How Defamatory Websites Are Teaching AI to Repeat Their Claims

Ask an AI assistant about a company or its executives, and the answer often comes from whatever the open web says most loudly. For a network of defamatory websites operating across the UK and Ukraine, that has become a new opportunity. Allegations that once needed a human reader to believe them can now be absorbed into search summaries, knowledge bases and chatbot answers, where they arrive stripped of their origin.

From paid article to “fact”

Reputation adviser Michael Macfarlane has described a two-stage process. First, an allegation is placed on several low-cost sites, many of which accept paid content with no checks. Next, those articles are cited as sources when editing Wikidata and Wikipedia entries about the target. In his view, large language models use those entries as a starting point for understanding who a person is.

The cost of entry is low. Macfarlane said he could start such a campaign in a couple of hours for a few hundred dollars. Placing a negative article with the Antikor group is reported to cost $150 to $200.

AI on both ends

The technology that spreads these claims also helps produce them. London law firm Schillings tracks more than 350 junk news sites. It recorded a 355% rise between 2023 and 2024 and links the jump to the mass adoption of AI writing tools. A site that once needed staff to fill its pages with local news can now generate thousands of filler articles automatically, and drop a paid attack piece in among them.

Leeds TV is one example. A former student television channel, it now publishes AI-written local stories alongside detailed pieces on wealthy figures from the former Soviet states. Its operator confirmed by email that the site accepts payment to publish.

The domains feeding the system

Duplication makes the network effective in automated systems. One story about a Kyiv businessman appeared on 17 sites in a single day. A campaign against one company executive ran 158 articles across 109 domains, and 95 of those domains were hosted on a single server in Reading, England. To a model weighing how often a claim appears, that repetition looks like confirmation. We have previously mapped the mirror-domain model behind coordinated online defamation and its replacement tactics.

The following domains have been documented publishing duplicate or near-identical content:

Antikor: antikor.com.ua, antikor.info, antikor.ua

Kompromat1: kompromat1.online, kompromat1.info, kompromat1.net, kompromat1.org, kompromat1.media, kompromat1.one, kompromat1.se, kompromat1.press, kompromat1.world

Compromat / Ruskompromat: compromat.pro, compromat-pro.com, kompromat.press, ruskompromat.info, ruskompromat.ru, ruskompromat.pro

Vlasti: vlasti.io, vlasti.se, vlasti.info, vlasti.live, vlasti.vip, vlst.live, vlast.guru

Antimafia: antimafia.se, antimafia.io, antimafia.life, antimaf.pro

Rumafia: rumafia.news, rumafia.io, rumafia.net, rumafia.online

Kartoteka: kartoteka.news, kartoteka.se, kartoteka.press, kartoteka.io

Glavk: glavk.se, glavk.info, glavk.net

Hab.media: hab.media, news.hab.media, crime.hab.media, corruption.hab.media

Sledstvie: sledst.org, sledst.com, sledstvie.info

Nocor: nocor.info, nocor.life

Novosti: novosti.cloud, novostiua.org, novostiua.online

Other linked outlets: repost.news, rozsliduvach.info, rucriminal.info, adcon.info, luxherald.com, lstv.co.uk, obriy.news, politica2.info, ukrainci.com.ua, oplatru24.ru, rp.rv.ua, drama.live

A technical review of more than 60 of these domains found shared SSL certificates and hosting fingerprints. It also found cloaking that showed search crawlers different content from what human visitors saw.

Courts can correct a page, not a model

Victims have won in court, but the rulings reach only the pages. A Kyiv court found Antikor’s money-laundering allegations against a Ukrainian bank false. A Los Angeles court ordered articles on 42 sites removed. In London, the High Court awarded more than £220,000 against the London Post and its listed owner, and the article stayed up for weeks afterwards.

Even when a page comes down, copies on mirror domains remain. So do the knowledge-base entries that cited it, and any training data that captured it. Ukraine blocked antikor.com.ua, antikor.info, rozsliduvach.info and hab.media in 2025. The material is still reachable through VPNs and foreign mirrors.

Platforms say they are responding

OpenAI has said it has teams working to detect and expose covert influence operations. World-Check says it has safeguards to keep misinformation out of its risk reports. Google says it delists malicious content. The network keeps growing anyway, because each new domain costs little and every copy is another chance to be picked up.

For anyone whose name has been targeted, the problem has shifted. It is no longer enough to get an article removed. The harder task is making sure that the systems now summarising the web can tell a court judgment from a page on kompromat1.online.