Independent fact-checking for what people share online
Factiz reads a message or a screenshot and says whether it is true, false or misleading, with the sources that show it. Often a newsroom has already checked it. We bring that check to the moment someone doubts.
- People in group chats
- Fact-checkers and newsrooms
- Researchers and platforms
Why we do this
Facts are a common good, and they need upkeep
A screenshot lands in a family group chat. It shows a headline that never ran, under the logo of a newspaper everyone knows. Within an hour, half the chat believes it and the other half is angry about it.
Often the truth is already out there. A newsroom checked this exact claim, maybe a year ago, maybe in another language. Nobody in that chat will find it in time. So the lie keeps travelling, and the check stays on the page where it was published.
That matters more than it seems. A democracy works when people can disagree about what to do while agreeing on what happened. That shared ground is wearing thin. Making a convincing fake now costs almost nothing, and checking one still takes a person, a day and a newsroom. In January 2025, Meta announced it would end its fact-checking programme in the United States.
We are not naive about what a tool can do. A label will not change the mind of someone deep in a bubble. What we can do, and measure, is make lying more expensive. Whoever fills two hundred websites in ten languages with little effort is making copies, and copies can be recognised. Then one check by one newsroom can cover a thousand copies.
We do this because we believe in a public that can find out for itself. Not by trusting us, but by seeing the sources. That is why Factiz never asks you to take its word for anything, and why it belongs to no platform, media group or AI company. A check button owned by a platform inherits the distrust of its owner.
Who it is for
For everyone who has to decide what to believe, quickly
The person in the group chat
Someone forwards a screenshot, and they want to know within seconds whether it is true, without an account.
Fact-checkers and newsrooms
Their work deserves to reach the moment of doubt, and nobody should have to check the same claim twice.
Researchers, platforms and institutions
They need to see a network of fake sites while it is being built, not a year later.
What it does
A fact-check machine for forwarded screenshots
Send Factiz a screenshot or a message, and it gives one answer, with its sources. When it is false, it says why. The colour comes from the sources: red only when a reliable source contradicts the claim, green only when reliable sources confirm it.
- True
- Reliable sources confirm the main claim.
- False
- Reliable sources contradict it. Factiz says why, and links to them.
- Misleading
- Partly true, old news presented as new, or a real photo in the wrong context.
- Not confirmed
- News this big would have been reported, and no reliable outlet reports it.
- Satire
- Meant as a joke, and a source says so.
- No claim
- A greeting, an opinion or an ad. Nothing to check.
- Uncertain
- Nothing decisive was found, and Factiz says so instead of guessing.
Each sentence of the explanation carries the number of its source. A sentence without one is left out.
How it works
Five steps, the way a fact-checker works
Read
Factiz reads the text in the screenshot and blacks out phone numbers, email addresses, handles and names in chats before anything goes to a search engine or a language model.
Remember
It looks for the claim among the fact-checks newsrooms have already published.
Search
It searches the web the way a fact-checker would, starting with the site of the outlet the message names, then the actual news and the original source.
Weigh
A language model reads what was found, but a fixed list of rated news sites picks the colour. The list is built from public ratings, including the IFCN and EFCSN signatories, the Journalism Trust Initiative and the Trust Project.
Answer
One answer, a short explanation, and every source with its link.
Where we are
Not public until each language clears a bar
Factiz works in English, Spanish, German, Portuguese, Dutch and French, in our own tests. A language goes public only when it clears a bar we set in advance.
- 1 in 100At most 1 in 100 real news stories flagged as false or misleading.
- 1 in 100At most 1 in 100 fake stories marked as true.
- 80 in 100At least 80 in 100 known fake images flagged, with the proof.
- 60 in 100At least 60 in 100 new fake stories flagged or marked not confirmed.
- 15 secondsAn answer within 15 seconds, with visible progress.
Right now we are collecting the fact-checks of newsrooms in each of these languages, from the ClaimReview data they publish on their own sites, and measuring every language against that bar.
We are also adding signals about where a message comes from, such as known network domains, copies of real news sites and fake media logos. During the riots in France in October 2026, a Russian operation spread videos made to look like Western media, according to an internal EU assessment reported by Reuters. That is the kind of message Factiz has to recognise.
What comes next, in this order
- A public report on how language models handle facts, one for each new model.
- Factiz for fact-checkers and newsrooms, with a person reviewing the answers.
- A paid service for organisations.
- Later, a free check for everyone.
Where this leads
The goal is an open map of how false claims travel: in which languages, through which networks, and, as far as the evidence goes, back to where they started. Every arrow on that map will carry its receipt, the source that put it there.
What we hold to
We publish our misses
Every release comes with its measurement, including what got worse. Factiz shows sources and counts, never a percentage for how true a claim is.
Talk to us
We are looking for partners in each language
Fact-checkers and newsrooms who want to try it on what reaches them. Researchers who track influence operations and want to compare data. Organisations that need to know what a message is before they act on it.