Collaborative fact-checking and disinformation response

Follow the claim.
Find the facts.

Disbunk is a collaborative platform where journalists, researchers and civil society verify claims, preserve evidence and coordinate responses to disinformation.

Submit a Claim See how it works

For journalistsFor researchersFor civil society

The verification trail

01 — 03

Submit

Report a suspicious claim with source links and context.

Investigate

Analysts verify claims with evidence, coordinated in teams.

Respond

Publish findings rapidly — verified, debunked, or escalated.

Every finding should lead back to its evidence.

The working parts

Tools for collaborative fact-checking

Move from an isolated link to a shared investigation, with the context kept intact.

Claims

Capture a claim, preserve its source, and follow its progress.

Evidence Vault

Keep files, fingerprints, and supporting context together.

Campaigns

Connect related claims and document the wider pattern.

Teams

Give each investigation a clear owner and a shared next step.

How Disbunk works

How claim verification works

One place to turn a suspicious claim into a documented, coordinated response.

Read our methodology

01
Submit
Report a suspicious claim with source links and context.
02
Investigate
Analysts verify claims with evidence, coordinated in teams.
03
Respond
Publish findings rapidly — verified, debunked, or escalated.

TECHNOLOGY IN SERVICE OF FACTS

AI-assisted grouping for claim investigations

At Disbunk, our mission is to identify and counter disinformation quickly, through evidence and human review.

Reviewing every claim or link one at a time takes significant effort. The internet is full of noise: repeated content and material unrelated to an investigation. Technology helps us organize that material so researchers can concentrate on verification.

In use · human reviewed

Bring similar claims together for review

HDBSCAN · Semantic grouping of multilingual claims

HDBSCAN helps us find groups of similar claims, instead of asking researchers to discover every connection manually.

For example, if dozens of claims recorded in Disbunk repeat the same narrative in different words, the analysis can suggest a group for researchers to investigate together. Similarity alone does not establish that posts are false or coordinated.

Connects similar claims

We combine multilingual semantic representations with HDBSCAN to suggest groups within each team. Researchers examine the context and evidence, then decide whether a group belongs in a documented campaign.

Leaves unrelated material ungrouped

HDBSCAN does not force every claim into a group. Material without sufficient similarity stays outside the suggested groups. No claims are deleted, and every case remains available for review.

Supports earlier responses

Periodic analysis of recorded claims can help the team notice recurring patterns and topics sooner. Automated monitoring of news and social media is a future direction, not an active capability of this system.

Why does this matter for Disbunk?

Our human resources are limited. The aim is for technology to handle part of the initial organization and reduce repetitive work. This gives the team more time to verify sources, assess potential harm and prepare evidence-based responses.

Every suggestion is reviewed by a researcher. Grouping helps prioritize investigations; findings depend on source verification and evidence.

Technology and planned integrations

We are expanding our research tools with translation, documentary sources and team communication.

Google Cloud Translation

Draft translation to support work across languages, preserving the original and human review. The feature has been prepared in Disbunk; service activation is still being configured.

Explore the tool

DocumentCloud

A planned integration for organizing documents and source material that support investigations. Documents and evidence remain part of a documented verification process.