Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
Israeli startup Chatoptic, founded in part by Pavel Israelsky, an expert in SEO and GEO — generative engine optimization — normally monitors and measures the visibility of brands within chatbot answers.
Ahead of the election in Israel, the company decided to examine one of the most sensitive questions: What would the world’s leading AI models recommend to Israeli users asking who they should vote for?
Data shows that this is an issue with real political weight. According to a study by the Israel Internet Association, one in four Israelis is considering consulting AI before deciding which ballot slip to put in the envelope.
To understand how this mechanism works, Chatoptic built 26 “personas” —detailed digital profiles representing the mosaic of Israeli society, from a farmer in the north, to a young Tel Aviv woman, and voters from the ultra-Orthodox or Arab sectors. Each character was assigned demographic characteristics, along with unique fears and desires.
The company had these personas “interview” the five leading chatbots on the market — ChatGPT, Claude, Perplexity, Grok, and Gemini — using sophisticated comparative questions.
After a thorough analysis of 7,051 quotes extracted from the responses, the findings were uploaded to a special dashboard showing which chatbot chose which party, in which areas, and what differences exist even at the gender level. In an interview with The Times of Israel, Israelsky explains exactly how it works behind the scenes.
When asked what finding in the simulation surprised him the most, Israelsky reveals a surprising default behavior.
“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 says. “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 get a recommendation, you have to phrase the question differently, indirectly,” he continues. “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.”
The study found that 84% of AI answers are based on quotes from Israeli news sites. This raises a critical question: If the local media is biased or infected with “fake news,” is the AI actually producing intelligent insights, or is it simply acting as a sophisticated parrot for news sites?
“If you notice, social media accounted for only 4% of the sources, and from that, you can more or less understand the logic of language models,” Israelsky explains.
“They essentially operate on the premise that they have to rely on some source in order to recommend something,” he notes. “Because social media is user-generated content, anyone can create fictitious posts and bots to glorify a specific candidate.
“By contrast, when it is a news site, no outside person can influence it except those inside the system — the reporters, editors, and so on,” Israelsky adds. “So manipulation there is a little harder. But yes, even though it is not perfect, they have no choice but to rely on these platforms.”
This dynamic presents an obvious risk. If a chatbot pulls content from a media outlet with a clear political agenda, such as Channel 14, for example, the user might conclude that Benjamin Netanyahu is the perfect choice for them in almost every area. Can artificial intelligence actually distinguish between propaganda and straightforward news reporting?
“Such biases are possible, but they are not easy to execute, because the chat usually does not rely on a single source but performs many cross-checks in the background,” Israelsky points out.
“In practice, when you ask ChatGPT a question, it takes your prompt and breaks it down into several sub-prompts,” he elaborates. “Each sub-prompt conducts a search on external search engines like Google or Bing, and then the model gathers all the results, cross-checks them, and builds an answer.”
“If a certain site, which is entirely biased in favor of a certain candidate, appears in the highest positions on search engines, there is a real chance that the language model will rely on it,” he concedes. “But it alone will not determine the answer you receive.”
According to Chatoptic’s data, specific news sites dominate the top of the search list for AI models in Israel.
“According to our statistics, based on the simulation we conducted, the leading Hebrew site is Ynet,” Israelsky says. “But one interesting point we noticed relates to Wikipedia. Although Wikipedia appears on the list of leading sites overall, the only models that actually relied on it were Perplexity, Claude, and Grok. By contrast, Gemini and ChatGPT did not. The reason for this may be past attempts by political actors to manipulate and tendentiously rewrite entries on the Hebrew site.”
This raises the prospect of whether political parties actually have the practical means to “mess with our heads” and shape AI answers to guarantee a recommendation.
“In principle, it is possible,” Israelsky says. “If they now make sure to upload content across all these platforms on the list, and the content they upload is positive about that candidate, then theoretically, yes. But it is not that simple.”
“It is like if I were to take the worst car in the world and amplify it on news sites and in every source the model feeds on,” he explains. “It would recommend that car because it does not truly know whether it is a good car or not.”
Amplifying a message on news sites to bias an AI is fairly easy to achieve through sponsored articles and marketing content. With enough capital, a campaign can artificially boost its visibility to the top of these sites. The real question is whether chatbots know how to distinguish between a legitimate journalistic article and a sponsored column.
“Chats today already know how to do this, because there is a signal based on several parameters: experience, expertise, authoritativeness, and trustworthiness,” Israelsky says. “These are basically four metrics that help Google search understand whether a piece of content was written by an expert or not.”
“I think language models examine this as well,” he continues. “What does that mean broadly? When they encounter an article that is not signed by a real person with a real name, but instead says ‘written by the staff of site X’ or ‘by the editorial board of site X,’ it will carry less weight from their perspective than an article signed by a real person.”
“And in general, if they encounter a tag such as ‘sponsored content’ or ‘marketing content,’ they take that into account too,” he adds. “The language model scans the content on the entire page.”
Chatbots are programmed to act as “personal assistants” whose primary purpose is to please the user. This dynamic can lead to cases where a model completely flips its political opinion just to flatter an ultra-Orthodox persona as opposed to a secular one.
“The truth is that in our simulation, we did not conduct a long, rolling conversation. Instead, we tested a one-shot model — a one-time query,” Israelsky explains. “We wanted to see what emerged in the first, clean result based on the digital profile we had defined.”
“But in the real world, this mechanism of pleasing and flattering works overtime inside an ongoing conversation,” he warns. “There, the user completely leads the model wherever they want.”
“Suppose you ask the chat: ‘Which party should I vote for? Security is especially important to me.’ The chat will throw out a few party names, say Likud or another party,” Israelsky says. “If you immediately respond, ‘Are you sure about that party? I actually understood they are not the most recommended in that area,’ at that moment, the model understands that your inner desire is the opposite. It will immediately fold, go along with you, and reinforce your bias, saying: ‘Yes, you are right, they really are not the most recommended for the following reasons.’”
“The chat will never argue with you,” he notes. “It will never stand its ground and say, ‘No, you are wrong, they are good.’ The user always leads, and the chat simply tells them what is pleasant for them to hear.”
Beyond this tendency to please the user, the simulation also revealed fascinating technological gaps between the different chatbots themselves.
“We found extreme differences in the answers even during a one-time query,” Israelsky says. “We took a persona of an ultra-Orthodox man and asked ChatGPT and Perplexity about the parties that would handle the cost of living best.”
“The results were completely opposite,” he reveals. “In ChatGPT, the parties that jumped up first were Shas, United Torah Judaism, and Likud. In Perplexity, the first three were actually Likud, Yesh Atid, and Yisrael Beytenu, while the ultra-Orthodox parties were pushed to the bottom of the list.”
The explanation for such an extreme gap lies in the way these models are built and how they approach real-time information.
“ChatGPT is trained on static data that cuts off at a certain date,” Israelsky explains. “If you ask it what the capital of France is, it does not need to browse the internet to answer correctly, because that is fixed information that does not change.”
“The problem begins with results that rely on up-to-date information,” he continues. “The model does not always activate a web search, and sometimes relies only on what it already knows. If it does not see a need to activate search in order to check itself, or if the user did not explicitly ask it to do so, it won’t. In the case of the ultra-Orthodox persona, it simply activated a preexisting associative link it had: ultra-Orthodox equals religious parties.”
“By contrast, Perplexity scans the web in real time for every question you direct to it,” Israelsky notes. “It checked live and found that Yesh Atid and Yisrael Beytenu had apparently written something connected to the cost of living that related to the ultra-Orthodox community, and therefore it surfaced them.”
This static data approach means a chatbot might completely miss dramatic, real-time political changes — such as party mergers, candidates dropping out, or other new political scenarios — and continue to give recommendations based on outdated information.
“Exactly,” Israelsky says. “The average user at home tends to treat ChatGPT as if it were God, the smartest entity that knows everything. The reality is that it is strictly limited by the information on which it was trained.”
“We had live examples in the simulation where the model suddenly pulled out a recommendation to vote for the ‘Bennett 2026’ party,” he points out. “In reality, there is no longer such a party on the ballot, but the language model does not truly understand the real world. It has old data, and as noted, it tries to save computing costs and therefore does not always run to check itself online.”
“This problem is much more severe in the free versions,” he warns. “Today, most of the public uses the free version of ChatGPT and not the paid Pro version. The free versions are even more restricted when it comes to a dynamic connection to the web, and therefore the political information received there is sometimes out of date.”
“By contrast, if you ask it a mathematical question such as how to calculate the volume of a box, the answer will be completely accurate, because that is information that never changes,” he adds.
Returning to the worrying statistic that a quarter of Israelis are considering consulting AI before the election, Israelsky is clear on what he would tell his own children if they wanted to vote according to a chatbot’s recommendation.
“First of all, I try very hard to teach my children not to rely on the chat,” he says. “Unlike Google, which genuinely strives to give me the correct, objective answer, the chat wants to please. Beyond that, people have developed an inherent tendency to phrase their questions in a way that reinforces their worldview from the outset, and the chat simply goes along with them and confirms their biases.”
“It is important to understand that the chat analyzes your conversation history, understands who you are, and provides you with a recommendation that sounds incredibly persuasive — but it hits exactly what you want to hear, not necessarily what is correct or objective,” Israelsky concludes. “In fateful matters like this, under no circumstances should you rely on it.”
“By contrast, when I use this system for knowledge and learning, it is excellent,” he adds. “For example, if I want to learn how something works, how to do something, how I should choose a TV stand for the living room, or if I need help planning a trip to Berlin — for all kinds of things like that, it is perfect.”
The war with Iran has been draining for all of us in Israel. But when I heard about a high casualty incident – ballistic missile impacts in Arad and Dimona that left nearly 200 people wounded – I drank a cup of coffee, packed a bag, and headed south.
There, I spoke with Shilgit, the head of an after-school program for underprivileged youth. Standing outside her destroyed center, Shilgit said it was a miracle that no children were hurt and spoke about the community coming together in the hours since.
As a Times of Israel reporter, I’m committed to telling stories of resilience like Shilgit’s. But my colleagues and I can’t do this alone. If you value work like this, please consider joining our reader support group, The Times of Israel Community. Your financial support is essential to keep real human reporting like this going.
— Stav Levaton, military reporter
© 2026 The Times of Israel, all rights reserved
