According to an investigation by Responsible Statecraft reporter Nick Cleveland-Stout, the Israeli government spent $900,000 through public relations firm Piro to build a fake think tank called the Hanover Institute, with the goal of influencing how AI chatbots answer questions related to Gaza.
Piro framed the effort as “AI Story Optimization”
The report says the Hanover Institute published more than 100 articles in 11 days. The pieces carried no author names and were described as content built not for human readers first, but for large language models that assess credibility and reuse online sources.
Piro reportedly referred to the method on its website as “AI Story Optimization.” In the report’s account, the strategy involved producing polished, neutral-sounding articles with structured formatting and source-heavy presentation so that systems such as ChatGPT and Gemini would be more likely to cite them in answers about Gaza and Israel.
The report ties the tactic to LLM poisoning
The article describes the method in industry terms as “LLM poisoning,” meaning an attempt to feed models a large volume of orderly, authoritative-looking material so it shapes the basis of later answers. It compares the logic to Anthropic’s recent research on small-sample poisoning, while noting that the target here was not a security weakness but public opinion.
The Hanover Institute’s about page says it studies the causes of antisemitism in the United States. Cleveland-Stout randomly selected 12 articles from the site and ran them through GPTZero. Eleven were classified as highly likely to be AI-written, and one received a medium-confidence result.
Copilot and Gemini were said to have used the content
A separate investigation by Drop Site found that Microsoft Copilot and Google Gemini had used material from the network for training. The report also says other chatbots frequently cited the sites without indicating that the material was part of an Israeli influence effort.
According to the article, Hanover Institute reports often opened with questions that appeared neutral on the surface, such as asking what caused the displacement of Palestinians in 1948, then wrapped the argument with polling figures and government sources.
NewsGuard analyst Alice Lee said sites like this “perfectly mimic the look of an American think tank,” including a red, white, and blue color scheme. The report adds that large language models tend to favor content with specific figures and cited sources, which matched the format of Hanover’s reports.
Focus on manipulation as AI becomes an information gateway
The piece argues that as AI chatbots become a first stop for many people trying to understand the world, money may buy more than ad placement. It may also buy answers that appear neutral and objective on the surface.