MEP of the week

Methodology & data sources

Everything on this site traces back to official, public records. This page documents where each figure comes from and how we process it. The pipeline is run daily; every entity page shows the date its data last changed.

On this page

1. MEP lobby-meeting declarations (the core dataset)

Under Article 7 of the MEP Code of Conduct, MEPs must publish meetings held with interest representatives; since November 2023 this applies to all MEPs. The source of record is each MEP's declaration documents on europarl.europa.eu, linked from every profile page here. The Parliament lists each member's meetings newest first, by meeting date, ten to a page. We read these daily on a rotating schedule: every member's first page at least once every two days, and the next pages while everything on them is new, up to a hundred meetings; and, in turn, each member's whole list at least once every two weeks, which picks up declarations published late for older meetings (see how they are counted). We merge them with Transparency International's Integrity Watch EU bulk feed (the historical base; that feed stopped updating in June 2025, since when our own daily collection is the only source of new declarations). For a meeting on or before 29 April 2025, the feed's last day, a declaration the feed holds is shown once, as the feed recorded it, and our own collection adds those it lacks.

2. Matching counterparts to the Transparency Register

Declared counterparts are free text. A counterpart entry is register-linked at source only when the MEP's declaration carries a Transparency Register ID. Separately, we try to resolve unlinked names to a Register identity; those inferred matches let spelling variants share an organisation profile and ranking, but never turn an unlinked declaration into a source-linked one. Where a match rests on who wrote the name - an acronym of three letters or fewer, or a Register name that adds a country the member did not write - it holds entry by entry, only for the members whose own record backs it: they linked that organisation on another entry, or sit for the country concerned - unless a reviewer has checked that spelling against the declarations and recorded it (a curated alias). Another member's evidence is not enough, so the same spelling can name a registrant in one member's entry and nothing in another's. Roughly 49.8% of eligible counterpart entries carry a declared Register link. Entries naming diplomatic or government bodies (which cannot register) are excluded from the site-wide headline, political-group, monthly, weekly and annual rates. Individual MEP profiles instead show the raw declared-link share across all of their counterpart entries, including diplomatic and government entries, and compare it with the same raw Parliament-wide denominator.

3. Organisation category, sector and spend

Category (company, business group, NGO, think tank, union …) follows the register's own taxonomy plus a reviewed classification for unregistered counterparts. An organisation in the Register has one type, decided once for its Register ID, and its page, the sector tables, the chart of counterpart types and the business share all use it. Where the Register files it only under its catch-all, “Other”, its page says Other and the chart leaves it out of the types until a reviewer places it. Industry/sector is a label this site assigns, not an official field: reviewers chose it for the most-met organisations, and a keyword classification over the organisation's name, stated goals and register sub-category gives it for the rest. Declared lobbying spend comes from LobbyFacts.eu, which republishes register filings.

The most-met organisations - 2,267 of them, keyed on the Transparency Register ID so every spelling of a name inherits the same answer - carry sector labels reviewed organisation by organisation. 1,434 were reviewed in August 2026: each was labelled independently by two or three reviewers working from different evidence and adjudicated (they agreed outright on 1,385, and 11 boundary cases were settled by the tie-break rules below), then tested against blind samples the reviewers never saw (about 96% right). 728 more came from a two-lens review in September 2026 and October 2026, of organisations the keywords had left in Cross-sector / other: two reviewers labelled each one on their own, one from what the organisation is and one from its Register text alone, and a label was kept only where the two agreed; another 26 were settled on a third look, where one reading was clearly right. The same review added two sectors, Construction & real estate (58 reviewed labels, counting organisations with too few meetings for a profile) and Democracy & civil society (71 reviewed labels, counting organisations with too few meetings for a profile), for organisations most of which had been filed under Cross-sector / other. Of those labels, 40 are August ones the review moved there: 15 from Cross-sector / other, 12 from Chemicals & manufacturing, 11 from Human rights & humanitarian, plus one each from Finance and Think tanks & research. 39 more were set in a later editorial review. The remaining, less-met organisations carry the keyword classifier's label, which is measurably weaker: on a frozen reference set of 540 organisations whose sectors were reviewed by hand, it is right 380 times (70%), and a blind sample of 100 small organisations in August 2026 put it at 59%. Sector labels on low-volume organisations should be read with that in mind; every label is open to correction.

When the keywords do not point clearly to one industry, the classifier turns to the policy areas the organisation ticks in its Register entry - the Register's own list of forty, from "Energy" to "Migration and asylum". It uses them only when they are concentrated: at least two of them name the same sector, and those are more than two-thirds of the areas that name any sector and more than a third of all the areas ticked. Registrants tick seven on the median, and many of the areas name no industry at all, so this places few organisations, mostly human-rights and humanitarian groups; on the reviewed labels it is right more often than the keywords are. Otherwise the classifier gives no answer rather than a guess, and the organisation is filed under Cross-sector / other. That page counts the two kinds it holds: organisations a reviewer placed there because no single sector on this site fits them - cities and regions, churches and faith groups, sport bodies, the legal professions, standards and testing bodies, social-welfare networks, and cultural organisations, among others - and organisations there because the automatic classifier could not match their names and Register entries to one sector, and no reviewer has placed them yet.

More than one sector

An organisation can genuinely belong to several. The German municipal utilities association states in its own filing that its members work in energy, water, wastewater, waste and telecommunications - reporting only the strongest of those would be a loss of information rather than a simplification. Where a second or third sector scores within 55% of the leading one, it is shown after the headline sector as also. The headline sector is what the rankings and sector pages use; the others say where else the organisation belongs.

Four tie-break rules

Four boundaries recur often enough that leaving them to judgement produced inconsistent answers - the same kind of organisation landing in different sectors depending on the wording of its filing. They are settled by rule:

  1. Rights before subject. An organisation that campaigns for rights takes Human rights & humanitarian even when its subject is technology. A body that sets standards or runs infrastructure takes Technology & digital. So digital-rights groups sit with human rights, not with the industry they scrutinise.
  2. Animals: welfare, farming or medicine. Animal welfare and protection campaigning is Environment & climate; breeding, livestock and farm bodies are Agriculture & food; the animal-health industry and veterinary professions are Health & pharma.
  3. Press: freedom, publishing or labour. Press-freedom campaigners are Human rights & humanitarian; publishers and broadcasters and their trade bodies are Media & publishing; journalists' unions are Labour & trade unions.
  4. Watchdogs follow their subject only if they have one. A watchdog devoted to a single industry takes that industry - a finance watchdog is Finance. A watchdog of governance, lobbying or transparency in general takes Democracy & civil society.

These are editorial choices, not facts about the organisations, which is why they are written down here rather than left implicit in a keyword list.

4. Counting rules

Two different units appear in this dataset and we keep them apart, because they differ by about a quarter:

  • A declared meeting is one entry in a member's declaration - one member, one date, one subject. A meeting that names six organisations is one meeting. There are 102,583 of these.
  • A counterpart entry is one organisation named in one meeting, so the six-organisation meeting above contributes six. There are 127,688 of these. This is the right unit for counterpart-level coverage and classification, and the wrong one for "how many meetings".

Every figure the site labels "meetings" is the first unit. The second unit is used only when the property belongs to a counterpart - for example, whether an organisation was linked to the Transparency Register or how it was classified - and is explicitly named as counterpart entries.

  • Site-wide totals cover the full record (2016–2026) for current members.
  • Per-MEP and per-party figures cover the current term (from 16 July 2024), so every member is measured over the same window.
  • Members who have left the Parliament during this term keep their profile, with the record of their time in the seat. They are not ranked, and their meetings are in none of the site-wide, group, country, sector or organisation figures, which cover the members seated now. Roll-call results include their votes, as Parliament recorded them.
  • All figures are counts of declared meetings. We do not estimate undeclared activity.
  • Rankings order MEPs by declared meeting records; they are statements about declaration volume, never about integrity or compliance.

5. The two labels: "the black box" and "the sloppy"

Two labels describe how a member fills in the declaration, not the member. Both are computed from the same data as everything else on this site, and both are stated here in full so the claim and its limits arrive together.

The black box

Given to a member who took their seat at least two years before the build date, and for whom no lobby meeting appears anywhere in the record we hold - not in the current term, and not in the 2016–2026 archive. The two-year rule is what separates this from "has declared nothing this term": a member sworn in last month has published nothing because they have not had time to, which is a different fact. Seating dates come from the Parliament's own record of each member's term, with its list of incoming members filling any gap; a member whose seating date is unknown is never labelled. 39 members carry the label, and each of their profiles gives the seating date it rests on.

What it does not mean. An empty record is not evidence of a rule breach and not proof that no meeting took place. Since November 2023 the obligation under Article 7 covers every member's scheduled meetings with interest representatives on matters relating to Parliament's business, so an empty record means no such meeting was declared - not that none was due. The label records the absence of a declaration, which is the only thing this site can observe. The Parliament also publishes members' declarations of attendance at events organised by third parties; that is a separate obligation, not a meeting, so it does not count here either way.

One exclusion. The President of the European Parliament is not included: the Presidency publishes its engagements through a separate channel from the MEP meetings register this site reads, so a zero there would measure where we look rather than what was declared.

The sloppy

Given to a member who, across what they published this term, leaves out the Register ID in at least 40% of the entries where they name an organisation that is in the EU Transparency Register, on at least 10 such entries and at least 5 left out. Those are the entries where a link existed to be made: the ones the member linked, and the ones where we can find the organisation in the Register although the member did not link it. Across the Parliament, 15% of such entries carry no ID. 17 members meet this test, and each of their profiles gives the two numbers beside the label.

Because every entry a member did link is part of that count, a member with the label also leaves at least 40% of everything they publish without a Register ID. The reverse does not hold. Many counterparts - individuals, local bodies, organisations that never registered - cannot be linked at all, so a low overall link share is not by itself carelessness. That is why the label looks only at organisations that are in the Register, and why a member can link a larger share of all their entries than the average member and still carry it, or a smaller share and not.

The unit is the counterpart entry, not the meeting, because a Register link attaches to an organisation: a meeting naming three bodies can have two linked and one not, so "carries no link" has no meaning at meeting level. Members with fewer than 10 such entries are not judged at all - at that size one entry moves the share by ten points or more, and the label would describe a row rather than a person.

What is excluded, and why. Counterparts that cannot hold a Register entry at all - diplomatic missions and government bodies - are left out entirely. An unlinked entry counts as naming a Register-listed organisation only on evidence strong enough to print beside a name: not when the match rests on a free-text string that names several organisations at once, not when the only evidence is one other declaration of the same name, and not for meetings held before the organisation joined the Register, when there was no entry to link to. A name of three letters or fewer ("ICE", "RTE") stands for too many bodies to go on the name alone: it counts only when the member's own record backs it - they linked that organisation's Register ID on another entry, or they sit for the country where it has its head office - or when a reviewer has checked that spelling against the declarations and recorded it. Entries where the member typed the Register number into the free-text field count as disclosed: that is disclosure in the wrong box, not a failure to disclose. This makes the count stricter than the "Unlinked but matchable" list on each profile, which also shows looser matches.

The match is ours, not the member's. We infer the organisation from the name typed into the declaration. As in section 2, that inference never turns an unlinked declaration into a source-linked one, and it is not a finding that the member was obliged to link it. It can also be wrong: an acronym held by one registered organisation can be the name of another that is not registered, so a member's count can be out by an entry or two.

Changed on 7 October 2026. Until then the label used a two-step test: at least a quarter of all entries without an ID, and at least a quarter of those naming a Register-listed organisation. Because the second share was taken over the unlinked entries alone, a member who left few entries unlinked needed only a handful of Register-listed ones among them to pass, while a member who left many unlinked could leave out the ID for more Register-listed organisations and still not reach it. So the test labelled members who left the ID out of a smaller share of their entries naming a Register-listed organisation than the Parliament as a whole, and spared members who left it out of a larger share. The rule above measures that share directly, and at a threshold of 40% - more than twice the Parliament's rate of 15% - so the label marks members who leave out Register-listed organisations far more often than their colleagues. On the data of this build the old test would label 200 members, 130 of whom link a larger share of their entries than the Parliament-wide 52% their profiles are compared with; the rule above labels 17.

6. How the "MEP of the week" is chosen

Since week 2026-W30, the MEP of the week is the member with the most meetings published in the last completed week that were at most 14 days old at publication and whose declaration carried a Transparency Register link. "Published" means the declaration first appeared in our daily collection of the official EP pages - the closest observable proxy for the moment of publication. A completed publication week never changes afterwards, so the choice is final the moment it is made. A week still in progress has no edition yet, and the earlier meeting-date formula is never used in its place. Ties break on the number of distinct registered organisations, then on total publications. The public archive begins at 2026-W30, when the site went live; we do not back-fill editions for weeks before then, even though an earlier meeting-date formula could compute them retrospectively.

Known limitation: "first seen" is when this site first read a declaration, not when the member published it. Since 6 September 2026, while collection is running, each daily run reads most members' pages, taking them in turn, so every member's page is read at least once every two days: the lag is at most about two days and varies by member within that. Before then the rotation was much slower. Collection has also stopped outright at times, before that date and since - every stop is listed below. When it resumes, the first three days of reading - a full pass over every member's page, which takes up to two days, plus a day to spare - pick up declarations that may have been published at any time during the stop, so those catch-up days cannot date a publication either: declarations first read on them count everywhere else on the site, but not for the MEP of the week. A week is scored only if this site observed it from its Monday for at least four days - the days of a stop and its catch-up days do not count as observed. A week observed in full is scored as usual; a week cut short after four or more observed days is scored, if anything was published in them, and marked with the number of days observed; any other week is not scored at all, rather than scored as empty. For example, 2026-W30 was observed Monday to Thursday and is marked as such. Every stop on record, and why it happened:

  • No collection from 24 July to 5 September 2026; catch-up days 6-8 September 2026. The European Parliament's website had begun answering automated requests, like the ones this site sends each day to read members' pages, with an empty page instead of the declarations; people visiting in a browser still saw them as normal. This site's own checks did not notice for sixteen days, because an empty page looked like a member with no new meetings. A way round the block, through Cloudflare, was built on 8 August, but it was never connected to the daily update, so the update read nothing for four more weeks. Collection resumed on 6 September 2026, when reading was routed through this site's web host, Cloudflare, whose requests the Parliament's website answers normally. Each read then took only the ten most recent meetings on a member's list, so declarations published during the stop for older meetings were not collected at first. Each member's whole list is now also read, at least once every two weeks, which picks them up as the reads reach them, counted as found late.
  • No collection from 20 to 23 September 2026; catch-up days 24-26 September 2026. This site's own daily update failed on those days: each run read members' pages, but stopped before saving what it had read. On 20 September one of the site's automatic consistency checks, which recalculates the published figures independently to catch mistakes, stopped the run: its copy of the rule for the "black box" label was out of date, and it disagreed with the site about a member who had just taken their seat. From 21 September a different step failed: the service that runs the update had run out of the storage it allows for keeping copies of the built site. Collection resumed on 24 September 2026, when both faults were fixed.

One day's reading was lost although collection did not stop. On 13 September 2026 a run read members' pages, but GitHub rejected its save with a server error, which the daily update then ignored. The next day's run read the same members again, so the declarations the lost run had found count from 14 September - in the week of 14-20 September 2026 rather than 7-13 September, which has no edition - and some members' pages went three days between saved reads. None of them is Christian EHLER's, the MEP of the week for 14-20 September, so the choice would be the same without them. The save now tries again when it is rejected, and fails the day's update if it still cannot save.

A declaration published late - during a stop or at any other time - for a meeting older than a member's newest ones sits further down their list than the daily reads go. Those reads used to take only the first page of each list, and a check on 29 September 2026 of twelve members whose newest declarations were all or nearly all new to this site when reading resumed on 6 September found 442 declarations on the Parliament's website, published after this site's full read of their lists in mid-July 2026, that this site had not collected. Each member's whole list is now also read, in turn, at least once every two weeks, so declarations like these are picked up as the reads reach them. A declaration that a whole-list read finds more than a page (ten meetings) below the meetings new since the member's last read may have been published at any time before, so it is counted as found late - unless the member's whole list had been read in full in the three days before: then it was published since, and counts from the day it was found, like a declaration at the top of the list. Everything found for a member this site had never read in full and held no declaration for, such as a newly seated member, is counted as found late too. A declaration found late counts in every total on the site, but in no publication week, and so never for the MEP of the week. (For a meeting on or before 29 April 2025, the last day of the older bulk feed, a declaration that feed holds is shown once, as the feed recorded it; one it lacks is added, and counts in no publication week however it was found.) The Parliament also edits declarations it has published, renaming counterparts and rewording titles. When a read finds, in place of a declaration held, the same meeting under a new counterpart name - the same Register entry, or a name that is plainly the same organisation's - or the same counterpart under a reworded title, the declaration already held is kept, so it is not counted twice. An edit read before this site recognised edits, on 7 October 2026, or by a read that had not yet reached the rest of that day's meetings, was added as a new declaration, so each complete read of a member's list also notes the declarations held that the list no longer shows, as they were or edited. Where it shows, in place of one of them, a version this site first read later - on the same meeting day, with the same counterpart, the same Register entry, or, under the same title, a name that holds the other - the version the Parliament has replaced is left out (250 counterpart entries of 183 members so far), and the one it lists counts once, from the day the first was read. This applies to meetings after 29 April 2025; for a meeting on or before that day, a declaration the older bulk feed holds is shown once, as the feed recorded it. A Register link a member gave a name that has since been replaced still counts as evidence of which organisation that name means, wherever else it is written. A declaration the list no longer shows, with nothing in its place, is still counted.

7. Quality control

Each daily build re-runs the counterpart-type classifier (company, NGO, union …) against a reference list of named organisations and fails the build if, with at least 30 tested, fewer than 97% reproduce. This is not the sector classifier described above. In this build, 263 of 263 reference labels were reproduced. This is a repeatability gate, not an estimate of the classifier's accuracy on every possible organisation. Separately, the build passed 8 of 8 automated integrity checks; the audit detail ships with the public dataset.

8. Roll-call votes

Vote data comes from the European Parliament's own Open Data Portal, which publishes each recorded vote with the full list of members who voted for, against and abstained. Members are identified by the Parliament's own numeric id - the same one used throughout this site - so no name matching is involved.

What is and is not here. Only roll-call votes are recorded by name. Votes taken by show of hands, and secret ballots, are not: they are absent from this dataset entirely rather than counted as anything. About one roll-call in ten is the binding final vote on a text; the rest are amendments, split votes and procedural motions. "Final vote on a text" means a vote classified as such from the Parliament's own label for it, and the share of labels we cannot classify is published as a quality check.

A missing vote is not absenteeism. The Parliament publishes who voted, never who was absent or why, and keeps no public excused-absence register. A member with no vote recorded may have been on parliamentary business, ill, on leave, or present without voting. We therefore report "roll-calls with a vote recorded" and never "attendance" or "participation". The President of the Parliament chairs sittings and does not vote, so no voting record is published for the office-holder.

Denominators. A member is measured only against the roll-calls held while they actually held the seat, taken from their membership record - which is also how members who have since left keep the denominator they had. Political group is resolved as of the date of each vote, never as of today: 22 members changed group during this term, and scoring their earlier votes against their current group would misrepresent every one of them.

Group majorities. A group has a majority on a vote when more than half of its members who cast a vote went the same way. With three positions available a group can split without any majority, and those votes are excluded from comparisons rather than resolved by a plurality. Members sitting with no group are excluded from group comparisons: they are not a bloc with a line to take. Groups are not bound to vote together in any case - a Member's mandate is personal and exercised freely - so voting with one's group is not a virtue and differing from it is not a fault.

Corrections. The Parliament publishes vote corrections and voting intentions after the fact, stating that they "do not alter the result of the vote as announced in plenary". We record them and never count them. Because corrections keep arriving for about two weeks after a sitting, each sitting is re-fetched over a 21-day window before its record is treated as settled.

9. Known limitations

  • Declarations are self-reported and published with delay; the record lags reality.
  • A declaration published late, for a meeting older than a member's newest ones, is collected only when that member's whole list is next read, at least once every two weeks. Unless the list had been read in full in the three days before, nothing says when it was published: it is counted as found late, in every total but in no publication week; for a meeting on or before 29 April 2025, the end of the older bulk feed, it is added only if that feed lacks it. After a stop in collection there can be many (see declarations found late).
  • Free-text counterparts can defeat matching; we surface the unmatched share.
  • The keyword classifier behind the less-met organisations' sectors is heuristic; each sector page says how many of its labels reviewers chose and how many the classifier gave.
  • Lobby meetings carry no dossier reference, so they cannot be linked to a particular law. Nothing here claims a meeting was about a given vote.

Licence & reuse

EU Transparency Register and EP data: © European Union, reusable under the EP/EC reuse policies. Roll-call vote data: © European Union, reusable under CC BY 4.0. Which roll-call is the final vote on a text is our own classification, derived from the Parliament's published label for each vote - no third-party vote dataset is used, so nothing here is encumbered by a share-alike licence. LobbyFacts data: © Corporate Europe Observatory / LobbyControl. Our derived dataset and computations: CC BY 4.0 - cite "MEP of the week" and link the page you used. Journalists and researchers: contact@mepoftheweek.com for the full dataset or questions about method.