Transparency
Open Parliament makes the work of the Chamber of Representatives readable. We use artificial intelligence to help you search it, never to write in place of the official documents.
Page updated on .
AI helps us prepare some texts, such as the plain-language summaries of votes. A team member reviews them before publication. The assistant only answers when you ask it something, and you always know it.
The assistant cites the official document behind every claim. The Chamber’s documents are authoritative, not the AI.
Attendance, votes, questions and speeches are calculated with fixed, public rules that are the same for everyone.
| On the site | AI? | How it is produced |
|---|---|---|
| The assistant’s answers | Yes | A language model writes the answer from our database, live, for you alone. |
| Search “by meaning” in the debates | Yes | A model ranks speeches by how close they are to your question; it always returns the original text. |
| Plain-language titles and summaries of votes | Partly | Prepared with the help of AI from the official title alone, then reviewed and approved by a team member before publication. Each vote links to the Chamber’s official file. |
| Member profiles and statistics | No | Calculated with public rules each time the data is updated: for example, attendance is the share of roll-call votes the Member took part in while sitting. |
| Reports, votes, written questions | No | Taken from the Chamber’s official publications, with their document number. |
| Data open to other assistants (connector) | No | We provide the data; the answer is produced by the AI tool you choose, under its publisher’s responsibility. |
A Member’s party is always the one they belonged to on the day of the event. No overall score, hidden ranking or secret weighting is used.
We have neither developed nor trained any model. We use two existing services; if either changes, this page will be updated.
Anthropic, a Claude family model
It reads your question, queries our database read-only, then writes an answer. Its instructions: rely only on this data, give the source and date of every claim, say when it does not know, stay neutral and answer in your language. While it works, you see the searches it runs.
Google, Gemini Embedding
It turns each public speech and your search into a digital fingerprint of their meaning, so a passage worded differently can still be found, in French as in Dutch. It writes no text.
Even fed with official data, an AI can be wrong: misunderstand the question, leave something out, summarise a debate badly or confuse what a vote was about (an amendment and the text it amends, for example).
What is covered. The Chamber of Representatives (federal level), since the start of the legislature in July 2024. The Senate, the regional parliaments and earlier legislatures are not. Written questions are available up to March 23, 2026. The figures on the site are updated regularly from the Chamber’s publications; they currently run up to October 22, 2026.
No account, no history kept: your conversations stay in your browser for the duration of your visit and are not stored in our databases.
The parliamentary data concerns the public activity of office holders in the exercise of their duties. Details in our privacy policy.
The EU Artificial Intelligence Act (EU) 2024/1689 classifies uses by their risk. An assistant that gives access to public documents falls under the transparency obligations, which apply since 2 August 2026.
| What the Act requires | Reference | What we do |
|---|---|---|
| Tell people they are interacting with an AI | Art. 50(1) | A visible notice in the assistant’s field, with a link to this page. |
| Mark AI-generated content | Art. 50(2) | Answers are presented as those of the AI assistant and carry a machine-readable marking. |
| Disclose AI texts published to inform the public | Art. 50(4) | Vote titles and summaries prepared with the help of AI are reviewed and approved by a team member before publication; the team holds editorial responsibility for them. |
| Do not use prohibited practices (manipulation, social scoring, profiling…) | Art. 5 | None: no visitor profiles, no scores for Members, no incitement to vote. |
| Strictly govern high-risk systems, including those intended to influence a vote | Annex III, point 8 | Not applicable: the tool is not designed to influence an election or voting behaviour. |
| Train the people who operate the AI | Art. 4 | The team is trained in the uses and limits of the tools used. |
The obligations that concern the models themselves (documentation, copyright, safety) lie with their providers, Anthropic and Google. We remain responsible for how we use them.
Open Parliament is an information tool. It is neither an official publication of the Chamber, nor a legal service, nor political advice.
To the extent permitted by law, we cannot be held liable for damage resulting from the use of the site or of an answer from the assistant, except in the case of intentional or gross misconduct on our part.
The assistant is provided free of charge to learn about the Chamber’s work. By using it, you agree not to:
We may limit, suspend or block access to the assistant in case of misuse, without notice. For automated use of the data, contact us.
Origin. The data comes exclusively from the official, public publications of the Chamber of Representatives (lachambre.be / dekamer.be): reports, roll-call votes, written questions and the register of Members. Each item keeps its reference to the original document. The site is neither affiliated with nor endorsed by the Chamber.
Members and ministers. We process information about public office holders as the Chamber itself publishes it: speeches, votes, questions, group membership, mandates, official portrait, date and place of birth. This processing serves a purpose of general interest, the transparency of parliamentary work, and relies on legitimate interest (Article 6(1)(f) GDPR). We collect no other data about them.
Your rights. Anyone concerned can ask for access to the data about them, its correction if it does not match the official source, or object to its processing on grounds relating to their particular situation, at the address below. A proven error is corrected as soon as possible. You can also lodge a complaint with the Data Protection Authority.
Our data can also be consulted from a third-party AI assistant, through a connector. In that case, we only provide the data: the answer is written by that tool, under its own rules and its publisher’s responsibility. We control neither those answers nor what those services do with your questions. How to connect your assistant.
Our own AI providers (Anthropic and Google) act as processors, solely for running the assistant and the search. The obligations linked to the design of their models lie with them. Links to external sites, including the Chamber’s, are given for reference; their content is their publishers’ responsibility.
This page is governed by Belgian law. Any dispute falls under the competent Belgian courts, subject to the mandatory rules protecting consumers.
It describes the tools in use on the date shown at the top of the page. We update it as soon as a tool, a provider or a rule changes, and whenever the legal framework evolves. It complements our legal notice and our privacy policy.
A false or misleading answer, a figure that does not match the official document, a speech attributed to the wrong person: write to contact@opentheparliament.com, with the page or question concerned. Every report is examined by the team, and data errors are corrected at the source, for the whole site.