Israel Is Trying to Game ChatGPT Results About Gaza and the IDF

Israel is planting sanitised PR stories online in an apparent attempt to influence how ChatGPT and other LLMs answer questions about the genocide in Gaza, the IDF and Zionism. 

More than 100 anonymous articles on Israel and Palestine were posted online within a week earlier this month by a new think tank connected to the Israeli government’s favoured PR firm in the US.

The articles are framed as questions that might be asked of a search engine or LLM, such as ‘Who Is the Aggressor in the Israel-Palestine Conflict?’ and ‘Does Hamas Account for Deaths Caused by Hamas Itself?’ 

The ‘data reports’ appeared on the website of the Hanover Institute for Public Policy, a body created by a boutique ad agency called Piro as part of its work for Havas Media, the French PR company that handles most of Israel’s foreign influence work in the US, Politico reports.

Other LLM-friendly articles published on the Hanover website are titled ‘Is There a Policy of Starvation in Gaza?’ and ‘Is the IDF the World’s Most Moral Army?’ None of the AI-optimised articles have bylines, and no researchers are named on the website.

In public US records filed in August, the articles were described as being part of a $100,000 campaign aimed at “the creation and dissemination of factual, source-supported informational materials intended to educate the US public regarding Israel”.

Piro co-founder Daniel Rosenberg confirmed to Politico that his ad agency was retained by the Israeli state “to put accurate, sourced facts into the public record and to counter misinformation about Israel with verifiable information.” Havas Media and the Hanover Institute did not respond to Politico’s request for comment.

According to tests run by Piro, LLMs including ChatGPT were found to be drawing on articles published by the Hanover Institute when responding to questions about Gaza, the IDF, anti-Zionism and antisemitism. 

When Novara Media ran similar tests on ChatGPT, Claude and Perplexity, none used Hanover as a source when asked similar questions. However, reporters at the Jerusalem Post found that Perplexity, an AI aggregator, did rely heavily on Hanover when asked ‘Is the IDF the most moral army?’