Israel is paying tens of millions of dollars to seed pro-Israel narratives about Gaza into artificial intelligence systems, using a network of apparently independent websites designed to influence chatbot responses and future training data, according to an investigation by Drop Site News.
The operation is being overseen by Brad Parscale, a former campaign manager for US President Donald Trump, through his digital influence company, Clock Tower X. The firm holds a $46.5 million contract with the Israeli government and has published hundreds of articles since October 2025 as part of what experts describe as an attempt to “poison” large language models.
The content promotes Israeli state narratives while seeking to undermine evidence of Israel’s atrocities in Gaza. Articles published through the network have attempted to smear Palestinian journalists killed by Israel as members of armed groups, reject accusations that the apartheid state is committing genocide and cast doubt on the killing of Palestinian child Hind Rajab by Israeli forces in 2024.
Parscale made the intention to target artificial intelligence explicit in his initial agreement with Israel. The contract promised to deploy “websites and content to deliver GPT framing results on GPT conversations”, according to Drop Site.
Clock Tower X created ten websites, each dedicated to a particular aspect of Israel’s propaganda campaign. Paxpoint.org portrays Israel as committed to peace, Allyvia.org promotes US-Israel military cooperation and FactSignal.org presents itself as a fact-checking platform while publishing material which discredits Palestinian journalists and denies genocide in Gaza.
Other sites claim that Israel complies with international law, protects Palestinian civilians and facilitates humanitarian assistance, despite extensive evidence from UN agencies, humanitarian organisations and international rights groups documenting attacks on civilians, restrictions on aid and the destruction of essential infrastructure.
The websites attract only a few hundred human visitors each month, Drop Site reported. Their principal audience appears instead to be search engines, automated web crawlers and AI chatbots.
When Drop Site asked Perplexity, an AI-powered search engine and chatbot, whether greater US military cooperation with Israel would be beneficial, it replied simply, “Yes”, citing Allyvia as its leading source. Microsoft Copilot, Microsoft’s AI assistant, also cited websites from the Clock Tower network..
In another test, Gemini, Copilot and Perplexity cited Clock Tower websites when answering a question about US-Israeli technology cooperation without informing users that the sources had been created on behalf of the Israeli government.
US law requires each website to display a Foreign Agents Registration Act notice identifying Clock Tower X as an agent of Israel. However, Perplexity and Copilot did not always disclose this relationship when presenting the sites as sources. Claude, ChatGPT and Gemini flagged the Israeli government connection when asked about material from Paxpoint.
READ: ‘Things have only gotten worse’: Israel fumes as millions fail to stop collapse in US support
The operation has also begun penetrating Common Crawl, a vast public archive of online material used in the development of many large language models.
Drop Site used an AI tool to examine Common Crawl’s monthly records and found that the ten Clock Tower websites were crawled 912 times between January and June. The number rose from only two crawls in January to 376 in May. Researchers manually checked a sample of 50 archived links and reported no errors in the results.
This does not prove that every AI company has incorporated the material into its models. The Digital Forensic Research Lab has stressed that inclusion in Common Crawl does not guarantee inclusion in any particular model’s training data because developers apply their own filtering, quality checks and data-selection processes.
Nevertheless, research suggests that actors do not necessarily need to flood the entire internet to contaminate AI training.
A joint study by Anthropic, the UK AI Security Institute and the Alan Turing Institute found that as few as 250 malicious documents could create a backdoor vulnerability in language models, regardless of the size of the models or the overall volume of their training data.
The study showed that “A small number of samples can poison LLMs of any size”.
The study also tested Microsoft Copilot, ChatGPT, Google Gemini, Claude and Perplexity to assess whether the Clock Tower material appeared in their underlying training data. The investigation found evidence of the network’s content in Gemini and Copilot, while Perplexity appeared to contain some of the websites.
The tests did not find evidence that ChatGPT or Claude had trained on the network’s material, although both systems could retrieve and cite the sites while searching the internet.
Hervé Letoqueux, chief executive officer of the disinformation monitoring organisation Check First, cautioned that while researchers could identify evidence that some models had encountered the websites, it remained difficult to measure how strongly the material influenced individual answers.
Experts told Drop Site that Clock Tower appears to format its articles in ways that make them more likely to be picked up by AI chatbots. The websites use clear summaries, bullet points and numerous links to other sources.
Alice Lee, an analyst at NewsGuard, said the material is also harder for AI systems to identify as propaganda because it often mixes opinion, disputed claims and selected facts, rather than relying on statements that are obviously false.
Clock Tower X’s Israeli contract has expanded from $1.5 million to $4.5 million a month. The operation forms part of a much wider Israeli campaign involving online influencers, conservative media, targeted advertisements and mass text messages aimed at reversing the collapse of public support for Israel in the US.
READ: Israel floods US phones with AI messages in campaign to manipulate public opinion
