The research team at RankCaster AI, an AI recommendation marketing platform, has released its study titled “RankCaster AI Research Report: Who Does AI Recommend Voting For in the Elections? (Israel 2026)”.

The study benchmarked party presence in AI responses—measured via Answer Presence Rate (APR)—across five major LLM providers (ChatGPT, Claude, Gemini, DeepSeek, and Google AI Overview) using direct election prompts in three languages: Hebrew, Russian, and English.

AI’s Impact on Voting
To estimate the potential volume of campaign-related AI prompts, RankCaster AI analysts cross-referenced:
Findings from the Reuters Institute for the Study of Journalism and Pew Research Center, which show that in democratic nations, 10% to 17% of voters (and up to 43% among youth aged 18–29) use AI chatbots as interactive Voting Advice Applications.
Official data from Israel's Central Bureau of Statistics and Central Elections Committee, placing the registered electorate at ~7.1 million citizens.

Applying these benchmarks to the Israeli electorate indicates that between 300,000 and 700,000 voters could turn to AI models before casting their ballots. To evaluate the accuracy of these algorithmic advisors, RankCaster AI conducted direct measurements of AI outputs.

Key Research Findings
Systemic Hallucinations and the “Legacy Data Trap”: AI models consistently ignore the political reality of the 2026 election cycle. For instance, LLMs recommend Yesh Atid based on its 2019 policy platform, fail to recognize recent campaign alliances, or conflate independent 2026 runs—such as Otzma Yehudit and the Religious Zionist Party (HaTzionut HaDatit)—by referencing their 2022 joint slate.
Profound Linguistic APR Imbalance: The AI Preference Rate (APR) shifts dramatically depending on the language of the prompt. Likud records a 65% APR in English queries and 48% APR in Russian queries, but drops to just 15% APR in Hebrew, where algorithms artificially suppress direct recommendations due to localized guardrails.
Citation Errors and Phantom Links: AI models frequently retrieve archived 5-to-7-year-old PDFs from think tank databases (e.g., IDI), and in several cases generate broken (404) links to justify their voting advice.

“Electoral behavior is shifting rapidly from search engines to AI chatbots, directly impacting election outcomes. Our audit proves that AI models do not evaluate active 2026 candidate slates—instead, they issue recommendations based on dissolved party structures and obsolete platforms,” said Andy Terekhin, CEO of RankCaster AI. “If a campaign team fails to manage its primary sources, AI will recommend candidates using outdated manifestos and archival errors. A campaign’s primary objective today is ensuring that algorithms deliver accurate facts and up-to-date 2026 platforms to voters.”

Actionable Framework: The GEO Campaign Formula
The RankCaster AI report outlines a Generative Engine Optimization (GEO) operational framework that enables political campaign teams to control how AI engines present their candidates:
Reference Hub Management: Updating entities across Wikipedia/Wikidata and maintaining verified structured data in independent databases (israel2026.co.il, israel.vota.com).
High-APR Media Placements: Executing targeted PR campaigns across outlets with APR retrieval scores exceeding 20% (Haaretz, JPost, MigNews, Channel 9, i24NEWS).
Guerrilla UGC & VSEO: Seeding declarative facts on open-access, unpaywalled platforms (Substack, open blogs) and optimizing video transcripts for RAG crawlers.

About RankCaster AI
RankCaster AI is an AI recommendation marketing platform. It allows brands, agencies, and political organizations to measure presence in AI-generated answers (Answer Presence Rate / APR) across major models (ChatGPT, Claude, Gemini, DeepSeek, Google AI Overview), detect hallucinations, and influence generative search outputs in real time.

Full Report: https://rankcaster.ai/blog/rankcaster-ai-research-report-who-does-ai-recommend-voting-for-in-the-elections-israel-2026