded. Before every output, the system verifies the facts against these sources. In this way, Mercury.ai significantly reduces the risk of fabricated answers.
Every answer references its underlying source. Alongside the response, the chatbot can display a link to the relevant page on the municipality's website. Your team can trace every answer back to its source and version. If information is not available in the sources, the chatbot states this clearly and refers the user to the responsible department.
How does your municipality keep information up to date?
Your team manages knowledge in the Knowledge Hub. It scans your municipality's website and synchronizes content continuously. Documents from SharePoint, Confluence, or your CMS can be integrated via APIs. When a page changes, the chatbot subsequently responds using the updated information. If a source contains a date, the answer can reference it directly.
The Knowledge Hub displays whether each source was ingested without errors. Your team can build standard workflows directly within the No-Code Studio. User roles and permission levels define who can edit and publish content. Every change is verified by your team in a staging environment first.
What are the limitations of a citizen chatbot?
A citizen chatbot cannot provide legally binding information. It does not replace administrative notices or individual legal counsel. Your team defines the topics for which the chatbot should merely redirect users to the appropriate department. For such inquiries, a standardized dialogue with pre-approved messaging is deployed.
Your team determines which information the chatbot requests. The chatbot hands over conversations to your staff whenever a case requires human intervention. The article What a Chatbot Should Not Answer details which inquiries are best escalated.
How does the digital citizens' office reach every resident in your community?
The chatbot can converse in up to 98 languages. Citizens with limited proficiency in German can submit inquiries in their native language. The chat widget complies with WCAG accessibility standards. It operates without an iFrame and announces new responses to screen readers. It adapts seamlessly to both mobile devices and desktop screens.
With Mercury Voice, the same chatbot can also handle phone calls. Citizens who prefer calling receive the exact same information as they would via chat. The article on Voicebots in Customer Service explains how this operates over the phone.
What are the requirements for municipalities regarding data protection and the AI Act?
Mercury.ai processes all data via AWS Frankfurt, ensuring it remains within Germany. The language models are hosted directly by Mercury.ai without calling third-party AI providers. Mercury.ai does not train models on your citizens' data. Your municipality executes a Data Processing Agreement (DPA) with Mercury.ai pursuant to Art. 28 GDPR.
By default, the chat widget identifies the chatbot as artificial intelligence, compliant with Art. 50 of the EU AI Act. As the operator, your municipality receives the necessary EU AI Act documentation. The platform provides built-in defenses against prompt injection, data exfiltration, and unauthorized manipulation of the knowledge base. Administrative access requires two-factor authentication.
The article Demonstrating AI Act Compliance in Chatbot Operations outlines the required operational documentation. Technical details are available in the EU AI Act Security Paper.
What belongs in the requirements specification for a citizen chatbot?
The following checklist summarizes the core criteria municipalities typically define in their requirements.
Hosting in Germany or the EU, with German-language support.
A Data Processing Agreement and full compliance with GDPR and the EU AI Act.
Responses generated solely from approved municipal sources, with clear citations.
Seamless handoff to the responsible authority if the chatbot cannot answer.
Multilingual capabilities with automatic language detection.
Protection against prompt injection, data leaks, and tampering with the knowledge base.
Strict exclusion of data usage for training public models.
Automated updates synchronized with the municipality's website or CMS.
Require concrete evidence from vendors for each requirement. Additionally, request a pilot test using authentic inquiries from your own citizen service. Further review points regarding data privacy and the AI Act can be found in our guide to GDPR- and AI Act-Compliant Chatbots.
Frequently Asked Questions
Does the chatbot replace the on-site citizens' office?
No. The chatbot handles recurring questions and helps citizens prepare for visits. Appointments, official applications, and personalized consultations remain the responsibility of your staff.
What happens if the chatbot cannot answer a question?
It clearly states its limitation and provides the contact details for the responsible department. If desired, it can route the conversation directly to your team.
How long does implementation take?
Initial deployment typically takes a few weeks. Most of this time is dedicated to content alignment and testing with real-world inquiries.
Can the chatbot distinguish between private citizens and businesses?
Yes. The chatbot can ask up front whether someone is inquiring as a private individual or on behalf of a company, and then tailor its responses accordingly.
Are you planning a digital citizens' office for your municipality? In a 20-minute consultation, we can help you determine the best starting point. Talk to us.

About the Author
Dr. Hendrik ter Horst
Chief Product Officer @Mercury.ai
Dr. Hendrik ter Horst is Chief Product Officer at Mercury.ai. He holds a PhD in Computer Science from the CITEC Institute at Bielefeld University, with a focus on machine information extraction and data processing. For ten years, he has been working on dialogue-based AI systems, overseeing the conception, implementation, and validation of AI-powered applications from initial idea to production deployment.





