For municipalities

How do I help residents find an answer they cannot find themselves?

Answer by vragen.ai

Your municipality's digital front desk

Residents can handle a passport application themselves. Figuring out which support schemes exist when their income stops is a lot harder. That information sits spread across your main site, your subsites and your schemes. Vragen.ai gives residents one entry point. They ask their question in their own words and get an answer from all those sources at once. With the source included and a clear next step.

You are asking an AI assistant from vragen.ai. Answers come from our own content, with the source included. Why we mention this (Dutch)

These organizations already use vragen.ai

Residents will find the answer anyway, or they will ask ChatGPT.

Answer by vragen.ai

Better that the answer comes from you

The information is there. Just not in one place. In the National Action Plan for Web Policy the national government writes about a “proliferation of government websites”. For the average visitor it is unclear where to find information.

Someone who does not find it asks a general AI service instead, and tells it their own situation right away. That service answers with whatever sits somewhere on the web. That is not necessarily your current information, and sometimes it is simply wrong.

Those conversations also happen out of your sight. You do not know what is being asked and you cannot steer anything. Your own front desk changes two things: the answer comes from your sources, and you see what residents actually ask.

Sources: letter to parliament from the State Secretary for the Interior and Kingdom Relations, 1 July 2025, and Staying out of chatbots puts municipalities behind, Binnenlands Bestuur, 11 March 2026.

“I would very much like to play a part in offering sovereign AI, AI from Dutch soil. Where you decide from A to Z what answer comes out of a chatbot.”

Matthias Snoei · founder of vragen.ai and owner of SWIS

Read why we built it this way (Dutch)

Municipalities already have chatbots, and they do not work well.

Answer by vragen.ai

The criticism of municipal chatbots is fair

In early 2026, independent research tested the chatbots of 39 municipalities with simple questions. Not one of them could say when household waste is collected.

Anything that changes from one day to the next went wrong. That is because they are decision trees. Someone once typed those answers in by hand in a separate system, and they go stale the moment the website changes.

64% of the simple questions were answered incorrectly

1 in 10 of the questions were answered correctly by the chatbot itself

“When do you collect household waste this week?” 0 out of 39 correct
“From what time can I vote?” 34 out of 39 wrong
“Are you open all day tomorrow?” 2 out of 39 correct

Source: conversation analysis of 39 municipal chatbots, Digimonitor, December 2025 to February 2026.

Vragen.ai is set up differently. It searches your knowledge first and writes the answer only after that, and for a complicated question it searches a few times over. You manage that knowledge in one place.

So what makes vragen.ai different?

Answer by vragen.ai

Your knowledge, your conversation

Whoever builds the knowledge should stay the owner of the conversation with residents. That is why vragen.ai is not a chatbot but a Conversation Management System. In one place you decide:

which sources take part
which legislation counts
what an answer looks like
which instructions apply
what you do with the insights

That way every conversation stays yours.

Can we make all our websites searchable at once?

Answer by vragen.ai

One entry point across all your sites

You decide which knowledge vragen.ai uses: your main site, your subsites, your schemes and other sources. You add to that list yourself, with a campaign site, a new scheme or the site of a joint arrangement. The resident asks one question and gets one answer, from all those sources at once. Everything stays where it is.

Your sources

Main site
Waste calendar
Work and income
Care desk
Permits
Sports facilities
Public notices
Moving house

My partner has died. What do I need to arrange?

One answer, assembled from the sources that hold the separate pieces, with the source per part and a next step.

The resident does not have to know which site holds the answer. That is exactly the question people get stuck on today.

You can also run several front desks side by side, each with its own sources, its own instructions and its own tone. In vragen.ai such a front desk is called an agent. Someone asking about special assistance needs something different from a business owner with a permit question. And your colleague at the service desk needs something else again. You decide per agent which sources it may use, so internal documents stay internal.

For residents

“When is my waste collected?”
- Waste calendar
- Main site
For business owners

“Which permit do I need for a terrace?”
- Permits
- Public notices
For your own service desk

“What is the processing time for this application?”
- Work instructions
- Internal service catalog

Example of three agents in one environment. The same engine, different sources and a different tone.

The DigiToegankelijk dashboard shows how many websites and apps your municipality is responsible for.

How do we know the answer is correct and current?

Answer by vragen.ai

An answer that is always current

No one types this answer in beforehand. It takes shape the moment a resident asks the question.

It searches by meaning, not by separate words. If someone types “I can no longer pay my rent”, they end up on your page about debt assistance.
The answer comes from your content as it stands right then. Change your opening hours and the answer changes with them.
It knows the day and the time, so “tomorrow” really does mean tomorrow.

Vragen.ai makes nothing up. If the answer is not in your content, the agent says it is not there. Good, current content is therefore a precondition. That limit gives you something too: you see which questions residents ask that you have no answer to.

If a question leans on a text you do not publish yourself, there is LexAgent: an additional source we switch on at your request. Then vragen.ai also consults Dutch law, municipal regulations such as your by-laws, official publications and court rulings. You see the version that applies at that moment, with the source included.

What do we learn from it ourselves?

Answer by vragen.ai

You see what residents actually ask

Every question is a signal. You hear in their own words what residents want to know, which topics keep coming back and where they get no answer. Per theme you see how many questions come in, what feedback visitors give and how the answers score on reliability, relevance and source relevance.

You put those insights to work. Usually it is a matter of adding to or rewriting content. Sometimes it turns out information sits in three places and contradicts itself. That is how you improve your own information step by step, and in the end that is what helps the resident most.

Insight from the conversations Example

Where knowledge is missing

Parking at city hall
Special assistance 2026
Evening opening hours at civil affairs
+9 topics

Questions asked

2,480

Last 30 days ▲ 12%

Most asked themes

Waste and collection days 318
Passport and driver's license 204
Work and income 176

Answer quality

Reliability 96
Relevance 92
Source relevance 78

Example of the analytics environment. The figures are illustrative. Every score comes with an explanation, so you can see why an answer scores lower.

Can we deploy this safely and responsibly?

Answer by vragen.ai

Using it safely in your municipality

The backing you need internally is not something you have to gather yourself.

Hosted in the Netherlands. The infrastructure and your data are on Dutch servers. If you choose an external language model, that runs within the EU. By default vragen.ai stores no personal data and the widget sets no tracking cookies.
Accessible and in plain language. About one in six Dutch adults has difficulty reading. You set the tone and the reading level yourself, and anyone who does not speak Dutch can ask their question in their own language. The widget is built on WCAG 2.1 level AA.
We walk you through the frameworks. The VNG AI Governance framework, AI literacy, the Algorithm Register and the risk classification from the AI Act. For each of these we set out what is yours to do and what we supply.
You do not have to wait for GPT-NL. We take care of the underlying language model and keep it current, so you are not stuck with today's choice. What you build now stays usable.

SWIS, the agency that builds vragen.ai, is ISO 27001 and ISO 9001 certified. Literacy figure: PIAAC 2023, Kohnstamm Institute: 15.9% of 16- to 65-year-olds have low literacy.

Read how we apply AI responsibly within government

Can I see somewhere how this works?

Answer by vragen.ai

Postbus 42, our own answer desk about the Dutch national government

To show what is possible, we built Postbus 42. Inspired by the old Postbus 51, with today's technology. Every answer is based on the information from Rijksoverheid.nl, with the source included. It is a Dutch-language example we set up ourselves and not an official channel of the national government.

Ask it a question yourself and see what comes out. Try Postbus 42 (Dutch)

The Postbus 42 homepage with the Dutch question "Vertel, wat wil je weten?" and the search field — Postbus 42 for real. You ask your question in your own words and get an answer from the information on Rijksoverheid.nl, with the source included and a next step.

Postbus 42 is our own example. Customers run it too. The VNG and the Dutch Association of Audit Offices work with it on their own site, and implementation is under way at the first municipalities.

Can we help shape it?

Answer by vragen.ai

Join our co-creation program

Together with us and with other municipalities you work out how to replace the search function on your municipal site with an AI agent. A learning track with a working product at the end of it.

One use case: vragen.ai on your own website content
Your team builds AI knowledge, from how it works to where it stops
You finish with advice on what is needed to go live
And together we work out how this becomes reusable under the Common Ground principles

That last one matters most to us, and we need input from municipalities for it. What one municipality works out about setup, sources and instructions should be usable for the next one.

Vragen.ai is registered for this as an initiative in the Common Ground portfolio, in the interaction layer, with Deurne as the municipality involved. That portfolio makes initiatives visible and is not a quality mark. We are open about the fact that we are still working out what this should mean: which parts become reusable, which standards we follow and what we make openly available.

There is no fixed enrollment round. Interested? Joint arrangements are welcome too. Let's look together at what fits. Get in touch.

Vragen.ai comes from SWIS, a digital agency in Leiden, the Netherlands. 46 thinkers and doers ready to help.