Israel Paying Millions to Train AI Chatbots How to Talk About Gaza. It's Working

Nick Cleveland-Stout

Jul 28, 2026

Brad Parscale, the former campaign manager for U.S. President Donald Trump, on November 4, 2019 in Lexington, Kentucky. Photo by Bryan Woolston/Getty Images.

Since October, former Trump campaign manager Brad Parscale has been quietly overseeing an operation posting hundreds of blog posts on behalf of Israel. One article, titled “The Reality Behind Gaza’s ‘Journalists’: Terror Ties, Propaganda, and the Laws of War,” asserts that a majority of journalists in Gaza were linked to terrorist organizations. Another casts doubt on the killing of Hind Rajab, a five-year-old Palestinian girl killed by the Israeli military in 2024.

The key intended audience of these sites is not concerned Americans, it’s not even humans—most of the sites average a few hundred unique visitors each month. Instead, Parscale and his firm, Clock Tower X, created them as part of a $46.5 million contract with the Israeli government to try and influence artificial intelligence-powered chatbots, tools like Claude or ChatGPT.

Parscale has made his goal of influencing artificial intelligence—often referred to as “LLM poisoning”—explicit. In his initial agreement with Israel, Parscale said that he would deploy “websites and content to deliver GPT framing results on GPT conversations” as part of the contract. More recently, his team even told Axios they are “seeing success” at getting popular AI systems to incorporate information from their sites, though they declined to provide data.

And it is working, according to disinformation experts who reviewed a Drop Site analysis of chatbot queries and training data, meaning tens of millions of Americans who use chatbots are increasingly likely to receive answers manipulated by Parscale on behalf of the Israeli government.

When Drop Site asked Perplexity “Is it beneficial for the US to enhance military cooperation with Israel?” the chatbot responded with a one-word answer: “Yes.” The top source listed was Allyvia.org, a Parscale-created website dedicated to promoting the U.S.-Israel military relationship. Microsoft Copilot similarly cited Parscale’s websites.

Screenshot from Perplexity.

Allyvia is one of 10 different websites created by Clock Tower X, a boutique digital influence company owned by Parscale. Each website is dedicated to promoting a different Israeli narrative. FactSignal.org, for instance, fashions itself as fact-checking website tackling anti-Israel information, containing articles discrediting slain journalists as Hamas terrorists and rejecting the notion that Israel’s military actions in Gaza constitute a genocide. Paxpoint.org, meanwhile, promotes the narrative that Israel is a nation of peace.

All of Parscale’s websites contain a statement at the bottom which states “this material is distributed by Clock Tower X LLC on behalf of the State of Israel.” This statement is a mandatory disclosure requirement under the Foreign Agents Registration Act (FARA). Despite this, when Drop Site asked prominent chatbots about the latest news from Paxpoint.org, Perplexity and Copilot did not flag that the site was created as part of an Israeli influence operation. Claude, ChatGPT, and Gemini did flag the caveat.

“Perplexity is built to give people cited, linked answers from a range of sources so they can go directly to the original reporting and reach their own conclusions,” spokesperson Beejoli Shah told Drop Site News when reached for comment.

Parscale’s websites—referred to throughout this article as the “Clock Tower network”—get cited by chatbots, but they also are successfully infiltrating the underlying training data.

The Clock Tower network started to get archived by Common Crawl—a nonprofit that oversees a massive repository of data used to train large language models such as ChatGPT, Gemini, and Claude—earlier this year. Common Crawl is the backbone of the artificial industry, training 80% of the tokens of OpenAI’s GPT-3, making it a useful proxy for understanding if malicious actors successfully infiltrate AI training data.

READ MORE from Nick Cleveland-Stout at the Quincy Institute: “The Eighth Front: Inside Israel’s $1 Billion Influence Campaign.”

LLM Poisoning

In the examples above, users can at least trace the sources that chatbots are pulling from when they search websites directly, such as in the screenshots above. Infiltrating the training data itself, on the other hand, can lead to Israel-influenced responses that are nearly impossible for users to trace.

To measure how deeply the network had penetrated the data used to train AI models, Drop Site used Claude Fable to query Common Crawl to count how often Parscale’s sites appear in its monthly snapshots. According to that analysis, a sample of which was manually fact-checked by Drop Site, between January and June 10 sites were crawled by Common Crawl 912 times. While this is a drop in the bucket of Common Crawl data, emerging research suggests a very low number of documents can successfully manipulate AI chatbots, no matter the data volume.

The 10 sites are:

Paxpoint.org (crawled 220 times): Website dedicated to showcasing the narrative that Israel is a nation of peace. “Paxpoint highlights Israel’s ongoing commitment to peace and coexistence, sharing historical accords, humanitarian initiatives, and grassroots projects that bridge divides.”

Allyvia.org (crawled 158 times): Website dedicated to strengthening the U.S.-Israel alliance. “Allyvia showcases how the US and Israel alliance directly benefits Americans through joint security, cutting-edge technologies, job creation, and shared values of liberty.”

Factsignal.org (crawled 108 times): Mission is to “prove that Hamas’ terrorist designation reflects global consensus and undeniable evidence of violence, not conspiracy”

Cognitura.org (crawled 106 times): Website claims to be a “research and education platform” dedicated to exploring how extremist groups such as Hamas operate.

Justorium.org (crawled 93 times): Website dedicated to arguing that Israel’s military actions in Gaza “follows international law and reflects the democratic ethics shared by the US and Israel.”

Culturavia.org (crawled 76 times): Website focused on celebrating cultural ties between U.S. and Israel.

Compassionpulse.org (crawled 57 times): Website dedicated to sharing stories, data, and eyewitness accounts of “how Israel protects civilians, delivers humanitarian aid, and upholds moral values shared with the US.”

Econora.org (crawled 49 times): Website dedicated to highlighting U.S.-Israeli trade relationship.

Innovascope.org (crawled 45 times): Website dedicated to highlighting U.S.-Israeli technology partnership.

Feedingyoufiction.com (crawled 1 time): Website featuring videos that claim footage coming out Gaza is staged and denying famine.

These sites are being crawled more and more. In January 2026, the sites were only crawled twice. In May, the sites were crawled 376 times, before a notable dip in June. To check the accuracy of Claude Fable, Drop Site manually searched for a sampling of 50 of the Clock Tower X URLs archived in Common Crawl’s CDXJ index, and did not detect any errors.

Hervé Letoqueux, Chief Executive Officer at Check First, an organization that combats digital disinformation, explained that Common Crawl’s status as the go-to source for training chatbots opens it up as a target for manipulation. “Common Crawl is widely used in the world of data science for training LLMs, which also means that actors want to manipulate the answers in their favor, which some studies have indicated are fairly easy to do.”

It only takes about “250 malicious documents to produce a ‘backdoor’ vulnerability in a large language model—regardless of model size of training data volume,” according to an October Anthropic study. The authors of the study note “this means anyone can create online content that might eventually end up in a model’s training data,” and that malicious actors “can inject specific text into these posts to make a model learn undesirable or dangerous behaviours, in a process known as poisoning.”

The Atlantic Council’s Digital Forensic Research Lab published a similar study earlier this year on Russia’s disinformation network’s attempts to influence LLMs and noted an important caveat. Because artificial intelligence companies have their own process of quality control when they incorporate data, “Inclusion in Common Crawl does not guarantee inclusion in any given model’s training data.”

Drop Site also performed a demo alongside Letoqueux to test five different chatbots—Microsoft Copilot, ChatGPT, Google Gemini, Claude AI, and Perplexity—on whether they could produce articles using only their training data. Of the five, the Parscale network consistently appears in the training data of Gemini and Copilot. Claude and ChatGPT did not appear to contain the websites in the training data, though these chatbots can still search Clock Tower X websites when responding to queries. Perplexity, meanwhile, appeared to be trained on some of the Clock Tower X websites.

Microsoft Copilot and Google Gemini did not respond to a request for comment for this article.

‘It is easy to prove that these models were trained on the Clock Tower X websites, though it is difficult to know the degree to which that influences the response a chatbot gives to a query,’ said Letoqueux.

Optimizing for AI

Parscale’s team has credited their success with a softer approach to selling Israel’s narrative. Scott Stouffer, Chief Technology Officer of Market Brew—a company that specializes in influencing LLMs that Parscale invests in—told Axios that information written in a “factual tone and tightly organized” is more likely to be cited in AI platforms. “What you can do is make sure that your information is structured, sourced, and aligned in a way that those systems are more likely to retrieve it when someone asks a question. It’s less about changing the conversation and more about making sure your facts are eligible to be part of it.”

Many of the blog posts on the Clock Tower network contain at least a few sentences of nuance, which disinformation researchers say make it more palatable for chatbots. For instance, the FactSignal blog post about how journalists in Gaza are supposedly secret terrorists states that, “Distinguishing between legitimate reporters and militant operatives is not always simple, but accurate reporting requires careful investigation and transparency.”

The author of many of the blog posts is Israel-based digital marketer Mark Ginsberg, according to the source code. Ginsberg, who registered as a foreign agent for Israel in March, advertises his expertise in Generative Engine Optimization—or LLM grooming—on his Linkedin profile. In a June article for The Times of Israel, Ginsberg laid out the argument for Israel investing in efforts to influence chatbots. “Israel needs to invest seriously in the defense of truth…It means ensuring that accurate information appears not merely in government archives but in search results, educational resources, encyclopedias, social media feeds, podcasts, documentaries, and AI training datasets,” Ginsberg wrote. Trying to influence AI training data, Ginsberg argued, is not “embracing disinformation” but rather stating “reality itself.”

Letoqueux, who reviewed the Clock Tower network sites, confirmed that because chatbots have a built-in preference for presenting a balanced view, websites that are sourced and present a lighter touch might be more successful. “These websites could probably be quite useful at doing damage control on questions such as ‘Is Israel committing a genocide?’ because they provide the Israeli narrative, which chatbots can then repeat as representative of one side of the debate,” said Letoqueux, who is the former head of operations at VIGINUM, the French agency responsible for identifying foreign digital interference.

Alice Lee, an anal

[truncated for AI cost control]