How Repsense catches coordinated disinformation with Apify

Repsense, a Lithuanian narrative intelligence company, reported that moving the social media collection layer of its Havel AI engine to Apify Actors let it collect 3x as many Facebook posts and almost 11x as many Facebook comments per month within three months, after its previous provider's keyword-based model left gaps. The switch added Facebook groups and comments, Instagram posts and stories, LinkedIn and TikTok to a stack that had stopped at Facebook pages, letting Havel trace a single narrative across platforms. Havel monitors the information space in more than 80 languages and has processed over 50 million content items, including 793,000 Facebook posts in a study of Lithuania's "Žemaitaitis network" and 1.5 million TikTok videos screened in a German study.

Summary Repsense https://repsense.io is a Lithuanian narrative intelligence company that detects, analyzes, and forecasts foreign information manipulation and interference FIMI . Its work includes coordinated-campaign analysis for the NATO Strategic Communications Centre of Excellence https://repsense.io/solutions/counter-disinformation-stratcom and the European Parliament https://repsense.io/case-studies/the-european-parliament-under-siege-unmasking-russian-information-warfare . How early it catches a FIMI campaign depends on how much of the information space its Havel https://repsense.io/platform AI engine can see, and until late 2025, that view stopped at Facebook pages. Moving the social media layer of its collection stack to Apify Actors https://apify.com/store opened up Facebook groups and their comments, Instagram posts and stories, LinkedIn, and TikTok. More data let Repsense connect the dots: tracing a single narrative as it moved from a Facebook group to a TikTok video to a comment thread, instead of studying each platform in isolation. Three months in, Repsense was collecting 3x as many Facebook posts and almost 11x as many Facebook comments per month. When an urgent investigation needed data we simply couldn't get, Apify let us collect it the same day. It has been the social media layer of our collection stack ever since. -- Mykolas Katkus, CEO, Repsense Inside Havel It was no coincidence that Repsense named the engine after Václav Havel, the Czech playwright and dissident chosen as a 20th-century symbol of free speech. Havel https://repsense.io/platform detects, analyzes, and forecasts FIMI, monitoring the information space in more than 80 languages, across online, broadcast, print, and social media. Everything between collection and output is automated: content gets translated, tagged for entities and metadata, and scored for sentiment, whatever language it arrives in. Video gets the same treatment, with speech and on-screen text extracted from TikTok, Reels, and Shorts. From there, Havel looks for the same claim resurfacing in different clothes: a text message on Telegram, a spoken-word TikTok, an Instagram carousel, a Facebook post, a national broadcast panel. Missing any one of those formats breaks the chain that proves it's the same narrative. The matching itself runs on text-similarity and timestamp analysis, built to track a claim even after it changes language or form. That method, combined with the breadth of sources Apify feeds it, is what answers the questions that matter: where a narrative began and where it's heading, whether it was amplified artificially or organically, what its propagation network looks like, and who's winning it. Impact scoring and analyst reports come out at the end. Havel's investigations run large: one study of the "Žemaitaitis network" in Lithuania covered 793,000 Facebook posts. A German study screened more than 1.5 million TikTok videos out of 3.1 million content items. Eight months of election monitoring in Armenia https://repsense.io/case-studies/armenia-diaspora-mobilization-and-domestic-narratives-in-pre-election-environment processed 1.8 million media items across 12 platforms. In total, Havel has processed over 50 million content items, and the published research https://repsense.io/case-studies runs from the 2025 Moldovan elections https://repsense.io/case-studies/moldova to wind-turbine disinformation in the Czech Republic https://repsense.io/czech-republic-wind-turbine-disinformation-and-cross-border-opposition-networks . The challenge: missing data means missing patterns Havel detects coordination by comparing content across accounts and platforms. Miss part of the stream, and the pattern disappears with it. Collection has to cover everything, in near real time, video included, at volumes no team could gather manually. The original data provider the team used fell short in two ways: a keyword-based model that left Repsense with little control over results, and Facebook access limited to pages, with no groups or comments, and several other platforms missing entirely. A second provider ran on the same keyword model and still didn't cover every media type needed. Building scrapers in-house was never realistic either, given how often platforms change. December 2025: an investigation that needed data they couldn't get One investigation made the data gap too costly to ignore. On December 20, 2025, Havel flagged a FIMI campaign targeting Lithuania's planned military training ground at Kapčiamiestis, a border town in the Suwałki Gap, three days before it gained measurable traction on Lithuanian social media. The analysis needed content from Facebook groups, which the existing setup could not reach, so the team went looking for an alternative and found Apify. They tried it on the live investigation and finished the research in half a day. Setup was straightforward, and the trial showed them they could extend future research to the media types they had been missing, so they stayed with Apify. The published analysis covered 5,623 Facebook comments, 95 video posts, and 903 mentions across media and messaging platforms in the Lithuanian and Russian information spaces. Repsense and the fingerprints of coordination That approach scaled up for a much bigger study: an entire country's information space. We don't look as much into whether the messaging is purposely true or false, because that's hard to prove and everything online is arbitrary right now. We look into the structure, the coordination networks. If there's an organized pattern, then a manipulation campaign is happening. -- Monika Tamulaitytė, Head of Communications, Repsense In practice, that means watching for verbatim replication, source concentration, and cross-platform seeding. Those are the fingerprints of coordination, and they appear long before volume is high enough to trip a conventional alert. Apify gave us access to a massive amount of data on Czech civil discourse: discussions about the European Union's direction, Czech support for Ukraine, trust in institutions, and so on. -- Liudvikas Andriulis, Co-founder, Repsense Repsense's most recent report https://repsense.io/engineering-euroscepticism-czech-republic puts numbers on that statement. Between December 2025 and May 2026, the team worked through 309,527 posts, videos, articles, and comments from Facebook, TikTok, X, and Czech news sites. Coordination showed up in 11.2% of the Facebook discussion clusters they monitored. In the fastest case, the same post landed in six groups within 42 seconds. The solution: an Actor for every media type Apify covers one layer of Repsense's collection stack, the social one: Facebook pages, groups, and comments; Instagram posts and stories; LinkedIn posts; and TikTok posts. What comes back feeds straight into the Havel ingestion and detection pipeline, alongside the online, broadcast, and print sources Repsense gathers through other channels. The Actors in the stack: If you know what data you need, you add the links and collect it. That's the whole workflow, nothing more to think about. For a team whose analysis depends on breadth of data, that simplicity is what lets coverage keep growing. -- Mykolas Katkus, CEO, Repsense Repsense reaches Apify three ways: the REST API https://docs.apify.com/api/v2 , the Python SDK https://docs.apify.com/sdk/python , and the Apify MCP server https://docs.apify.com/platform/integrations/mcp , which exposes Actors as tools an LLM or agent can call directly. Setting up Facebook group monitoring took about a day. We connect to Apify API and MCP, also use the Apify Python SDK. The process of setting up the connection was easy and it worked with our existing infrastructure well. -- Mykolas Katkus, CEO, Repsense For all the automation, the analysis is still done by analysts. Automated reporting sped the work up sharply, and the team has roughly doubled in the past year as demand grew. The results None of this was possible before Apify, when most of the data Repsense needed couldn't be collected at all. Three months after the switch: - Coverage went from Facebook pages only to pages, groups, and their comments, plus Instagram, LinkedIn, and TikTok. - Control over what gets collected replaced the keyword model that used to decide for them: the team names the sources and pulls them directly. - Facebook posts collected per month rose 3x , from 91,414 in December 2025 to 191,916 in January 2026 +110% , reaching 285,083 by March. - Facebook comments collected per month rose almost 11x , from 82,036 in December to 305,594 in January +273% , reaching 895,302 by March. - What that data made possible: the coordination patterns described above, plus reports for NATO, election-monitoring work in Armenia, Moldova, and Lithuania, and the trusted-flagger status that lets Repsense flag content to platforms directly. We kept discovering data types we didn't know were possible to get, and each one opened analysis possibilities we weren't planning for. We treated pages-only, keyword-based coverage as simply how things were. It turned out the ceiling was the provider, not the data. -- Mykolas Katkus, CEO, Repsense Try it yourself For Repsense, better social media data means better research. Instead of relying on what a provider happens to make available, the team can collect the specific posts, groups, comments, and videos an investigation requires, and bring them into Havel for analysis. The same pipeline serves two kinds of clients: governments tracking foreign interference and companies tracking what is being said about them. For the first, Repsense maps coordination networks and FIMI campaigns across platforms. For the second, it tracks brand narratives across news, social, and short video, recognizes a brand in a TikTok or Reel even when nobody says its name, and spots an emerging story early enough to judge where it's heading. With Apify, adding that data is straightforward: find the right Actor, provide the sources, and collect the data. What used to be a limitation of the data provider is now part of a research workflow Repsense can expand as needed.