July 19, 2026 (Conversation)
The political and electoral uses of generative artificial intelligence (AI) are increasingly discussed in the public debate. According to some, they could reshape our democratic systems and our relationship to politics. What is the reality of these uses today in France? A case study in Lille during the last municipal elections.
You may have read it in the press: 1 in 4 French people say they intend to use an “artificial intelligence tool like ChatGPT, Gemini, Claude, or Le Chat during the presidential campaign.” 22% would like to use AI “to find out about the candidates’ platforms” and 10% “to help them make their voting choice.” While this survey raises methodological questions by asking respondents to imagine themselves using these tools several months in the future, it nevertheless demonstrates the interest in them and the expectations they generate in the public debate.
In line with how the political uses of social networks were presented a few years ago, the uses of generative AI in politics are alternately suspected of weakening the functioning of democracy ( foreign influences , spread of “fake news” , increased distrust of the political system) or, on the contrary, praised as new opportunities, likely to reconnect citizens to politics and increase their participation ( “popularization” of political issues, access for young people ).
Thanks to the results of a survey conducted in Lille just after the 2026 municipal elections, we attempt here to shed light on this debate.
Minority uses
Generative AI can be defined as a category of artificial intelligence that focuses on the autonomous creation of content. Due to its very recent development, research on the uses of this type of AI by citizens for political purposes remains relatively scarce.
Certainly, the general uses of generative AI in France are very widespread : in 2025, 48% of the population say they use it (85% for 18-24 year olds), including 73% for information retrieval.
However, the few studies available put these uses and their effects into perspective. For example, regarding news retrieval, the use of AI is still limited: in 2024, a study comparing usage in six industrialized countries, including France, suggests that only 5% of the population would use AI to access the latest news.
Our survey confirms the limited use of AI by citizens for political and electoral purposes. In Lille, during the 2026 municipal elections, only 14% of respondents used AI tools to learn about politics in general and 8% about the elections in particular. However, 76% of respondents do use AI tools. Thus, just because a digital practice is widespread among the population does not necessarily mean that its political application is also widespread.
Neither a political use like any other, nor a compensatory practice
Our analyses show that political activity on social networks, the number of social networks on which people are registered, the level of interest in politics in general and the following of the municipal campaign in particular have no statistically significant link with the use of AI for political purposes.
Thus, it is neither a “normalized” practice by people very interested in politics, nor a practice to compensate for low follow-up or low interest in politics.
No difference in behavior was observed based on gender or education level. Age, however, did have a significant effect: younger people were more likely than their elders to use AI to find information about politics, mirroring the trend observed for other uses of these tools . But this practice remains a minority one: even among those under 30, only 24% use it to find information about politics in general, and 12% to find information about elections.
Limited influence on electoral choices
Although few citizens have used generative AI to learn about politics and elections, it remains to be seen whether the voting behavior of those who use generative AI for political purposes differs from that of the rest of the population.
First-round Municipal Election Vote by Studied Digital Practices (%)
| Category | PS | LFI | The Ecologists | Renaissance | RN | Abstained | Not Registered |
|---|---|---|---|---|---|---|---|
| Entire sample | 18 | 16 | 12 | 8 | 7 | 16 | 15 |
| Respondents reporting political use of AI | 18 | 22 | 6 | 5 | 10 | 16 | 16 |
| Respondents reporting electoral use of AI | 16 | 32 | 3 | 5 | 14 | 15 | 11 |
Political Party Abbreviations
- PS – Socialist Party
- LFI – La France Insoumise (France Unbowed)
- The Ecologists – Green Party (Les Écologistes)
- Renaissance – Centrist party led by President Emmanuel Macron
- RN – National Rally (Rassemblement National)
Footnote
Fieldwork: Sample of respondents (n = 800); weighted data. Votes for other lists are not shown because their sample sizes were too small to be statistically reliable.
Our survey shows that voters who used AI were more likely to vote for LFI, or even RN. Thus, while LFI voters represent 16% of our sample, they represent 22% of respondents reporting political use of AI and 32% of those reporting electoral use.
Similarly, while RN voters represent 7% of the sample, they represent 10% of those declaring a political use of AI and 14% of those declaring an electoral use of AI.
Respondents reporting electoral use of AI also appear to be slightly less unregistered than average (11% versus 15%), but no less likely to abstain.
However, on the one hand these differences are small in view of the size of the sample and, on the other hand, they could be explained much more by the much younger age of LFI voters , or even RN voters .
Some key points to remember ahead of the 2027 presidential election
The results presented above therefore suggest a need to significantly qualify media narratives announcing a profound and rapid transformation of political behavior under the influence of AI. Above all, they highlight that this technology is still in a phase of adoption, with effects that remain uncertain and largely dependent on the contexts of use.
Regarding user profiles, they indicate that the political and electoral uses of AI do not correspond to clearly identifiable social or political segments: they are neither the domain of highly politicized individuals nor those with little interest in politics. This suggests that these uses stem more from an exploratory logic on the part of citizens than from structured appropriation.
Furthermore, this allows us to qualify the idea of a structuring of political uses around an “AI divide” which would take up the contours of the now well-known “digital divide” , which would particularly weaken older people with a lower than average level of education.
While an age effect is clearly observable, other sociodemographic and political variables play a limited role, suggesting that these uses are still not yet socially established. Finally, regarding the political effects of AI use, our analyses show that these practices are relatively weakly linked to electoral choices.
Overall, our research highlights a significant gap between, on the one hand, public discourse on the role of AI, and on the other hand, the empirical reality of citizen use. This “new” digital use appears largely rooted in a continuity of political practices rather than a transformation. Generative AI, while opening up significant possibilities in political communication and persuasion, does not yet seem to constitute a structuring factor in ordinary electoral behavior.
