Politics is not immune to its growing use. Mobilized by the parties to produce visuals, leaflets and strategic plans, in daily support of parliamentary work, AI is now involved in voters’ decisions. At least some of them: 16% of French people used an artificial intelligence tool to guide their voting choice during the 2026 municipal elections, according to a Toluna Harris Interactive* study, the conclusions of which the Terra Nova Think Tank is publishing this Monday.
“One in six French people consulted a generative AI and said they were going to ask it questions about their electoral behavior, it’s an important phenomenon,” Jean-Daniel Lévy, author of the study and director, assures Public Sénat delegate of Toluna France. Enough to make last March’s municipal elections Act 1 of the electoral use of AI in France: “It is no longer reserved for a few isolated technophiles but is starting to become part of the ordinary political practices of an identifiable part of the electorate,” points out the pollster.
5% of French people changed their vote after consulting an AI
The survey, carried out from a poll on voting day of more than 4,000 voters spread across municipalities with at least 3,500 inhabitants, distinguishes several types of users. The figure of 16% thus adds up the French who were reassured by AI in their choice (7%), those who changed their mind upon contact with it (5%) and those who, not knowing who to vote for, used it to adopt a position (4%).
Confirmation, influence or decision support tool: AI offers various possibilities. Unsurprisingly, it was the public most accustomed to its use who took it up for these municipal elections. 20% of men have used it (compared to 10% of women) and 35% of those under 25 (compared to only 1% of those aged 75 and over).
If almost half of its followers therefore used it as a simple guarantee of control of their choice, the AI is worrying by its capacity to modify voting positions. Almost a third of users experienced it during the municipal elections. “AIs are black boxes and raise the question of how political opinions are formed,” warns Jean-Daniel Lévy, recalling however that this is still a very minority phenomenon.
“I don’t know what ChatGPT is going to recommend to vote on.”
Last November, Emmanuel Macron expressed concern about their growing influence in front of an audience of readers of La Voix du Nord. “In the next municipal and presidential elections, I will tell you what will happen: more and more of our compatriots will go to their AI agent and they will say “who should I vote for?” HAS”. There, we’re going to enter another world, because I don’t know what ChatGPT is going to recommend to vote on. HAS”
The impact of a prolonged dialogue with generative AI has already been highlighted by several researchers from the Massachusetts Institute of Technology (MIT) for the 2024 American presidential election. In their study published on December 4 in Nature, the authors observed a strengthening of the partisan preference of voters faced with a model designed to defend their favorite (Donald Trump or Kamal Harris) and, conversely, an erosion of their support after contact with an opposing model to their candidate.
CSPs are getting more information via AI
While awaiting such a conclusion in the French case, part of the electorate also used AI as a source of information on the municipal campaign. They are 11% in this case, which makes it only the fourteenth information channel, according to the Toluna Harris Interactive study, far behind leaflets (59%), word of mouth (47%), the regional daily press (36%) or even social networks (32%). Enough to qualify, here again, its importance.
The informational use of AI also reveals an over-representation of men (14%, compared to 8% of women) and younger people. But, even for the latter, AI remains last in the hierarchy of information sources used during the campaign. Less old than the average, supporters of La France insoumise are the most numerous to have adopted it (16%).
More surprising, this use “affects categories of population which we could have considered, a priori, that they are not part of the main users”, analyzes Jean-Daniel Lévy. CSPs, i.e. workers and employees, obtained more information via AI (17%) than managers and higher intellectual professions (13%). Voters who “feel less close to political information and whose opinion is sometimes less constituted,” specifies the pollster.
« Un outil de référencement important »
The appearance of AI in the political conversation was predictable. An Ipsos-bva study published in January, before the municipal election, revealed that almost one in two French people (48%) had already used or were considering using a generative AI tool to find out about politics during the municipal or presidential elections. This figure rose to 75% among 18-24 year olds. In addition to information on programmatic measures or the positions of candidates, 30% of respondents imagined using it to identify the party most in line with their ideas.
“AI has established itself as an important SEO tool. As with the emergence of social networks, political parties are a little late but they are seriously starting to get into it,” explains Jean-Daniel Lévy. Except that, unlike social networks whose imperatives can be partly decoded to appear prominently, the responses provided by conversational assistants remain difficult to understand, and therefore to influence. A major challenge for the 2027 campaign.
