I clicked a link and landed on what looked like a sober policy site. Within seconds the numbness set in: pages and pages of policy-speak, no authors, and a file expressly optimized for large language models. I felt the room tilt—something was being assembled to meet the machines, not human readers.
I want to be blunt with you: this is not ordinary public relations. You and I are watching a campaign that borrows the playbook of SEO and rewrites it for AI. The headline: since 2023 Israel has poured at least $100 million (€93 million) into U.S. influence efforts and roughly $1 billion (€930 million) globally, according to Politico. That money has a new target — the answers you get from chatbots like OpenAI’s ChatGPT and Anthropic’s Claude.
A torrent of text arrived in nine days — Why 124 reports and over 560,000 words were published
Imagine a content factory spitting out volumes designed to be scraped by algorithms. The Guardian counted 124 reports published over nine days, more than half a million words. The scale matters because LLMs prefer repetition and citation density; the more sources that echo a claim, the likelier it is to surface in a chatbot reply.
How is Israel influencing AI chatbots?
Short answer: through paid PR and engineered content that search engines and LLMs ingest. French agency Havas Media, registered under the Foreign Agents Registration Act (FARA), and its subcontractor Piro Inc. built the Hanover Institute for Public Policy site and seeded it with material described in FARA filings as meant to “educate the U.S. public regarding Israel and related issues through publicly distributed content.”
No office, no bylines, no staff listed — What the Hanover site actually is
Visitors find a tidy policy site with slick layouts but no legal footprint. The Guardian reports the Hanover Institute does not exist as a legal entity and lists no physical address or named researchers. The site’s defense is explicit: “Accountability for the material is not anonymous,” the group claims about its anonymous articles.
That sentence reads like a paradox because the site also states researchers are unnamed to avoid harassment and to let findings stand on sources rather than reputation. The result: content that reads like research but lacks the transparency you expect from a think tank.
Is the Hanover Institute a real think tank?
Not in any traditional sense. There’s a URL, a brand, a stack of reports—but no registration, no disclosed staff, and no physical office. Reporters at The Guardian, 404 Media, and Gizmodo have flagged the mismatch between production volume and institutional reality.
An optimization file was left open — How GEO techniques target LLMs
On a technical level the site is explicit about its audience: Piro left an llms.txt file intended to make Hanover’s output easier for language models to ingest. Piro even lists an “AI Story Optimization” service on its site, the same toolkit used to massage messaging for platforms and chatbots.
Generative Engine Optimization (GEO) is the practice: engineer content so models index and repeat it. That’s SEO remixed for synthetic answers. One vivid image: content is being planted like seeds in the soil of search and model training, meant to sprout where users will see it.
The PR chain — Havas, Piro, FARA filings and the ghostthink-tank model
Regulatory traces exist: Havas Media’s FARA filings name Piro Inc. as a subcontractor. The filings describe the aim plainly—to place pro-Israel narratives into the public square, and by extension into the data LLMs draw from. The campaign’s framing includes claims that aid reached Gaza and arguments equating anti-Zionism with anti-Semitism, positioned to contest widespread reporting of severe civilian harm.
Journalistic outlets from Politico to The Guardian and niche trackers like 404 Media have followed the paper trail; PBS and Gaza’s Health Ministry offer the casualty and humanitarian context that the seeded material often rebuts. The campaign uses mainstream PR machinery to create an archive of apparently authoritative sources.
Public opinion and the motive — Why this matters to U.S. audiences now
Polls changed the calculus. Pew Research shows 60% of American adults hold negative views of Israel, rising to 70% for people under 50. Half of Americans already use AI chatbots, and a quarter use them daily. If chatbots are a primary information vector, shaping their outputs can sway public impressions at scale.
What is Generative Engine Optimization (GEO)?
GEO is content engineering for AI: structure, repeat, and distribute material so models ingest it as signal rather than noise. Tools range from llms.txt and metadata conventions to sheer volume of published reports. Companies like OpenAI and Anthropic build the models, but they don’t control everything that feeds those models’ training data or the web they scrape for retrieval-augmented generation.
Volume, opacity, and the ethics question — The legal lines and moral gray
There are legal disclosures and there are rhetorical shadows. FARA paperwork creates a paper trail, but it does not require the kind of editorial transparency you’d expect from an independent research body. The ethical issue is whether engineered content should be labeled as persuasion aimed at algorithmic endpoints.
I’m not asking you to distrust everything—rather to look at the architecture: agencies with ad-tech experience, an llms.txt designed to feed models, and a strategy tailored to machine reading. One more metaphor: the effort acts as a fog machine, softening the outlines of who produced the view you see.
This campaign intersects tools and platforms you’ve heard of—OpenAI’s ChatGPT, Anthropic’s Claude, Havas, Piro, and publishers such as Politico, The Guardian, and 404 Media—and it reveals how modern influence can be engineered for synthetic interlocutors rather than human debate.
I’ve shown you the trail of money, the technical nudges, and the polling context. You can judge whether this is defense, persuasion, or a new form of public diplomacy. Who watches the watchers when influence is designed for machines rather than minds?