FactRank logo
Content Authenticity Detection · Media Journalism

FactRank

A free Dutch-language tool that automatically spots factual claims worth fact-checking. It continuously scans Belgian and Dutch parliaments, politicians' social posts and broadcast transcripts, then ranks each statement by check-worthiness rather than by truth.

Active Free plan Free API available Verified by Guidaio
Overview

What is FactRank?

FactRank is a free tool that reads Dutch-language public debate and points to the sentences a fact-checker should look at first. It does not decide whether anything is true. It estimates check-worthiness: how likely a sentence is to be a factual assertion that also matters socially. The publisher is blunt about the distinction, stating that FactRank is not an automated fact-checker and ranks statements only by check-worthiness, never by their reliability or correctness.

The monitored corpus is unusually well documented. Three parliaments are indexed: the Vlaams Parlement and the Kamer van Volksvertegenwoordigers in Belgium, and the Tweede Kamer in the Netherlands. Politicians' accounts on Twitter, Facebook, TikTok and Instagram are followed on both sides of the border, alongside YouTube via transcription. Twenty-seven television programmes are covered, from Ter Zake, De Afspraak and Het Journaal in Belgium to Nieuwsuur, NOS Journaal, Buitenhof and Een Vandaag in the Netherlands, plus ten radio programmes including De Ochtend and NOS Radio 1 Journaal. Each detected statement keeps its speaker, party, date, source and a link back to the broadcast.

Underneath is a deep-learning classifier trained on thousands of sentences labelled by hand by students at the University of Leiden, themselves trained by fact-checking practitioners. The first model learned from 1,800 classified sentences; the corpus later grew to roughly 7,000. User feedback closes the loop, feeding periodic retraining.

Three ways of using it follow from that. The claim browser lets you sort by score or date and filter by source, country, period and speaker. A matching view links statements to fact-checks already published by Knack, Factcheck Vlaanderen and Nieuwscheckers, with a strength column. And a scoring box accepts your own Dutch text, up to 4,000 characters, or the URL of a YouTube clip to transcribe and score. Measured on 2 September 2026, the database held 34,236 statements and 1,802 matches, with claims from the previous day already visible. An undocumented but working public API exposes the same engine.

What it does

  • Detect, inside Dutch-language text, the sentences that are verifiable factual claims
  • Score every sentence for check-worthiness and rank statements by that score
  • Continuously monitor three parliaments, five social networks, 27 television programmes and 10 radio programmes across Belgium and the Netherlands
  • Match a statement against fact-checks already published by Knack, Factcheck Vlaanderen and Nieuwscheckers, with a strength indicator
  • Analyse a text supplied by the user, splitting it into individual claims and scoring each one
  • Transcribe a YouTube clip from its URL and score what was said in it
  • Collect thumbs-up and thumbs-down feedback that feeds periodic retraining of the model
Audience

When to use FactRank / When not to

A quick filter to help you decide if FactRank is the right fit.

When to use FactRank

  • Newsroom fact-checkers working in Dutch, who need a daily shortlist of claims worth verifying
  • Political reporters covering the Flemish, Belgian and Dutch parliaments
  • News producers and editors at Belgian and Dutch television and radio programmes
  • Political and communication scientists studying public debate in the Low Countries
  • Computational linguists and NLP researchers working on Dutch, who can reuse the open model and the public API

When not to use FactRank

  • Anyone working outside Dutch: the model processes no other language, and the English interface does not change that
  • Anyone expecting an automatic true-or-false verdict, which the tool explicitly refuses to give
  • Newsrooms covering countries other than Belgium and the Netherlands, since the monitored sources stop at those two
  • Organisations that need contractual guarantees, as the site publishes no terms, no privacy policy and no service commitment
  • Mobile-first users and editorial teams wanting accounts, shared workspaces or workflow features, none of which exist
Get started

How to use FactRank

A typical end-to-end flow, from setup to results.

  1. Open factrank.org: there is no account, no sign-up and no payment at any point
  2. Switch the interface between Dutch and English from the language selector; Dutch is the default and the choice is stored in a cookie
  3. Accept or decline the analytical cookies in the dialogue that appears on arrival
  4. Open the claims browser to see the statements the model has flagged
  5. Sort them by check-worthiness score, newest first or oldest first
  6. Filter by source, by country, by time period and by speaker to narrow the list
  7. Open the matches view to see which statements already correspond to a published fact-check, and how strong the match is
  8. Go to the tool page, paste your own Dutch text of at least six words and at most 4,000 characters, then run the check to get a score per sentence
  9. On the same page, paste a YouTube link to have the clip transcribed and scored
  10. Use the thumbs-up and thumbs-down icons on any misranked statement; that feedback feeds the next retraining round
Quick read

Pros & Cons

Pros

  • Completely free, with no account, no sign-up and no announced usage limit
  • Fills a genuine gap, since comparable tools such as Claimbuster serve English rather than Dutch
  • Broad and exhaustively documented source list, refreshed daily; statements from the previous day are visible on the homepage
  • Methodologically honest: the publisher states plainly that the tool is not a fact-checker and is not perfect
  • Traceable output, with speaker, party, date, source and a link back to the original broadcast on every statement
  • Open-source code under the MIT licence and a published academic account of the method
  • Solid institutional backing, from KU Leuven, the University of Antwerp, the University of Leiden and VUB, with public funding

Cons

  • Dutch only: English is an interface language, not something the model can process
  • Coverage stops at Belgium and the Netherlands
  • No terms, no privacy policy and no mention of the GDPR, so nothing frames how submitted text is handled
  • Broken navigation: the about page links to a contact page and a search page that both return 404
  • The public API works but is announced nowhere on the site and carries no stability or support commitment
  • No mobile app, no accounts, no team or editorial workflow features
  • The public GitHub repository has had no commit since October 2024, even though the service itself runs daily
Pricing

Pricing & Plans

FactRank is offered free of charge. The publisher describes it explicitly as a free tool, and no pricing page exists on the site. There is no paid plan, no published amount in any currency, no usage tier and no paywall: every feature, including the public API, is reachable without an account. The service is funded by public grants rather than by sales, which is why no lowest price point can be quoted.

Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how FactRank handles your data.

GDPR overview

There is no mention of the GDPR anywhere on the site. No privacy policy, no data controller, no data protection officer, no legal basis, no description of access, rectification or erasure rights, and no Article 27 representative. The only data-protection measure actually visible is a consent dialogue shown on arrival, asking whether analytical cookies are accepted and allowing a genuine refusal; the accompanying cookie policy names Google Analytics as the sole third party and exists in Dutch only. The publisher, Feitisj vzw, is a Belgian non-profit and is therefore subject to the Regulation, but it never says so. This is an absence of statement, not a stated refusal to comply.

Who owns the data?

The site publishes no terms and conditions, so ownership is never addressed. Requests for a terms, legal or privacy page all return a genuine 404, and the only legal document on the whole site is a Dutch-only cookie policy that says nothing about content. What the tool indexes is public speech to begin with: parliamentary debates, posts by elected politicians, and subtitles or transcripts of broadcast programmes, each displayed with its speaker, date, source and a link back to the original broadcast. Text that a visitor pastes into the scoring tool is covered by no document at all. In practice nobody claims ownership in writing, and nobody grants any either.

Reuse rights

Because there are no terms of use, reuse is neither authorised nor forbidden in writing. No data licence is published for the statement database, and robots.txt disallows crawling of the individual claim pages under /claims/. Two things are nonetheless openly available. The project's source code is published on GitHub under the MIT licence, together with the historical corpus of 1,800 hand-classified sentences, so the code and that training set can be reused freely. And the public API at api.factrank.org answers without any key or authentication, returning statements as JSON. Anyone planning systematic reuse of the statement database itself should treat the silence as a gap to clarify with the publisher rather than as permission.

Data retention & training

Retention summary
No retention rule is published. The site states no storage period, describes no anonymisation, and offers no deletion procedure. The cookie policy covers functional and Google Analytics cookies but says nothing about the text a visitor submits to the scoring tool. The publisher does state that it collects user feedback and periodically retrains the model with it, without indicating what is kept, in what form, or for how long. In short, whatever you send is handled under rules that have never been written down.
Trains on customer data
Yes

Hosting summary

No hosting information is published anywhere on the site: no country, no region, no jurisdiction and no list of processors. The only observable element is technical rather than declarative. The site resolves to an IP address geolocated in Falkenstein, Germany, on Hetzner Online GmbH infrastructure, and the API is served from the same domain and the same address. That is an infrastructure observation, not a commitment by the publisher, and it should not be read as a hosting guarantee. For context, the publisher itself is established in Belgium and the domain registrar is Dutch. No clause covers transfers outside the European Economic Area, no data processing agreement is offered, and no subprocessor list exists. Anyone with a jurisdictional requirement will have to ask the publisher directly, because the site answers none of it.

Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting FactRank.

  • A high score means 'worth checking', not 'false' or 'dubious'. The publisher warns about this, but a percentage shown next to a named politician's words invites exactly the wrong reading.
  • Statements are attributed to real, named people and displayed with a score, so a misread ranking can turn into public pillorying of a speaker.
  • Relying on the ranking to decide what to verify quietly delegates editorial judgement, and the model cannot surface what its training corpus of roughly 7,000 labelled sentences never taught it.
  • There is no privacy policy, no terms of use and no mention of the GDPR, so text pasted into the scoring tool goes to a service that documents neither retention nor reuse nor deletion. Do not submit confidential or unpublished material.
  • The publisher states that user feedback feeds periodic retraining, and no opt-out is offered or even discussed.
  • Two Belgian legal entities appear in the footer, Feitisj vzw for management and Textgain for maintenance, and no page says which of them is responsible for processing your data.
  • The public API is open and unauthenticated but appears on no page of the site and carries the framework's default title, so nothing guarantees it will keep working or keep its shape.
Setup

Setup & Integrations

Technical difficulty

Effectively none for normal use. There is nothing to install, no account to create and no configuration step: open a page in a browser, paste text and run the check. The only input constraints are a minimum of six words and a maximum of 4,000 characters. Programmatic use is barely harder, since the API needs no key and a single JSON POST returns a score. The one real obstacle is discovery: the API is documented nowhere on the site, so you have to know that api.factrank.org exists before you can use it.

Deployment

Web appAPI

Integrations

YouTube

Supported languages

DutchEnglish
Company

Behind FactRank

Company name
Feitisj vzw
Founded
13/12/2019
Country of origin
🇧🇪 Belgium
UBO
INFORMATION_NOT_FOUND
UBO country
INFORMATION_NOT_FOUND
Domain registrar country
🇳🇱 Netherlands
Support contact

Fundraising

Grant of EUR 39,869 awarded to FactRank by the Vlaams Journalistiek Fonds in December 2018
European Union funding through the BENEDMO project, Grant Agreement INEA/CEF/ICT/A2020/2381738
Both are public grants to a non-profit association, not equity rounds; no investor or valuation is named anywhere
Official links

Resources

All the official URLs gathered for verification and reference.

Compare

Alternatives

Tools that compete with or complement FactRank.

C Claimbuster
FAQ

Frequently asked questions

Does FactRank tell me whether a statement is true?
No. It ranks statements by check-worthiness only, never by reliability or correctness. Deciding what is true remains the job of human fact-checkers; FactRank simply shortlists what deserves their attention.
Which languages does the tool actually process?
Dutch, and only Dutch. The interface can be switched between Dutch and English, but that is a display setting: the underlying model is trained on Dutch and cannot score text in another language.
How much does it cost?
Nothing. FactRank is free, with no account, no sign-up and no paid tier. It is funded by public grants rather than by users.
Which sources are monitored?
Three parliaments (Vlaams Parlement, Kamer van Volksvertegenwoordigers and Tweede Kamer), politicians' accounts on Twitter, Facebook, TikTok, Instagram and YouTube in both Belgium and the Netherlands, 27 television programmes and 10 radio programmes. The full inventory is published on the sources page.
Can I score my own text?
Yes. The tool page accepts Dutch text of at least six words and at most 4,000 characters, splits it into individual statements and scores each one. The same page also accepts a YouTube link, which is transcribed and then scored.
How was the model trained?
On thousands of sentences labelled by hand by students at the University of Leiden, who were themselves trained by fact-checking experts. The first version learned from 1,800 classified sentences; the corpus later grew to roughly 7,000.
Is there an API?
Yes, at api.factrank.org, and it needs no authentication. It offers single and batch ranking endpoints plus statement queries, with interactive documentation at /docs. Note that the site itself never mentions it, so treat it as unsupported.
Who is behind FactRank?
The Belgian non-profit Feitisj vzw manages it, with support from the Vlaams Journalistiek Fonds. Since 2021 Textgain has handled maintenance and further development, within the European BENEDMO project.
Can I correct a statement that is ranked badly?
Yes. Thumbs-up and thumbs-down icons let you say whether a statement deserves a fact-check, and that feedback feeds periodic retraining of the model.
Is there a privacy policy or terms of use?
No. The only legal page on the site is a cookie policy, published in Dutch only. Nothing describes how text submitted to the scoring tool is retained, reused or deleted.
Conclusion

Should you pick FactRank?

FactRank is a narrow tool that knows exactly how narrow it is, and that is its main strength. It does one thing: read Dutch-language public debate and surface the sentences a fact-checker should examine first. It refuses, explicitly and repeatedly, to say whether anything is true. For a Dutch-speaking newsroom that distinction is the difference between a useful triage instrument and a machine that would quietly hand its editorial judgement to a classifier.

The service is genuinely alive rather than a research leftover. Statements from the previous day appear on the homepage, the database held 34,236 statements when this record was compiled, and the model answers in real time. The corpus is broad and, unusually, listed source by source. Everything is free, no account is required, the code is published under the MIT licence, and a working public API exposes the same engine, even though the site never mentions it.

The weaknesses are structural rather than accidental. Nothing outside Dutch is supported, and coverage stops at Belgium and the Netherlands. More seriously for any professional use, the site publishes no terms, no privacy policy and no reference to the GDPR, so anyone pasting unpublished text into the scoring box is doing so without any stated rule on retention or reuse. Smaller signs of partial upkeep are visible too: two links on the about page lead to 404s, and the public repository has been quiet since late 2024.

Judged on what it sets out to do, FactRank is valuable and, for Dutch, without a known equivalent. Judged as a product, it remains a publicly funded non-profit service: free, open and useful, but offering no commitment to anyone who would come to depend on it.