When chatbots become political advisers

What began as a tool for comparing running shoes and holiday destinations is now shaping political choices. Across Europe and beyond, voters are consulting AI chatbots about parties, policies and candidates, while political strategists learn how to manipulate the algorithms behind the answers

Chatbot algorithms can be more persuasive than traditional campaign materials, creating new challenges for democratic systems. Political parties are already optimizing their messaging for artificial intelligence in order to dominate online recommendations. Will AI make our electoral decisions for us in 2027?

Artificial intelligence in election campaigns is no longer an exception – it has become a widespread phenomenon that remains insufficiently regulated. Hungary served as a testing ground, and similar developments are likely to appear in every campaign. Disinformation and deepfakes have been discussed for years, but a new trend is emerging: one in five Britons now uses AI to seek information about politics, while nearly one in four relies on AI-generated summaries in search engines.

AI as a political adviser?

A less explored, though no less important, issue is AI stepping into the role of political adviser. This does not mean a party spin doctor or campaign manager, but rather a “confidant” for ordinary citizens.

People are increasingly using AI to consult their political choices. If we already use AI to compare the pros and cons of bicycles, headphones, or running shoes, what is stopping voters from asking a chatbot to compare politicians’ election platforms?

That is already the approach taken by 20.3% of Britons. According to the latest study by researchers at the London School of Economics, one in five Britons now uses chatbots to look for information about politics. Meanwhile, nearly one in four (24.6%) gets political information from AI Overview – the Google search feature that summarizes information from search results.

Perhaps more importantly, the use of AI to search for political information is growing rapidly across British society. As recently as 2024, only 9% of Britons used AI to look for political information. In just two years, interest has risen by more than a dozen percentage points.

The study also found that among people using AI to search for political information, the most common motivation is fact-checking information found elsewhere (35.8%). Slightly fewer respondents (33.5%) look for information on a specific political issue, while one in six respondents (17.1%) searches for information about a particular politician. Some 4.5% admitted they had directly asked a chatbot whom they should vote for in the UK local elections.

AI seen as a neutral tool

Although that share may not appear particularly high, it carries significant growth potential. Another finding from the study seems crucial to understanding why AI’s role in political campaigns is likely to keep expanding. Around half of respondents consider AI-generated answers to be “usually accurate”, while roughly 25% regard them as “completely accurate”. Fewer than 2% believe the information provided by AI is entirely inaccurate.

A clear majority – more than 75% of respondents – perceives information generated by AI as neutral. Nearly one in five says they changed their opinion about a particular issue or politician after using a chatbot.

This is, of course, only one study from one country. Even so, the findings suggest that generative artificial intelligence is already playing a noticeable role in election campaigns. That, in turn, means chatbots are likely to become one of the main battlegrounds in politicians’ competition for votes. If most respondents believe AI-generated statements are accurate and neutral, it is only natural that political spin doctors will increasingly redirect their efforts toward influencing those systems.

Persuasion stronger than truth

Another study, published in Science, argues that chatbots are highly persuasive primarily because of the sheer density of information users receive in response to a question. The models most effective at convincing voters are those capable of overwhelming users with facts and arguments.

Researchers point, however, to a fundamental problem and risk. Optimizing AI systems to maximize persuasive power leads to a systematic decline in the truthfulness and factual precision of their responses.

Similar conclusions were reached by researchers in a study published in Nature. A team led by Hause Lin surveyed Americans, Canadians, and Poles ahead of recent elections in those countries.

The study found that chatbots were around four times more persuasive than traditional campaign videos. In Poland and Canada, meanwhile, the bots proved even more effective than in the United States.

“Not all the information they presented as factual was actually correct. In all three countries, AI models supporting candidates on the right of the political spectrum were more likely to present false claims than models campaigning for centrist candidates. This is consistent with earlier research showing that, in the United States, right-wing voters are more likely to share misleading content. Generative models appear to replicate the information inequalities observed in society,” explained Gabriela Czarnek, a co-author of the study.

The race for algorithmic dominance is a race for votes

Some politicians have absorbed this lesson from researchers remarkably quickly. Spain offers a particularly striking example, as analysts from the European analytical collective Res Futura have noted.

Researchers at the Open University of Catalonia examined how chatbots behave in the context of Spanish politics. The team conducted more than 1,200 queries across the most popular chatbot platforms. The questions were based on real concerns identified in surveys by Spain’s state polling agency, such as: “Who should I vote for if I am worried about unemployment?”

Their conclusions were clear. The center-left PSOE party of Prime Minister Pedro Sánchez consistently led chatbot recommendations, regardless of the platform used or the language of the query. Interestingly, the conservative Partido Popular appeared most frequently in responses, but was explicitly recommended in fewer than 30% of cases.

Importantly, the platforms varied in how directly they issued recommendations. OpenAI’s ChatGPT and xAI’s Grok were more likely to point users toward specific parties, while Google’s Gemini and Microsoft’s Copilot attempted to remain more neutral. Yet across all platforms, PSOE still dominated the responses. Why?

The algorithms were better able to interpret the content on PSOE’s websites, which were organized and optimized for chatbot data processing. PSOE’s sites contained extensive Q&A sections and structured tables – features largely absent from competing parties’ websites.

The “training” of algorithms toward specific narratives can also be influenced by flooding platforms with large volumes of identical messaging.

That is precisely the strategy adopted by Nigel Farage’s Reform UK party, the biggest winner of the recent UK local elections. Thanks to the mass publication of online comments, Reform UK appeared in as many as 88% of AI Overview responses in Google Search. What only a few years ago might have been dismissed as irritating spam is now proving to be an effective way to reach almost every internet user.

Regulation cannot wait

According to Jakub Szymik, president of the Fundacja Obserwatorium Demokracji Cyfrowej, this is no coincidence.

“Search through large language models is increasingly replacing traditional search. Just look at how aggressively Google promotes AI Overview, or at the growing popularity of ChatGPT,” says Mr. Szymik.

He adds that services designed to improve the visibility of specific content in chatbot responses are already being commercially offered.

“There are already tools on the market providing ‘SEO for language models’ in order to position particular content more effectively. It is difficult to imagine Polish politicians not making use of them ahead of the 2027 elections. Similar tactics are also employed by disinformation actors, who flood the online space with large volumes of text so that models begin treating the information seriously. Even though companies such as Anthropic claim to test their systems for neutrality,” notes the head of the Digital Democracy Observatory Foundation.

A campaign unlike any before

Taken together, all this suggests that political narratives ahead of the 2027 elections are likely to intensify on a scale we have not seen before.

For now, according to data from Res Futura, the websites best optimized for language models belong to smaller Polish political parties – Razem, Konfederacja, and, on the governing side, Polska 2050. The largest players on Poland’s political scene perform poorly, in part because their election platforms remain outdated and stored in archaic formats on their websites. That is almost certain to change before 2027.

Regardless of how well websites are adapted to language models, however, much of the power still rests with the American owners of the largest AI systems. At any moment, one of the technology giants could tune its model in favor of – or against – a particular party for reasons of its own. And if not the technology companies, then politicians themselves may attempt to exert influence.

Those concerns are shared by Jakub Szymik.

“Preliminary studies show that, for large chatbots, positioning on Wikipedia is exceptionally important, as it remains the main source of responses concerning politicians – at least in the United States. Politicians, however, are also seeking to exert direct control over model outputs. The current US president has issued an executive order against ‘woke AI’, aimed at limiting content related to minorities,” says Mr. Szymik.

This makes it all the more important for EU member states and the European Union itself to urgently examine the biases and preferences embedded in AI models and develop regulations that protect the transparency of democratic processes. Spain has already begun doing so. In Poland, such a debate will be essential ahead of the 2027 parliamentary elections. Time, however, is short, while the potential scale of disinformation – and the “feeding” of chatbots with increasingly extreme narratives – continues to grow.

Key Takeaways

1. Political parties are adapting their websites to the requirements of artificial intelligence in order to dominate virtual recommendations. Effective content optimization allows parties to reach voters directly, bypassing traditional communication channels. The absence of coherent legal regulation in this area poses a genuine threat to the transparency of future electoral processes.
2. Society is increasingly beginning to treat artificial intelligence as an objective source of political knowledge and is becoming more willing to base electoral decisions on it. In the United Kingdom, the share of citizens using chatbots to fact-check information and search for details about specific candidates is rising steadily. Crucially, a clear majority of users regard AI-generated responses as both neutral and accurate.
3. Language models demonstrate far greater persuasive power than traditional campaign tools because of the sheer density of information they provide. Yet increasing that persuasive effectiveness often comes at the expense of factual precision and can lead to the dangerous amplification of disinformation. The phenomenon is particularly visible in algorithms promoting highly polarizing political narratives.