Fake Thinktank Funded by Israel Published 560,000 Words in Nine Days to Game AI Chatbots

A pro-Israel messaging website badged with the name of a thinktank that does not exist published more than half a million words in nine days, built on a commercial platform that promises to optimize content so AI chatbots will cite it. That is the headline finding of a Guardian investigation published on 26 August 2026, and it may be the clearest documented case yet of a state-linked influence operation deliberately targeting generative AI systems rather than human readers.

The site operates under the name of the Hanover Institute for Public Policy. According to the Guardian, it apparently does not exist as a legal entity in any jurisdiction, has no physical address, carries no named staff, and publishes no bylines on any of its reports. Its own terms of use are governed by the laws of “the state in which the Institute is established” — a state the site declines to name. After the site’s existence was first reported on 14 August, a funding page appeared with a remarkable correction: an earlier version had claimed the Institute operated without external funding, and the new text conceded that “that was true when it was written.”

124 reports, 4,500 words each, all shaped like chatbot prompts

The scale and tempo of the operation are what make it notable. A Guardian analysis found that between 6 and 14 August the Hanover Institute published 124 reports totaling more than 560,000 words — an average of roughly 4,500 words per report. On 12 and 13 August alone it produced 73 reports amounting to almost 354,000 words in two days. Nothing has been published since the first media reports about the site appeared.

The format is telling. All but one of the 124 headlines opens with a question resembling one a user might type into a chatbot: “Is Anti-Zionism Antisemitism?”, “Did Israel Expel Palestinians From Their Land?”, “Is Israel Carrying Out a Genocide in Gaza?” The reports are presented in a quasi-academic style and punctiliously sourced, citing the World Bank, UN agencies, the Palestinian Central Bureau of Statistics, Amnesty International, and the text of the Genocide Convention. The polish is the point — Nick Cleveland-Stout, a research associate at the Quincy Institute who has reported on the site, said his first impression was that it looked impressively academic, but “the more you read it, you’re able to realise that a lot of it makes no sense.”

The sourcing is real; the framing is the payload. Where human rights organizations have concluded that Israel is committing genocide or apartheid, the reports reliably observe that no court has entered a verdict on the claim — a formulation the Guardian counted in 20 of the 124 reports. Casualty and aid figures are attributed to “a party to the events,” a caveat used in 55 reports and applied as readily to Israel’s own counts as to Palestinian ones. The cumulative effect is to leave the reader — or a language model summarizing the page — with the impression that little about Gaza can be established at all.

One statistic stands out: a single 2022 study of antisemitic comments on the Facebook pages of British, French, and German media outlets is cited 454 times across 88 of the 124 reports, including on pages about land registration in the occupied territories, AIPAC’s influence on US media coverage, the Gaza death toll, the 1948 expulsions, and the blockade of Gaza. Almost every article, Cleveland-Stout noted, ends by suggesting that merely discussing the topic could contribute to antisemitism.

Follow the money: FARA filings and a Frankfurt intermediary

The paper trail runs through the US Justice Department’s Foreign Agents Registration Act (FARA), the 1938 law requiring anyone working for a foreign government to disclose it. The New York media production company Piro Inc filed the Hanover Institute material this month as work distributed for the Israeli government, and is contracted for $1 million across two work orders.

But that sum is dwarfed by the wider campaign. Piro is a subcontractor to Havas Media Germany GmbH, a Frankfurt company held through Havas’s Spanish arm and ultimately by Amsterdam-listed Havas NV. Justice Department filings record Havas paying Clock Tower X — the firm run by former Trump campaign manager Brad Parscale — $15 million in six installments between October 2025 and March 2026, and Targeted Communications Global $9.89 million between October and December 2025. In Israel, procurement records show the government advertising agency LaPam awarding Havas entities contracts without competitive tender, including $58.3 million in March 2023 for planning and purchasing media abroad — though no award covering the period in which Havas was paying $24.9 million to American firms could be found in the published record.

The structure has a convenient side effect. Seven days after Havas’s first payment to Clock Tower X, Havas’s own US subsidiary terminated its registration as an Israeli government agent. Once the German entity began paying American contractors directly, disclosure obligations fell away, leaving only piecemeal filings from individual contractors rather than a visible picture of the campaign as a whole.

The GEO layer: an industry built to get cited by LLMs

Perhaps the most technically interesting detail is the platform underneath. The Guardian’s analysis found the site initially carried an llms.txt file — a machine-readable manifest that tells AI systems what a site contains and how to read it — but it was not written for the Hanover Institute at all. It was the standard file of Res, an “AI-native content platform” that “helps B2B teams get cited by ChatGPT, Perplexity, Claude, and Gemini,” selling what the industry calls generative engine optimization (GEO). Res’s founder, Hai Tran, registered with the Justice Department on 22 July as a subcontractor to Piro on the Israeli account, listing his position as “digital strategist.” After the Guardian archived the file on 13 August, it was replaced with a bespoke version, and Res is no longer named on the site.

Piro’s own website advertises a service called AI Story Optimization, under which it will “author content engineered for how LLMs evaluate credibility,” with the work shipping “on your site or on trusted third-party properties we build.” In other words, the supply chain for this influence operation is partly off-the-shelf: a media production company, a GEO platform, and a FARA filing.

Cleveland-Stout argues the more consequential target is not live search citations but training data itself. There are two ways to do it: publish sites and hope chatbots with web access cite them, or — “the more concerning and potentially more effective thing” — get material into repositories such as Common Crawl that supply the underlying training data for commercial models. “When it regurgitates narratives the chatbot is trained on, it won’t even cite the sources,” he told the Guardian. “You won’t be able to fact-check it.” The precedent is already there: seven sites run by Clock Tower X appeared 294 times in Common Crawl’s July index. The Hanover Institute’s reports went up after that crawl closed — it is at the start of the same pipeline.

Does it work?

Early evidence suggests scrutiny has had an effect. When the Guardian asked ChatGPT’s web-connected chatbot for new research published by the Hanover Institute, it offered links to four reports but also warned that the institute is subject to “significant controversy over who funds it and why it exists,” citing the Quincy Institute’s reporting. Perplexity, asked about the site’s foreign-agent disclosure, reportedly suggested users could visit the website and judge its trustworthiness for themselves — a response Cleveland-Stout greeted with skepticism: “But how many people are doing that?”

The operation also appears to be contested at home. Israeli officials briefed Ynet in July that the public-diplomacy campaign had failed, with one saying the government “paid a lot of money, but the situation only got worse.” A former senior figure in Israel’s public diplomacy apparatus called it “a terrible hoax on the scale of Pollard.” Foreign Minister Gideon Sa’ar had allocated $181.6 million to international public diplomacy on taking office, followed by a further $666.3 million for 2026 and 2027.

Why this matters beyond one conflict

Strip away the geopolitics and the structural lesson is general. The economics of influence have shifted: search engine optimization fought for rankings; GEO fights for citations and for a presence in training corpora. A website that looks academic, cites real sources, and is structured as question-and-answer pages is precisely the format large language models are trained to treat as high-quality reference material. The defense — a footer disclosure filed under an 88-year-old law and a chatbot’s hedged warning — is thin against an operation that published 560,000 words in nine days.

For AI developers, the case is a concrete demonstration that provenance signals (FARA disclosures, funding pages, llms.txt manifests) are now attack surface. For everyone else, it is a reminder that the next time a chatbot summarizes a contested topic with unusual confidence and a tidy citation, the citation itself may have been engineered — written not to persuade you, but to persuade the model.