Why Newjee

Don't trust us.
Check us.

A verdict you can't check is just another opinion. Here is how Newjee reaches its results, what it can do today, and what it can't do yet.

Why you can trust it

These rules are built into our code. They don't depend on the AI behaving well.

Scores are calculated, not guessed

The AI doesn't write the score. It labels each claim as verified, unverified or contradicted, and Newjee calculates the score from those labels: a verified claim counts 1, an unverified one 0.5, a contradicted one 0, and the most important claims count double. The score always matches the claims you see. When fewer than three claims are settled (verified or contradicted), or the unverified ones outweigh them, Newjee shows no score at all rather than a number that rests on claims nobody could check. Scores are for articles and posts. A topic you search gets none: averaging the facts found around a person or a meme grades nothing. When the topic is itself a claim, such as a theory or a rumour, Newjee gives a verdict on that claim, and only calls it contradicted or supported when sources it found point that way.

You can see what backs every step

Newjee searches the web when it runs an analysis, so it reads current coverage, not an AI's old training data. Each step of a story's spread is marked Observed if a page from that search backs it, with the link, or Inferred by the model if none does. That label is set by our code, not by the AI.

No borrowed reputations

Newjee doesn't give outlets a trust score, because a score like that would come from an outlet's name rather than from what it published. It judges the article, not the masthead. It also never cites certifications, awards or third-party ratings unless they appear in the coverage itself.

Ten copies of one wire story are one report

When AP, Reuters, AFP or ANSA copy runs in ten papers, Newjee counts one report, not ten confirmations. Newjee can't tell whether two outlets are truly independent, so it never calls them that. Our code takes the word out of the analysis even if the AI writes it, and says how many outlets carried the claim instead.

Coverage splits are counted from the outlet list

The left, centre and right percentages are counted from the outlets listed underneath, so the bar and the list always agree. Outlets Newjee can't place are shown as unclear rather than quietly counted as centre, and the bar says how many it is based on, for example “8 of 11 outlets placed”.

An editor approves every Daily brief

Daily briefs are drafted overnight, and none is published until a Newjee editor approves it. The brief that goes live is the exact draft the editor approved; nothing rewrites it on the way out.

Our methods are peer-reviewed

Three papers by Fooladi & Bottino were accepted in 2026, at NLPAICS and at the PoliticalNLP and InDor workshops at LREC. They're on the ACL Anthology and arXiv. Read the research.

What you can do with it

All of this works on newjee.eu today.

A Reality Card for any article

Paste a link to get its claims and whether each one is verified, how its sources hold up, its citations, a grade with “Why this grade?”, and suggested next steps.

A Reality Map for any topic

Type a topic, a headline or a name to see the facts, the framing, how the story spread, the people and organisations involved, and trends, with an AI analyst you can question.

Framing and manipulation, measured

Bias direction, sensationalism and emotional appeal, the narrative frames in play, and six manipulation signals: cherry-picking, false equivalence, appeal to authority, loaded language, omission and framing bias.

Who says what, and where outlets clash

See which outlets carry which frame, how the coverage splits from left to right, and an alert when two groups of outlets frame the same story in opposite ways.

Any language in

Articles and searches in any language are analysed and explained in English. Outlet names are never translated, so “Il Post” stays “Il Post”.

People and organisations

Search for a person or an organisation. If the name could mean more than one, Newjee asks which one you mean before profiling their coverage.

Where the story splits today

A live panel of the day’s stories where different outlets and regions tell the most different versions.

Share and export

Share any analysis with a link, browse analyses others made public, and copy a summary or full report, or export the data as JSON.

Newjee Daily and the newsletter

Editor-approved briefs on the stories that matter today, each with a timeline, a claim tracker and how outlets frame it, plus a morning email on the topics you choose.

Ask on any page

The chat on every Newjee page fact-checks a claim you type and answers with a verdict.

For teams and developers

Ask us about access: Newjee for business.

Newjee inside your AI assistant

An MCP connector brings six Newjee tools into Claude and other assistants: check an article, fact-check a claim, analyse a topic, investigate a person or organisation, trace a story over time, and find stories where outlets disagree. Available on request.

Science claims against the research

Check a claim against published papers from OpenAlex, Crossref, arXiv and Europe PMC. Newjee weighs each paper’s evidence quality and says whether the claim sits inside, on the edge of, or outside the scientific consensus. Part of the Newjee Suite.

A stable API

Every analysis is also available through a versioned API, with a response format that is fixed for each version. Access by token, on request.

What it can't do yet

We'd rather tell you than have you find out.

  • Analysing YouTube and Instagram posts is built but not open to the public yet.
  • The browser extension isn’t in the Chrome Web Store yet.
  • Reports export as text and JSON; there is no PDF export yet.
Ask me to verify a claim ✓