Parliamentary monitoring is not judged by the number of alerts it generates. It is judged by one question: the day your leadership or your client asks for a position on a bill, do you have what you need to write it, and can you show where every statement comes from?
This guide describes a working method, not a tool. It sets out the scope to cover, how to turn that scope into queries, the cadence to hold and the reporting format to choose. It then details the official sources, what they actually publish and where they stop — citing the Règlement de l'Assemblée nationale, the standing orders that set most of these rules and that few practitioners have read.
It is written for public affairs consultants, for in-house and trade-association public affairs leads, and for analysts setting up a monitoring practice for the first time or inheriting a predecessor's. The final part covers automation: the conditions under which it saves time, and the conditions under which it wastes it.
What parliamentary monitoring covers: five materials, five rhythms
The first instinct, when setting up a monitoring practice, is to follow "the bills". That is not enough. Parliamentary monitoring sits at the heart of a broader public affairs monitoring practice, and a legislative file plays out across five distinct materials, which share neither the same publication rhythm, nor the same degree of formality, nor the same signal value.
Bills tell you the state of the law being proposed. Amendments tell you the balance of power. Questions to the Government tell you the political temperature. Hearings give you the arguments and the commitments. The calendar tells you how much time you have left to act. A practice that covers only part of this produces notes that are always one beat late.
- Written questions are the most underrated instrument in parliamentary monitoring: they force the Government to take a position, in writing, on a specific point. The Règlement de l'Assemblée nationale provides that ministers' answers must be published within two months of the question's publication, and that this period runs without interruption (article 135(6)). Where the deadline is missed, group chairs may flag the question in the Journal officiel, and ministers are then required to answer within ten days (article 135(7)). The very existence of that catch-up mechanism tells you the two-month deadline is not always met — so track the questions asked on your topic, not only the answers.
- Questions to the Government are the opposite case: they leave no prior written trace at all. The Instruction générale du Bureau is explicit — they are "ni déposées, ni notifiées, ni publiées" (neither tabled, nor notified, nor published), article 15(II). You cannot anticipate them through any document feed; you discover them at the moment they are put, on the floor. This is the first place where purely text-based monitoring falls behind.
| Material | What you read in it | Publication rhythm | What it tells you |
|---|---|---|---|
| Bills (government and private members', reports) | The state of the law being proposed, article by article, and the version that came out of committee | By stage of the shuttle between chambers, in successive versions | What the debate will be about |
| Amendments | Who wants to change what, on which article, and with which co-signatories | In waves, ahead of each committee then floor reading | The real balance of power on the bill |
| Questions to the Government and written questions | What the Government is willing to say publicly, and to whom | A weekly sitting for oral questions; continuous for written ones | Trade-offs announced before they are written down |
| Hearings and committee proceedings | The arguments, the commitments made in front of members, the positions of the administration | Concentrated in the weeks a bill is examined | What actually decides the content of the bill |
| Committee and floor calendar | Examination slots and amendment tabling deadlines | Set, then revised during the week | How much time you have left to act |
The method: scope, keywords, cadence, reporting
A monitoring practice rests on four decisions, taken in that order. Reversing them — starting with keywords, or worse, with the choice of tool — is the most common reason a practice produces a great deal and serves very little.
- Scope. Write it down before anything else, on a single page: the institutions you follow (Assemblée nationale, Sénat, Journal officiel, regulators, EU institutions), the themes, the named stakeholders, the time horizon. An unwritten scope expands with every news cycle, and the practice ends up covering everything badly rather than a few things well. That document is also what lets you decline an out-of-scope request.
- Keywords. A public affairs topic almost never reduces to a single word. Plan three families of queries: the vocabulary of the provision itself (the terms the legislator uses, often different from your industry's), hard references (bill number, code article, legislative file title) and stakeholders (rapporteur, committee chair, political group). Hard references produce few false positives and form the backbone of a clean practice; vocabulary needs re-reading every quarter, because it shifts as the bill does.
- Cadence. Set it against parliamentary deadlines, not against your own diary. At the Assemblée nationale, members' amendments on the floor must be tabled by 5 p.m. on the third working day preceding the start of the bill's discussion (article 99(1)); in committee, they must reach the committee secretariat by 5 p.m. on the third working day preceding the start of examination (article 86). And at first reading, outside the accelerated procedure, at least ten days must separate the electronic publication of the text adopted by the committee from the start of its examination on the floor (article 86). Those three markers define your useful windows: a daily pass is enough outside examination periods, continuous watch is required in the days before a tabling deadline.
- Reporting. Decide who reads, at what time, and what the reader needs to be able to do with the document. A ten-line daily brief read by the whole team beats a six-page weekly note read by nobody. One rule is non-negotiable: every line carries a link to its source. A note without references cannot be defended in a meeting, and re-checking it costs more than writing it did.
The official sources, and where they stop
All of this material is public. That is what makes parliamentary monitoring possible with no budget — and it is also what makes it expensive: the data is open, but scattered, heterogeneous, and delivered in formats designed for archiving rather than for monitoring.
- The Assemblée nationale open data portal (data.assemblee-nationale.fr). The richest of the four: amendments tabled in committee and on the floor, recorded votes, legislative files, members and bodies, meetings, questions. Published in XML and JSON under the Licence ouverte. Limit: the "Débats" dataset covers the records of public sittings — committee proceedings do not get the same treatment.
- The Sénat open data portal (data.senat.fr). Legislative files since October 1977, amendments in committee since October 2010 and on the floor since October 2001, questions since April 1978. Practical limit: the datasets are distributed "au format PostgreSQL 8.4" — a database to restore, not a feed to consume. Substantive limit: the records dataset covers the full verbatim records of public sittings since January 2003. Neither committee proceedings nor video.
- Légifrance and the Journal officiel. The reference for law as published: consolidated texts, the JORF, implementing decrees. Programmatic access runs through the Légifrance API exposed on the PISTE platform, with account creation and an OAuth 2.0 token obtained through the Client Credentials flow — a small integration project, not a feed you plug in over an afternoon.
- Regulators and EU institutions. Each has its own portal, calendar and format: consultations, opinions, decisions, files under examination in Brussels. Nothing is hidden; nothing is connected either. None of these portals knows that an amendment tabled at the Assemblée nationale bears on the same provision as a consultation a regulator opened that same month.
Committees and hearings: where it is decided, where monitoring falls behind
On most bills, the substance is settled in committee. That is where the text is rewritten before the floor stage, where the rapporteur arbitrates, where the administration and stakeholders are heard. Positions defended in a hearing announce the amendments of the following weeks — it is the most predictive material a public affairs professional has.
These proceedings are public: committee work is public under article 46(1) of the Règlement de l'Assemblée nationale. But public does not mean trackable. The standing orders themselves explain why, by treating the floor and the committee in markedly different ways.
- The written record. For public sittings, a full verbatim record is drawn up for every sitting and published in the Journal officiel (article 59(2)). For committees, the obligation is looser: at the end of each meeting, a record is published setting out the committee's proceedings, its votes and the statements made (article 46(3)). A record — not a full verbatim record. The Conseil constitutionnel held on 25 June 2009 that the requirements of clarity and sincerity of parliamentary debate, which apply to committee proceedings, demand that statements made before them be reported precisely — which shows the question of granularity is a live one.
- The video. For the floor, a complete audiovisual record of public sitting debates is produced (Instruction générale du Bureau, article 19 bis). For committees, recording is conditional: the Assemblée provides image and sound recording of committee proceedings "lorsque les installations techniques le permettent" (where the technical facilities allow) and "à la demande des commissions" (at the committees' request), article 18 bis. In other words, filming a hearing is neither automatic nor guaranteed — and where it exists, it is often the first record available to consult.
- The verbatim. This is the real bottleneck. A three-hour recording is not searchable: you cannot run a full-text search on it, you cannot tell in thirty seconds whether your topic came up, and you cannot pull an exact quote from it without listening again. A committee that finishes late in the evening therefore leaves you, the next morning, with a video nobody has time to watch and a written record that is not always available yet.
Whether to automate, and on what conditions
Automation is not an end in itself. It solves two specific problems — volume and latency — and creates a third if it is badly framed: a note produced quickly that then has to be checked in full saves nothing. Five signs tell you the moment has come.
- Volume exceeds human reading. On a sensitive bill, several hundred amendments can land overnight: nobody will read them all before the next morning's meeting, and nobody will spot by hand the identical provisions tabled by different groups.
- You are running more than three or four files. In a consultancy, monitoring is multiplied by the number of clients; in a federation, by the number of topics. The marginal cost of one more file becomes the real question.
- Latency is costing you opportunities. If you have already discovered an amendment after the committee vote, or a consultation the day before it closed, the problem is not your rigour: it is your cadence.
- Continuity is not guaranteed. A holiday, a handover, a week of travel: manual monitoring stops precisely when the workload rises. A practice that pauses in August is not a monitoring practice.
- You need fast verbatim. If your job involves quoting what was said on the floor or in a hearing, timestamped transcription is not a comfort: it is the condition for producing a note the same day.
How to judge a tool, and what Polyfact automates in this method
Three criteria let you judge a parliamentary monitoring software, and all three are checked in a demo, not in a brochure. First, traceability: every statement must link back to its source — the official document, the amendment, or the exact second of video. A tool that produces summaries without references is a black box; you will be able neither to quote it nor to defend the note. Second, actual coverage: ask for the list of sources genuinely collected, not the list of sources envisaged. Third, the spoken layer: ask whether debates and hearings are transcribed, and whether the transcript is searchable word by word.
Polyfact is an AI-powered public affairs monitoring tool, built around the two hard problems described above: the volume of amendments and the inaccessibility of the spoken record.
On the text side, the platform tracks bills article by article and automatically groups amendments carrying identical provisions, across a corpus of more than a million amendments and more than four million indexed documents. On the spoken side, over 17,000 videos from the Assemblée nationale, the Sénat and the European Parliament are indexed, more than 10,000 of them transcribed and indexed by speaker: you search for a sentence, you land on the exact second, you copy the quote with its source. That is what turns a three-hour hearing recording into material you can use the same evening.
The parliamentary calendar can be filtered by institution, committee or bill, and shows amendment tabling deadlines next to each slot. The markers set by articles 86 and 99 become dates in your calendar, rather than a calculation to redo for every bill.
Coverage is partial, and published: 60 institutions and sources are connected out of 489 catalogued, across ten countries where we already collect. We would rather display that than promise exhaustiveness. It is the same principle as line-by-line traceability — what can be verified is worth more than what is asserted.
Frequently asked questions
With the written scope, not with the tool. List on a single page the institutions you follow, the themes, the named stakeholders and the time horizon. Then translate that scope into queries, favouring hard references (bill number, code article, legislative file title), which produce few false positives. Set your cadence against the deadlines in the Règlement de l'Assemblée nationale — notably 5 p.m. on the third working day for tabling amendments (articles 86 and 99) — rather than against your own availability. Finally, fix the reporting format before writing the first note: who reads it, at what time, and with what level of sourcing.
Through the video, where it exists. The standing orders require a full verbatim record published in the Journal officiel for public sittings (article 59(2)), but only "a record" at the end of each committee meeting (article 46(3)); and the Instruction générale du Bureau provides for audiovisual recording of committee proceedings only "where the technical facilities allow", "at the committees' request" (article 18 bis). The recording is therefore often the first trace available — but it is not searchable as it stands. That is precisely the role of a timestamped transcript indexed by speaker: finding the passage that concerns you word for word, without listening again to three hours of hearing.
When one of five thresholds is crossed: the volume of amendments exceeds what one person can read before the next day's meeting; you are running more than three or four files in parallel; you have already found something out after the vote or after a consultation closed; monitoring stops during holidays; or your job requires quoting quickly what was said on the floor and in hearings. One caveat, though: automation is only worth it if it stays verifiable. A tool that produces summaries with no link to the source forces you to re-check everything, and saves you nothing.