Upload a document or a URL and the entities and relations are extracted for you. People, organisations, places and more, each with the verbatim quote that proves it. You confirm, JournalInt automates.
Free beta, by invitation.
Two problems, with their sources.
On the Pandora Papers, more than 600 journalists spent more than a year structuring, researching and analysing 11.9 million records. Only 4% of the files were structured.
Source: ICIJ, About the Pandora Papers dataset
JournalInt gives you a reviewable graph from the first document.
In a study of 22 public media organisations in 18 countries, 45% of AI assistants’ answers about the news had at least one significant problem and 31% had serious sourcing problems: missing, misleading or incorrect attributions.
Source: EBU and BBC, 2025: press release; News Integrity in AI Assistants report
JournalInt only proposes a relation if it finds the verbatim quote in your document, and you review it.
From your documents to a map you can defend, in three steps.
Upload PDF, DOCX, TXT or Markdown, or paste a web address. Scanned PDFs go through OCR.
The AI proposes the entities and you confirm them. The relations come next, each with the verbatim quote from your document. You correct or delete the ones that are wrong.
What you have reviewed forms your project’s graph. Every item carries its provenance label.
For anyone who has to explain who is connected to whom, and prove it.
Little time and unpaid research work
42% of journalists do not usually have enough time.
Source: Muck Rack, State of Journalism 2026
Upload the URLs and PDFs you already have and defend each claim to your editor with its quote.
More original investigation with fewer resources
A net balance of +91 in favour of more original investigations among 280 news executives.
Source: Reuters Institute, Trends and Predictions 2026
A view that combines your projects, to see who shows up in more than one investigation.
Financial pressure and the need for traceability when publishing
According to the IFCN survey, 3 in 4 fact-checking organisations are financially vulnerable or in crisis.
Source: IFCN via Duke Reporters’ Lab, 2026
Every relation carries its quote and every item its provenance label, so you can cite with traceability.
Complex control structures and files that have to be auditable
The EU anti-money-laundering regulation (AMLR, Regulation 2024/1624) applies from 10 July 2027 and requires enhanced due diligence on complex ownership structures.
Source: eucrim, the EU's new AML rulebook; Persona, what the AMLR is (a vendor)
Assisted reading of press and reports, with the quote of each relation for the file.
JournalInt does, and it is free during the beta.
| Criterion | JournalInt | Maltego | Datashare | NotebookLM | Aleph |
|---|---|---|---|---|---|
| Does it extract relations from your documents? | Yes, typed relations | No | Only entities and co-occurrence | No: it answers questions, with no persistent map | No: entities and public data, no relations extracted from text |
| A verbatim quote for each relation? | Yes, checked against the text | No | No: it links to the document | Cites passages, not relations | No: it links to the document |
| Human review? | Yes: you confirm the entities and correct or delete each relation | Manual editing of the graph | Manual editing or tagging | No | Manual editing or tagging |
| Price per user per year | Free during the beta | €3,000 (Entry, with prior verification) to €7,500 (Professional) | Free, open source | Free with limits; paid plan separate | Free for non-profit news organisations; commercial price not published |
Sources of the comparison: Maltego, pricing; ICIJ on Datashare’s graph; OCCRP, Aleph Pro FAQ
We give no price for NotebookLM: Google does not publish amounts on its plan pages.
SourcesWhat JournalInt does with your documents, and what it does not.
Every relation is kept with the sentence of your document that supports it, checked against the text. No quote, no relation.
You confirm the entities before the relations are extracted, and you can correct or delete any of them. Every item carries its label: detected by AI, confirmed by a person or added by a person.
Your account, your files and your graph are stored in the EU. To extract the relations, the text of your documents is sent to AI model providers based in the United States.
We use paid accounts with model providers, and training on your documents is not enabled on any of them.
When you delete your account, your documents, your graph, your files and the debugging record are removed.
Do not upload leaks or drafts: during the beta, only material that is already published or public.
An extraction is a lead, not a finding. Check each relation before you publish it.
JournalInt is measured against documents that people have annotated by hand. Journalists and researchers can add documents, annotate and review each other’s work, and define the types the product looks for. The work earns points, and points are redeemed for time on the paid plans to come.
20 of 20 seats are open.
about an hour
55 points: 1 month of the low tier
about 15 hours
616 points: 1 year of the mid tier
about 40 hours
1602 points: 2 years of the top tier
This perk also needs 7 of your documents to have been accepted.
Points have no cash value. Perks are reserved until paid plans exist, and the point costs above are provisional until the plans are announced. A perk you redeem keeps the cost you redeemed it at. Read the programme terms.
Request an InvitationAn account comes first; you ask for a seat from your profile.
An assistant answers a question. JournalInt builds a map: it extracts relations between people, organisations and places, each with the verbatim quote from your document, and what you review forms a graph that is kept. NotebookLM cites passages, but leaves no persistent graph.
The AI proposes them from the text of your documents, once the entities are confirmed. Every relation carries the verbatim quote, checked against the text. You review it and can correct or delete it. An extraction is a lead, not a finding: check before you publish.
Your account, your files and your graph are stored in the EU. To extract the relations, the text is sent to model providers based in the United States, on paid accounts where training is not enabled. When you delete your account, everything is removed. If you join the collaborator programme, the documents you contribute on purpose are used to measure the extraction.
Today it supports documents in English and Spanish, and we are working to improve the extraction in both. We plan to add French, German, Portuguese, Catalan, Galician and Italian. The web app is in English only for now.
During the beta, only material that is already published or public: do not upload leaks or drafts. An on-premise deployment is planned for sensitive material, and does not exist yet.
It is free during the beta, by invitation. We will announce the prices before the launch.
Yes: you upload PDF, DOCX, TXT or Markdown, or paste a web address, and scanned PDFs go through OCR. Audio and video are not processed yet.
Not exactly. Maltego is strong at enriching with external data and drawing the graph, but it does not extract relations from your documents. JournalInt does the opposite: it starts from your documents and does not enrich with registries or sanctions lists, which is planned.
Upload the PDF to a project. The AI proposes the people, organisations and places that appear, you confirm them and, from there, JournalInt extracts the relations between them, each with the sentence of the document that supports it. If the PDF is a scan, it goes through OCR first.
JournalInt only keeps a relation if it finds the verbatim quote in the text of your document, and it checks that the sentence exists. Even so, the model can misread a sentence and propose a wrong relation: that is why you see it next to its quote and can correct or delete it.
An extraction is a lead, not a finding. The quote takes you to the exact sentence in your document, but checking the claim against other sources and deciding what is published remains the journalist’s work.
It depends on what you have. Datashare and Aleph find and organise names in large volumes of documents and are free (Aleph, for non-profit news organisations). Maltego draws the graph and enriches it with external data, and is paid. NotebookLM or ChatGPT answer questions about your documents, leaving no map. JournalInt extracts the relations from your documents with their quote and keeps them as a graph.
Seats are limited. Ask for yours and you will get an answer by email.
