ISLAMABAD: Israel has spent more than $1 billion since late 2023 on public diplomacy and influence efforts aimed at shaping perceptions of Israel and its policies, according to US government filings and investigations by The Guardian, Time magazine, and the Quincy Institute for Responsible Statecraft.
The campaign includes public relations firms, apparently independent websites, paid influencers, social media networks and content designed to reach information systems increasingly used by AI chatbots.
The effort has taken on greater significance as Israel's military campaign in Gaza continues to generate reports of serious violations of international law.
The International Criminal Court (ICC) has issued an arrest warrant for Israeli Prime Minister Benjamin Netanyahu over reported war crimes and crimes against humanity in Gaza. Israel rejects the allegations and disputes the court's jurisdiction.
Israel is also defending itself in a genocide case brought by South Africa at the International Court of Justice. The court has issued provisional measures but has not delivered a final ruling determining whether genocide occurred.
A Pew Research Center survey conducted in March found that 60% of US adults held an unfavorable view of Israel, up from 42% in 2022. A February Gallup poll found that 41% of Americans sympathized more with Palestinians and 36% with Israelis, the first time Palestinians led in the poll's history, although Gallup said the five-point difference was not statistically significant.
The polling does not establish that Israel's influence campaign caused or failed to cause changes in public opinion. It provides the broader context in which the campaign's scale can be examined.
Fake think tanks
One strand of the campaign works directly on how the Gaza war is described online.
Foreign Agents Registration Act filings submitted to the US Justice Department showed New York media firm Piro Inc. built a website called the Hanover Institute for Public Policy, presenting itself as an independent American foreign policy think tank while distributing material for Israel's Government Advertising Agency, known as LaPam.
The Guardian, which exposed the operation on Aug 26, found that the institute had no legal registration, office, or listed staff.
The site published more than 560,000 words across 124 articles in nine days, according to the newspaper's investigation. The material addressed allegations of genocide, apartheid, civilian casualties and humanitarian conditions in Gaza.
Technology publication 404 Media found much of the material likely to have been AI-generated, while Israeli news site Ynet reported that 123 of the 124 reports published over nine days were phrased as questions resembling queries commonly entered into AI chatbots.
The content sought to challenge or soften allegations concerning Israel's conduct in Gaza.
Twenty reports addressing genocide or apartheid allegations repeatedly noted that no international court had entered a binding verdict, The Guardian found. Another 55 questioned casualty and hunger figures produced by humanitarian organizations by describing those organizations as parties to the conflict.
What AI systems see
The strategy resembles what researchers call Generative Engine Optimization, or GEO, a field concerned with increasing the visibility of material in responses produced by generative AI systems.
Unlike conventional advertising, the objective is not necessarily to persuade a reader directly. It is to place material into the information environment from which an AI system constructs an answer.
Drop Site News and Politico tested AI systems and found that material from websites linked to the Israeli campaign could appear in chatbot responses on Gaza and US-Israel policy.
The tests demonstrate that campaign material can be incorporated into AI-generated answers. They do not establish that it changed users' beliefs or US public opinion.
That distinction is crucial. Visibility in an AI answer is evidence of reach, not proof of persuasion.
AI-influenced industry
Developments since the original investigation show that the technology behind the campaign is becoming part of a much broader ecosystem of influence.
Anthropic's September threat-intelligence report identified and disrupted nine influence operations in which actors used Claude to create fake social media profiles, fabricated news outlets, political content and other material while concealing the origins of campaigns.
The operations involved actors from Russia, Iran, Türkiye and other regions and targeted audiences across six continents.
In one case, Anthropic identified a network of about 70 fabricated news websites that produced at least 8,913 articles in around 20 languages. The material was designed in part to improve search visibility, demonstrating how generative AI can turn influence operations into high-volume content production.
Anthropic also cautioned that many operations generated little authentic engagement and were disrupted before developing substantial audiences.
That caveat applies to the Israeli case as well. The ability to manufacture content at scale does not prove that the material has persuaded large numbers of people.
But Israel's campaign predates today's AI tools. It also builds on an older digital influence infrastructure.
Meta and OpenAI disrupted an Israeli-linked operation in 2024 involving hundreds of fabricated social media accounts and invented news outlets. Both companies linked the activity to STOIC, a Tel Aviv marketing firm. Investigations by The New York Times, Haaretz, and Israeli watchdog FakeReporter connected a $2 million contract to Israel's Ministry of Diaspora Affairs and Combating Antisemitism.
Earlier research examined Act.IL, a platform that organized volunteers to participate in online campaigns, including voting in media polls and reporting pro-Palestinian accounts.
The technology has changed, but the broader objective of shaping the information environment predates generative AI.
Influence networks
Time's investigation, based on Foreign Agents Registration Act (FARA) filings, reported that Havas hired Brad Parscale's firm, Clock Tower X, in September 2025 to run a digital campaign for the Israeli government.
The initial agreement called for 100 pieces of content a month, with at least 80% aimed at Gen Z audiences and at least 50 million monthly digital impressions. The contract later grew to more than $45 million.
Clock Tower X built websites, including Allyvia, Paxpoint, and Thetruthaboutiran, that did not prominently identify their connection to the Israeli government. Drop Site News reported that the campaign also included a chatbot-focused component.
The Quincy Institute identified 17 firms registered under FARA to represent Israeli interests in 2024 and 2025, more than in the previous eight years combined. It estimated that Israel had spent more than $100 million through FARA-registered US contracts since late 2023.
The Wall Street Journal separately reported in July that millions of Americans received AI-generated text messages from senders using names such as "Emma," "Sarah" and "John" under a group calling itself Friends for Peace.
AI debate comes home
The influence campaign is no longer confined to Israel's attempts to shape foreign perceptions.
With Israel's general election scheduled for Oct 27, AI-generated videos and synthetic political material have become part of the domestic campaign environment, too.
The Wall Street Journal reported this month that Israeli politicians are using AI-generated videos to influence voters and attack opponents, while regulators struggle to keep pace.
The development parallels the government's foreign public diplomacy efforts: AI is increasingly becoming part of both Israel's external information campaign and its domestic political environment.
Gaza remains at the center
The current AI debate also runs counter to an expanding international record of the human consequences of the Gaza war.
On Sept 4, the UN human rights office said 24 Palestinians had been killed in Israeli attacks over five days, including four children, and stressed that the pursuit of reported lawful targets does not suspend Israel's obligations under international humanitarian and human rights law.
Days later, UN Special Rapporteur Francesca Albanese warned that removal and crushing of rubble in Gaza could destroy evidence relevant to investigations of reported international crimes, including genocide.
The contrast is central to the information battle: while international bodies continue documenting civilian harm and investigating reported violations, Israeli government-linked campaigns are producing material aimed at challenging or reframing allegations surrounding the war.
The two uses of AI are distinct, but they underscore how deeply the technology has entered the debate surrounding Israel's war in Gaza.
What the evidence shows
The evidence documents a sustained effort by Israeli government-linked actors and contractors to shape the information available to US audiences through public diplomacy, apparently independent websites, social media, influencers and AI-related techniques.
It also shows that some campaign material can enter AI-generated answers.
What remains unproven is whether the campaign itself caused changes in public opinion.
But the more consequential question is whether AI systems can remain reliable information intermediaries when governments and commercial influence firms learn to engineer the information from which those systems generate answers.
As international institutions investigate reported violations arising from the Gaza war and UN officials continue documenting civilian harm and destruction, the battle over how that war is described is increasingly moving from newspapers, television and social media into the systems that generate answers themselves.
The technology may be new, but the political objective is familiar: shape the narrative around Israel's conduct toward Palestinians.
What has changed is the intermediary.
Increasingly, it is not a newspaper, television channel, or social media platform, but an AI system answering a user's question in a voice that can appear neutral even when the information beneath the answer has been deliberately engineered.