How LiesWatcheur works, what we measure, and why you can trust our results.
LiesWatcheur is an automated, apolitical, transparent media audit platform for real-time fact-checking of French TV and radio broadcasts. We monitor public statements, verify factual claims, and publish our findings openly — regardless of who made the statement.
We receive live broadcast streams and convert every spoken word to text automatically — the same way your phone can write down what you say.
Example: A minister says "Unemployment fell by 3% last year" — our system captures that sentence within two seconds.
We identify who is speaking at every moment — even when multiple people interrupt each other.
Example: During a debate with 3 politicians, each statement is attributed to the correct speaker automatically.
Our detector reads the transcribed text and identifies statements that make a factual claim — something that can be checked against data, statistics, or official records.
Example: "Unemployment fell by 3%" is a factual claim. "I believe things will get better" is an opinion — we skip it.
For each claim, we search trusted sources — official statistics, academic studies, government databases — to find evidence that confirms or contradicts it.
Example: INSEE data shows unemployment actually rose by 1%. The claim is marked false.
When a false statement is confirmed, we automatically cut a short video clip (15–30 seconds) showing the statement, the source that disproves it, and the confidence score — ready to share.
Example: A clip shows the minister saying the false statistic, with an overlay showing the real INSEE figure below.
Every finding comes with a confidence percentage — a combined score reflecting how certain we are of our analysis.
How the score is calculated
A score of 92% means:
We rely exclusively on verifiable, non-partisan sources:
Every claim is linked to its primary source. You can verify any finding yourself.
We treat false statements from all sides equally. Our system monitors all major political speakers and applies identical fact-checking standards regardless of their affiliation.
Our LLM prompts do not encode partisan bias. We have deliberately excluded any political framing from our AI instructions. The same claim from a left-wing and a right-wing politician is evaluated identically.
Our methodology, prompts, and scoring weights are published in our open-source repository for public review.
LiesWatcheur processes publicly broadcast audio in accordance with French press freedom and public-interest journalism provisions. All audio is sourced from publicly accessible broadcast streams.
We do not store personal data. Speaker identification is used solely to attribute factual claims to their source. No biometric data is retained.
Generated proof-clips are published under a public interest and fair use framework. They include attribution to the original broadcast source.
For legal inquiries, contact: legal@lieswatcheur.fr
© 2026 LiesWatcheur. All rights reserved.