Ankara: A new analysis found that a pro-Israel website under the guise of a fake think tank has published hundreds of reports to bias how artificial intelligence (AI) chatbots present information about Israel and the Gaza war.
According to Anadolu Agency, a Wednesday analysis by The Guardian revealed that the website, operating under the name of a think tank that appears not to exist as a legal entity, published more than half a million words in just nine days in an effort to shape AI responses. The website, under the invented name 'Hanover Institute for Public Policy,' published 124 reports totaling over 560,000 words on August 6-14, averaging 4,500 words per report. It released 73 reports, totaling nearly 354,000 words, on August 12 and 13 alone.
The website presents Israel's positions on issues including the torture of Palestinian prisoners, Israeli war crimes, and whether Israel has deliberately starved Palestinians in Gaza, all falsely presented as 'neutral' research. The website has no physical address, named staff, or bylines, and the 'institute' does not appear to be registered as a legal entity in any jurisdiction.
The Guardian reported that the website is part of a broader Israeli government-funded campaign aimed at influencing AI chatbots, with tens of millions of dollars reportedly routed through third parties, including ad company Havas Media. Nearly all of the website's headlines are reportedly phrased as questions resembling AI chatbot queries, such as 'Is Anti-Zionism Antisemitism?' and 'Is Israel Carrying Out a Genocide in Gaza?'
The reports use an academic style and cite sources including UN agencies, the World Bank, and Amnesty International, reportedly in a style to attract AI chatbots. However, The Guardian found that the reports consistently frame information in ways that reduce criticism of Israel. For example, references to Israel's genocide and apartheid are repeatedly followed by statements noting the absence of a final court ruling on the claims.
Nick Cleveland-Stout, a research associate at the Washington, DC-based Quincy Institute, told The Guardian that the site initially appeared highly professional but that closer examination showed 'a lot of it makes no sense.' Cleveland-Stout said the campaign's more significant aim may be to influence AI training data. He added that the campaign could have a greater impact if its content enters repositories such as Common Crawl, which can contribute to AI training datasets.
"When it regurgitates narratives the chatbot is trained on, it won't even cite the sources," he said. "You won't be able to fact-check it." Cleveland-Stout added that the institute "is and will continue to have success at influencing the responses that chatbots give." "But they're not changing the polls, which is what Israel actually cares about," he said, perhaps referring to growing public disapproval worldwide of Israel's genocidal policies.