Asking an AI chatbot about politics can feel a lot like asking a well-read friend for a quick summary. It is fast, confident and often easier than opening a dozen tabs. But a new report is a useful reminder that AI answers are not a neutral window on public life. They are shaped by the sources, patterns and online noise that the system finds and chooses to repeat.
The Guardian reports that research by AI search analytics firm Peec found AI platforms were more likely to reference Nigel Farage than other UK political leaders when prompted on British politics. The work tested leading AI systems with 5,000 structured prompts over several weeks, producing more than 280,000 data points. According to the report, visibility varied by topic and platform: Reform UK appeared strongly in some Google AI Overviews, while Labour was more visible in prompts about the NHS.
The point for ordinary readers is not that one AI answer is secretly deciding an election. It is more practical than that. If more people use chatbots and AI search summaries to understand policy, local candidates or political rows, the way those systems select and frame information starts to matter.
Why AI answers can feel more trustworthy than search results
A normal search page shows you a messy list of links. You can see, at least roughly, that one result is a newspaper, another is a party website, another is a campaign group and another is a social media post. You still need judgement, but the seams are visible.
An AI answer smooths those seams away. It gives you a single response in plain English, often with a tone of balance and certainty. That is helpful when you want a quick explanation of council tax, NHS waiting lists or what a manifesto promise means. It is also risky, because the answer can make a selection of sources feel like settled common sense.
That matters especially in politics, where the same facts can be framed very differently. A chatbot may summarise a party’s position accurately but leave out context, over-emphasise a loud online argument, or make a political figure appear more central simply because their name appears more often in the material it is drawing from.
Visibility is not the same as public support
The most important distinction is simple: being mentioned by an AI system is not the same as being endorsed by it, and it is not the same as being popular with voters. AI tools often respond to patterns in available material. If a party, campaign, media outlet or group has a large and active online footprint, that can influence what appears in answers.
The Guardian’s report says Peec’s research found that AI systems frequently cited Facebook, followed by sources such as the BBC, the UK parliament website and Wikipedia. That mix should make users pause. Some of those sources can be reliable in the right context; others can contain fast-moving claims, campaign messaging or posts designed to provoke attention.
ManyHands has previously written about why source-checking matters when people use chatbots to learn. Politics raises the stakes because a tidy answer might influence how someone understands a candidate, a policy choice or a local issue before they ever read the underlying material.
What to do before trusting an AI political answer
You do not need to avoid AI tools completely. They can be useful for turning a complicated topic into a first-pass explanation. The trick is to treat that answer as a starting point, not a verdict.
First, ask for sources, then open them. If the answer is based mainly on social posts, opinion pieces or campaign pages, treat it differently from an answer grounded in official documents, parliamentary records, full manifestos, regulator statements or established reporting.
Second, ask the same question in a more neutral way. “Why is Party A better on immigration?” is likely to produce a different shape of answer from “Compare the main parties’ immigration policies, including trade-offs and criticisms.” The wording of your prompt can pull the system towards a particular frame.
Third, ask what is missing. A useful follow-up is: “What important context or counterarguments might this summary leave out?” That will not make the AI perfect, but it can reveal whether the first answer was too narrow.
Fourth, check dates. Political positions, policy announcements and local candidate details can change quickly. If an AI answer does not show when its information was last updated, look for a more recent source yourself.
Watch out for confident summaries of live events
The risk is highest around breaking news, elections and viral claims. During live events, AI systems may lean more heavily on fresh web pages and social media posts because the information is not already settled in their training data. That can make them useful but also more vulnerable to rumour, repetition and coordinated posting.
The UK’s Electoral Commission has already been paying attention to AI-driven misinformation. In April, it announced a pilot to detect political deepfakes ahead of May elections in England, Scotland and Wales, warning that false audio and video can spread quickly. Deepfakes are a different problem from AI search summaries, but they belong to the same wider shift: voters now have to judge political information in a media environment where synthetic and automated content is easier to produce.
That is why we have also argued that AI-made content should be clearly labelled when it is trying to persuade people. Labels will not solve everything, but they help people understand what kind of information they are looking at.
A better way to use AI during an election
For everyday voters, a sensible workflow is: use AI to get oriented, then use primary or clearly identified sources to make up your mind. Ask a chatbot to explain what a policy term means, summarise the difference between two proposals, or list questions you might want to ask a candidate. Then check the actual party material, council information, parliamentary record, regulator notice or reputable reporting behind the answer.
Be especially careful when an answer sounds too neat. Politics is full of trade-offs. If a chatbot presents one side as obviously right, ask it to explain the strongest criticism of that view. If it quotes a statistic, ask where it came from. If it names a source you do not recognise, open it before sharing the claim.
It is also worth separating “what does this party say?” from “is this policy likely to work?” The first can often be answered from manifestos and speeches. The second needs evidence, expert analysis and judgement. AI tools sometimes blur that line by packaging both into one confident paragraph.
The practical takeaway
AI search and chatbots are becoming part of how people discover political information. That does not make them bad tools, but it does mean voters should bring the same scepticism they already apply to leaflets, adverts, viral posts and campaign speeches.
Before trusting an AI answer about politics, ask where it came from, what it left out, whether the information is current and whether the prompt nudged it towards a particular view. If the answer helps you ask better questions, it has done something useful. If it quietly replaces your judgement, it has gone too far.
The safest habit is to let AI explain, not decide. In politics, that difference matters.
