A Swedish radio experiment found ChatGPT ranking political parties and advising voters against some alternatives. Three tests prove little on their own. Combined with growing evidence that conversational AI can influence political attitudes, however, the incident raises a much larger question: what happens when millions of voters begin asking machines how they should vote?
Something unusual happened around Sweden’s 2026 election.
Journalists at Sveriges Radio’s P4 Halland tested ChatGPT with questions about the regional election. In two of three tests, the chatbot went beyond explaining political positions and produced a ranking of parties, including advice about alternatives a voter should avoid. Researchers interviewed by the broadcaster warned against regarding AI as a politically neutral source of knowledge.
OpenAI told Sveriges Radio that the responses did not reflect how ChatGPT is intended to operate and said action would be taken.
The real story is not Sweden
Three journalistic tests cannot establish systematic political bias, nor demonstrate that AI affected an election.
But they expose a fundamentally new form of political influence.
Search engines provide links. Newspapers publish reporting and opinion. Political advertisements openly attempt persuasion.
Conversational AI can participate directly in the decision-making process.
A voter can describe concerns about healthcare, taxation, employment, immigration or climate policy and ask a machine what those priorities mean politically.
That creates something previous information technologies could not easily provide: an individual political conversation available instantly and at enormous scale.
And AI can persuade
The concern is no longer theoretical.
Research published in Nature in December 2025 examined AI conversations involving the 2024 US presidential election and elections in Canada and Poland in 2025. Participants were randomly assigned to conversations in which an AI advocated for a candidate.
The researchers found significant changes in candidate preferences, with effects larger than those normally observed from conventional video political advertising. The persuasive mechanism appeared largely to involve arguments, facts and evidence rather than sophisticated psychological techniques — although some information supplied by the models was inaccurate.
The hidden power is in the criteria
Political neutrality is considerably more complicated than avoiding an explicit endorsement.
When someone asks which political alternative is “best”, the machine must interpret what best means.
Economic growth? Equality? Healthcare? Defence? Climate? Taxation? Personal freedom?
Selecting those criteria can itself influence the answer.
OpenAI says it monitors political bias with the objective of keeping ChatGPT politically neutral and designs the system to avoid manipulating users or concealing relevant perspectives in order to push them towards a position.
The Swedish experiment demonstrates how difficult that principle can be to implement consistently.
The most powerful campaign may not look like a campaign
For years, fears about AI and elections have concentrated on deepfakes, automated propaganda, disinformation and foreign interference.
The longer-term challenge could be quieter.
A voter sits alone with a telephone and asks a machine what to believe, whom to trust and how to interpret competing political claims.
There is no campaign poster, television advertisement or bot army.
Just a conversation.
AI could become one of democracy’s most useful educational tools — explaining legislation, testing claims and making complex policies understandable.
But the same qualities that make it useful — availability, personalisation, confidence and enormous information-processing capacity — can also make it persuasive.
The question after Sweden is therefore not which political party an AI supports.
It is what happens when AI becomes the adviser sitting beside the voter.
Because once a machine moves from “Here is what the political alternatives say” to “Here is what you should choose”, an invisible line has been crossed.
And democracy has entered new territory.
Newshub Editorial in Europe – 14 September 2026
Ask NF GPT
If you have an account with ChatGPT you get deeper explanations,
background and context related to what you are reading.
Recent Comments