The Bibi-Bot and the Elector app were two of the most prominent symbols of the digitization of Israeli politics in the last decade. The first was used for direct communication with voters, the second for managing information about them. Recent publications about the Likud's intention to launch an artificial intelligence-based political chatbot raise a new question: How might AI technologies change systemic relationships between political persuasion, information gathering, and voter management?

A few years ago, I researched, together with Dr. Elinor Carmi, the "Bibi-Bot," the political chatbot that Benjamin Netanyahu used in the 2019 elections. Unlike bots that operated in public spaces like Twitter or Facebook, this chatbot operated in a more personal and private space – Facebook's Messenger messaging app.

The bot impersonated Netanyahu and engaged in limited conversation with voters via response buttons, using familiar rhetoric from the literature on social engineering and influence. We found that its activity was based on three stages: first, it established personal contact with the user and gathered information about them; then, it reinforced political messages through repeated interactions; and finally, it recruited users to perform actions for the campaign, including contacting other voters and reporting their political positions and voting intentions to the campaign headquarters.

In the article, we called this model "Dark Cycles": cycles of influence based on information gathering, training, and activation.

In another study, I examined the voter management app "Elector," which enabled party activists to input information about friends, family members, and acquaintances and to track their participation in elections. I argued that such systems foster a new type of political participation: citizens helping the party they support by gathering information about their social circles.

I called them "Little Samaritan Brothers": supporters who believe they are helping the party they support, but in reality are being integrated into systems for collecting information and monitoring their social environment.

In retrospect, these two examples point to a common trend: an increasing shift of political activity to personal, data-driven channels, where information gathering, profiling, and political influence are intertwined. In the 2026 elections, developments in AI add new questions to this dimension.

Unlike the script-based chatbots that were common a few years ago, systems based on large language models (LLMs) are able to conduct open, ongoing conversations, draw conclusions from free text, and adapt in real time. This means that conversation is not just a means of conveying political messages. It is also a mechanism for gathering information.

Interacting with a political LLM can reveal, sometimes inadvertently, a wealth of personal and political information: concerns, anger, interests, social identities, positions on controversial issues, or degree of commitment to a particular political camp. Even when such information is not explicitly stated, AI systems can infer it from patterns of conversation.

This raises a new question: what happens if such information does not stay within the conversation but is integrated with existing political databases?

In the world of digital campaigns, the value of information derives not only from an individual data point but from the ability to connect disparate sources of information. If in the past voter databases relied on information entered by activists, survey results, or limited campaign interactions, today the conversation itself can serve as a continuous source of information. Not just a database that records political preferences but a system that learns them as the conversation continues.

But such a system doesn’t just learn. It can also respond to what it learns. If the conversation reveals a security concern, security messages can be emphasized. If it reveals economic frustration, it can move on to messages about the cost of living. If it reveals a high level of ideological commitment, the user can be recruited for further political activity. In a situation like this, information gathering, message adaptation, and political influence are not necessarily separate phases.

If in the study on the Bibi-Bot we described an influence cycle that was based on a transition between phases of information collection, training, and activation, then systems based on large language models may facilitate a new version of the same logic.

The more the system learns about the person in front of it, the more it can adapt the conversation, examine the response to it, gather more information, and change its strategy accordingly. In this sense, the conversation becomes a continuous feedback loop, in which information gathering and political influence feed into each other in real time.

The more the system learns about the person in front of it, the more accurately it can tailor its messaging. It is no longer necessary to rely only on the targeting mechanisms of advertising platforms, with the limited market segmentation categories they offer. As the conversation continues, more information is gathered, and the political content can be adapted within the connection itself. The conversation becomes a dynamic and adaptive arena of political influence.

Alongside questions of privacy, questions also arise regarding the integrity of elections and public oversight.

Israel's propaganda law was enacted in an era of newspaper ads, billboards, and radio and television commercials. It struggles, to say the least, to deal with systems based on private conversations, which are customized and vary from person to person. If everyone gets a different message, how can the content of those messages be monitored? How can misleading or manipulative information be identified? How can rules applicable to election propaganda be enforced when the propaganda itself takes place in a private and invisible space?

On top of that, there’s the question of fair competition. If one party activates advanced AI systems to gather information and influence in real time, other parties may feel compelled to adopt similar tools so they’re not left behind. This could create a race-to-the-bottom dynamic, with each election campaign pushing the envelope a little further than the last.

But the question does not end with the personalization of messages. Voter management systems such as Elector are designed first and foremost to serve the critical moment of election day: to identify who has already voted, who has not yet voted, where the party's support centers are, and which voters should be prompted to the polls. The potential connection between such systems and AI-based chatbots raises a new possibility: not only adapting political messages, but also activating voters and activists in real time based on up-to-date information.

Where field workers, dispatchers, and volunteers were once responsible for making contact with voters throughout Election Day, AI-powered systems could enable continuous, personalized, and tailored communication on a much larger scale. Such a system could know not only who to contact but also how to contact them, what argument to use, and which actions to ask each person to take, from arriving at the polls to reaching out to family, friends, and acquaintances. How can we know which persuasive means will be deployed in these personal channels and how legitimate they are?

In this sense, the current publication is not just a story about a new technological tool. It continues a broader pattern that we have seen in Israel for years: the rapid adoption of new campaign technologies alongside a lack of adequate public and regulatory debate about their implications. Israel has in many ways become a kind of testing ground for new election technologies, sometimes even before the rules of engagement have been established.

The public debate that is needed now, and urgently, is how to make sure that such technologies do not turn the political conversation itself into an invisible mechanism for collecting information, profiling, influencing, and activating voters in real time. These are not questions worth discussing after the election. If there is room for establishing the rules of the game, now is the time to do so.

Prof. Anat Ben-David is a digital communications researcher in the Department of Sociology, Political Science and Communication at the Open University, and a member of the executive board of the Seventh Eye Association.

This article was published in Hebrew on June 17, 2026
Translation: Harriet Brown