Defending Truth Against Disinformation in Ukraine
We monitor 22 Ukrainian Telegram channels around the clock, group what they publish into news clusters, check the factual claims, and release the result three ways: as live reports, as an interactive dashboard for every report, and as an open dataset anyone can download.
Aisberg in numbers
Figures as of 26 August 2026. Collection runs continuously. How the numbers are produced.
About Aisberg
Aisberg is a Ukrainian NGO working on disinformation research, fact-checking, and media literacy. Rather than publishing opinions about the information space, we measure it: what was posted, how fast it spread, who picked it up first, and how audiences reacted.
Our name reflects the mission. Just as 90% of an iceberg lies hidden beneath the surface, much of the information ecosystem remains invisible to society. We bring those hidden layers to light, and we publish the underlying data so others can check our work.
What We Publish
Eisberg News is the running output of our pipeline. Everything below is public, free, and available right now, without registration.
Live reports
When several channels start covering the same event, we group their posts into one cluster and publish a summary: what happened, which manipulation techniques appear in the coverage, what the fact-check found, and who published first. Reports go out as the news breaks, not on a weekly schedule.
Open the channelA dashboard for every report
Each published report has its own interactive dashboard: view and reaction dynamics over the first hour, reaction mix per channel, comment sentiment over time, and the publication chronology showing which channel was first and by how many minutes. Every dashboard has a permanent link you can cite.
How the dashboard worksOpen dataset
The raw material behind the reports, de-identified and released under CC-BY-4.0: monthly JSONL snapshots of posts and comments, plus the per-cluster JSON that powers each dashboard. Downloaded 8,165 times so far, around 3,000 of them in the past month. Need a fuller or custom cut? Ask us.
Browse the datasetBuilt for the platform where narratives start
Three choices that define the project
Telegram-first
Most monitoring infrastructure was built for the open web and large Western platforms. In Ukraine and Eastern Europe, narratives form and spread on Telegram first, so that is where our collection, clustering and analysis run natively.
Hours, not weeks
Reports publish within the first hours of a story's spread, when verification can still affect how far it travels. A retrospective case study explains what happened; a real-time report reaches readers while it is still happening.
Open by default
Every verdict ships with its evidence and every report with its data; the method is published together with its limitations. Findings from this pipeline can be reproduced and contested by anyone, which is what makes them usable in research and reporting.
Our Core Principles
The values that guide our work
Transparency
We publish our methodology, our data, and our limitations, including the cases where our own pipeline gets it wrong.
Media Literacy
We give people tools for thinking critically, not only information to absorb.
Diversity
We value different perspectives while staying true to facts.
Integrity
We remain independent and resist external pressure that could compromise our mission.
Our Work
The capabilities behind the published output
Continuous collection
Posts are captured the moment they appear, then re-measured four times over the following hour so we can see how views, reactions and comments actually develop.
Narrative clustering
Multilingual embeddings group posts covering the same event across channels, which is what turns a stream of separate messages into a trackable narrative.
Manipulation detection
Each cluster is scanned for a defined set of manipulation techniques, and coverage is checked for signs of coordinated positive or negative framing.
Fact-checking
Factual claims are verified against web sources with dated, country-tagged, scored evidence. 2,420 claims checked so far, including the ones that came back unverifiable.
How the fact-checker worksAudience analysis
Comment sections are analysed for sentiment and for patterns that distinguish organic discussion from coordinated activity.
Open data
Everything above is released as an anonymized public dataset so researchers can reproduce, contest, or build on our findings.
Roadmap
What we're building next. Published so you can hold us to it.
- A public web dashboard with search and filtering across every cluster, instead of one report at a time.
- Quarterly reports: written analysis of the narratives our pipeline surfaced over the previous three months.
- Media literacy materials that explain manipulation techniques using real examples from our own data.
- Partnerships that feed live narrative clusters into the verification workflows fact-checkers and newsrooms already run.
- Coverage of more than the current 22 channels.
Partnerships and People
We work with donors, fact-checking organizations, newsrooms and researchers, and we welcome volunteers who share our values. The pipeline is already running; the next step is scale.
Discuss a partnership Become a volunteer