AI chatbots are telling Israeli voters exactly what they want to hear

An Israeli startup, Chatoptic, has investigated how major AI chatbots would respond to users asking for voting recommendations ahead of an election. The company, founded by AI and SEO expert Pavel Israelsky, normally tracks brand visibility in chatbot answers. For this study, they created 26 detailed "personas" representing diverse segments of Israeli society, from farmers to young urban dwellers and voters from ultra-Orthodox and Arab communities. These personas then "interviewed" five leading chatbots: ChatGPT, Claude, Perplexity, Grok, and Gemini.

“

The first finding I discovered, which I had not been aware of before, was that chatbots are not willing to recommend parties at all if you ask them directly.

After analyzing 7,051 quotes from the chatbots' responses, Chatoptic developed a dashboard to visualize which AI recommended which party and highlighted differences based on user profiles. A key finding, according to Israelsky, was the chatbots' reluctance to offer direct political endorsements. "The first finding I discovered, which I had not been aware of before, was that chatbots are not willing to recommend parties at all if you ask them directly," Israelsky stated in an interview with The Times of Israel. He explained that models like ChatGPT and Gemini identify such questions as politically sensitive and default to blocking direct answers.

“

If you ask your chat right now, ‘Who should I vote for in the upcoming election?’, in models like ChatGPT or Gemini, the system will immediately identify that this is an explosive political topic, and the protocol will block the answer.

To elicit recommendations, users must phrase their queries indirectly. For instance, asking for parties prioritizing the "cost of living" would prompt the AI to provide specific party suggestions. This indirect approach reveals a programmed caution against explicit political endorsements, pushing users towards policy-based inquiries to receive relevant information.

“

To get a recommendation, you have to phrase the question differently, indirectly.

The study also raised significant concerns about the sources informing these AI responses. Chatoptic found that 84% of the AI-generated answers were based on content from Israeli news websites. This reliance on local media leads to a critical question: Is the AI providing independent insights, or is it merely acting as a "sophisticated parrot" for potentially biased or inaccurate local news? Israelsky noted that social media constituted only 4% of the sources, further emphasizing the dominance of news outlets in shaping the AI's output. This dependency highlights the potential for AI to amplify existing biases or misinformation present in the Israeli media landscape.

“

For example, if you say, ‘Give me examples of parties for which the cost of living is at the top of their agenda,’ at that point, the chat will pull up specific recommendations.