Israel-linked PR built a website to teach ChatGPT about Gaza. It's working.

Ask ChatGPT whether there's a policy of starvation in Gaza, and the answer may quietly cite a "data report" from an institute that lists no authors.

The Hanover Institute for Public Policy is funded by Israel's government advertising agency through a chain that runs Havas Media Germany to Piro, Inc., a U.S. subcontractor, according to U.S. Justice Department Foreign Agents Registration Act (FARA) filings. FARA is the law that requires agents of foreign governments doing political or public-relations work in the United States to register and disclose their activities; registration 7732 was filed in June 2026.

The campaign's purpose, as stated in the FARA registration, is "the creation and dissemination of factual, source-supported informational materials intended to educate the U.S. public regarding Israel and related issues through publicly distributed content." The disclosure names a roughly $100,000 budget. The materials are not aimed at journalists or television ad slots. They are aimed at large language models (LLMs) such as ChatGPT and Perplexity, the conversational AI assistants that now answer a growing share of the questions readers ask about current events.

Politico Influence, the Jack Blake newsletter that broke the story on August 14, reports that when its reporters ran neutral prompts on ChatGPT and Perplexity, both assistants cited the Hanover Institute's material in their answers. The campaign is not just live in the U.S. market; it is live at the retrieval layer where AI assistants decide which sources to surface.

The Hanover Institute publishes short articles it calls "data reports" with titles formatted as questions: "Is the IDF the World's Most Moral Army?" and "Is There a Policy of Starvation in Gaza?" That Q&A pattern mirrors the way users actually prompt chatbots. The site's footer disclosure names the full chain: Piro, Inc. on behalf of Havas Media Germany GmbH on behalf of the Israel Government Advertising Agency, known in Hebrew as LaPam. The articles themselves list no named authors and carry no byline.

Reporting by Politico and the trade publication O'Dwyer PR identified Res, a Seattle-based AI startup that markets "AI recommendation" optimization to clients including Nike and Intel, as the host of the Hanover Institute's site. Res's homepage once carried a mirror of the Hanover Institute content; the mirror went offline after Politico reached out for comment. The combination of a Q&A-shaped article corpus, an unstaffed masthead, and a vendor whose core product is positioning content inside AI answers is a campaign designed to be picked up, not read.

The Politico newsletter's reporting references prior chatbot-influence work involving former Trump campaign manager Brad Parscale, suggesting a continuing pattern rather than a one-off. The FARA short-form filings for the principals (Exhibit AB, Short-Form 1, and Short-Form 2) name the individuals who will personally act on the foreign principal's behalf. A separate, earlier Havas FARA registration 7662 covers related LaPam work filed in late 2025.

The FARA disclosure is the receipt, not a defense. A campaign can be legal, on the public record, and still represent a new category of foreign influence inside the products readers use to learn about the world. The disclosure tells readers that the material came from a foreign government; it does not tell the AI assistant that. When ChatGPT cites a no-author "data report" without a footer disclosure, the reader cannot tell whether the citation came from a foreign-agent-registered campaign, an advocacy group, an industry analyst, or a journalist.

For readers who want to pressure-test their own AI answers, the receipts are public. Search the DOJ FARA database for registration 7732 to see the full contract and the parties involved. Before trusting a cited source inside an AI answer, check whether it has named authors; check the site footer for a foreign-agent disclosure; and run the same prompt on a second model and compare the citations. If the cited material is real, it will usually reappear; if it is shaped for LLM retrieval, it usually will not.

Hanover is the first FARA-disclosed campaign with on-record test results showing that consumer AI assistants cite its content in response to neutral prompts. The FARA paperwork, the no-author Q&A corpus, the Res-built site, and the $100,000 budget are all public. The next prompt a reader types about Gaza will go through the same retrieval layer the Hanover campaign was built to feed.