In Israel’s AI election, who decides what is true?

Israel is heading into an election unlike any it has held before. Foremost is the question of false information. This is not new. Neither have politicians suddenly discovered how to manipulate voters. But artificial intelligence is beginning to change both the information we are being shown and the way we decide whether to believe it. A new survey conducted by the Geocartography Group for the Israel Internet Association (ISOC-IL), an independent non-profit organization that conducts research and policy work on Israel’s online environment, found that 71 percent of Israeli internet users surveyed said they frequently or occasionally encounter false election-related information. Some 56.8% reported encountering fake or impersonating accounts, while 48.2% said they frequently or occasionally encounter AI being used to harm or defame someone. The survey was conducted in July among 1005 Israeli internet users aged 18 and over, using a representative sample of the country’s Jewish and Arab populations. But the statistic that interests me most isn’t how many Israelis say they have encountered false information. It is how many of us may now turn to technology to help decide what to believe. In a separate survey published earlier this year, the ISOC-IL found that one in four Israelis was considering using AI tools or another online service to help inform a voting decision. The Times of Israel has already explored what that might mean. In June, it reported on an experiment in which 26 simulated Israeli voters questioned five leading AI chatbots about their political choices. The researchers examined ChatGPT, Claude, Perplexity, Grok and Gemini and found significant differences in the information and recommendations they produced depending on the model, the voter persona and the way the question was framed. But there is a bigger question here than whether ChatGPT, Gemini or another chatbot might nudge an Israeli voter towards one party rather than another. What happens when we increasingly rely on AI to help us make sense of an information environment that is itself contested, confused and sometimes deliberately manipulated? I have spent much of my career reporting wars, and perhaps that is why this question bothers me so much. I remember reporting from a UN school in Gaza after Israeli fire had struck the area. Palestinians at the scene told me civilians had been deliberately targeted, while the Israeli account was that militants had been operating nearby. Standing there as a journalist, I couldn’t independently establish which account was correct. You still have to report, but you also have to be honest about what you know, what somebody else is claiming and what you simply cannot establish. That experience makes me wonder whether we are concentrating too much on the most spectacular AI threat. We are fascinated by deepfakes: the politician who never said those words, the explosion that never happened, the photograph manufactured by a machine but convincing enough to look real. These are real threats, particularly during an election, but research into the Israel-Iran war suggests that the harder problem can be much less spectacular. The ISOC-IL analyzed hundreds of fact-checks from the June 2025 conflict and found that 88% of the war-related fact-checks in its analysis contained at least one decontextualised feature. 71% concerned false connections and decontextualised content, where genuine material was presented with contextual information that changed its meaning. By comparison, 20% of the content examined involved generative AI. The categories overlap, but the broader finding is striking: much of what fact-checkers were dealing with was rooted not in an entirely fabricated reality, but in genuine material being used misleadingly. That distinction matters. False or misleading information is not automatically disinformation. Disinformation involves deliberate deception and, unless intent can be established, we shouldn’t assume that everyone sharing something inaccurate is deliberately trying to deceive us. People misunderstand, mis-caption and repeat things they genuinely believe all the time. But for the person trying to work out what actually happened, both can contaminate the information environment. A photograph can be genuine but from the wrong war, a quotation can be authentic but stripped of the sentence that changes its meaning, while a statistic can be completely accurate and still be used to construct a profoundly misleading argument. Now AI enters that already complicated information environment, not simply as another technology capable of manufacturing content, but increasingly as the tool we ask to explain the content to us. That, to me, is where the really interesting problem begins. Until now, much of the discussion about AI and elections has focused on what AI can do to voters: deepfakes, synthetic propaganda, automated accounts and political content produced at extraordinary speed and scale. But if voters increasingly use AI to make sense of competing political claims, those trying to influence them potentially have another route. They don’t necessarily have to influence the voter directly. They may eventually try to influence the information environment the AI encounters before answering the voter’s question. To be clear, the Israeli research does not show that this is happening, nor does it demonstrate that an influence operation has successfully changed an AI system’s voting advice. What it does show is why the question is becoming increasingly important. An AI answer doesn’t appear from nowhere. Depending on the system and the question, it may draw on information produced by governments, journalists, political parties, academics, activists, businesses and ordinary people. Some systems can search current material online, select sources and synthesise competing accounts for us. And that information environment is not pristine. Influence operations can create professional-looking websites, apparently credible personas and bodies of material designed to reinforce a particular narrative. So the old advice to “check the source” remains important, but it becomes more complicated when the source itself can be made to look authoritative, or when several apparently independent sources ultimately originate from the same information operation. If an AI system retrieves and synthesises information from that environment, the challenge is no longer simply whether the machine can recognise a fake photograph. It is whether it can make sense of competing versions of reality when much of the material looks perfectly legitimate. Which, when you think about it, isn’t so different from the problem human beings already face. Israel has already introduced new rules requiring election campaigns to label AI-generated or AI-manipulated content that could be mistaken for authentic material. That is an important attempt to protect voters as synthetic images, audio and video become more convincing. But a label saying something was AI-generated or manipulated solves only one part of the problem. It tells us something about how content was made, but it cannot tell us whether the argument attached to it is true, whether an authentic photograph has been placed in a false context or whether an apparently independent source deserves our trust. I learnt that lesson long before artificial intelligence entered the conversation. Standing in a conflict zone with two completely different accounts of what had just happened, there wasn’t always a technological tool that could tell me which one was true. Sometimes the most responsible conclusion was simply: I don’t know yet. That willingness to live with uncertainty may become more important in the AI age, not less. AI can help us search, compare competing claims, find information we might otherwise miss and expose us to arguments we haven’t considered. Used properly, it could make us better-informed voters, but there is a difference between using AI to help us think and asking it to do our thinking for us. AI can help us decide what information to consider, but the decision about what deserves our trust cannot simply be outsourced to a machine. The question facing us in Israel’s AI election isn’t simply whether AI can tell truth from fiction. It’s whether we still can.