AI chatbots can be useful for explaining a complicated topic in plain English. A new Demos study reported by the Guardian is a reminder that they can also be a risky place to get hard facts about elections.
According to the report, Demos tested five free AI tools before May’s Scottish Parliament election by asking 75 questions about real constituencies. The thinktank said the systems gave misinformation in 34% of the questions it posed. The examples reported included invented scandals, wrong election dates, incorrect claims about voter ID rules in Scottish elections, and candidates being placed in the wrong contests.
That matters well beyond Scotland. Many people now use chatbots and AI search tools in the same casual way they once used a search box: to ask what something means, who is involved, what the rules are, and what they should do next. For low-stakes questions, a rough answer may be good enough. For voting information, rough is not good enough.
Why election questions are awkward for chatbots
Election information looks simple until you ask for something specific. A voter might want to know whether they need photo ID, who is standing in their area, which voting system is being used, when postal votes must arrive, or whether a claim about a candidate is true.
Those details change by nation, election type, date and constituency. They also depend on official sources being up to date. A chatbot that gives a confident answer from old training data, muddled web results or weak citations can sound helpful while being quietly wrong.
This is the same practical problem ManyHands has covered before in pieces such as why instant AI answers still need human checking. The faster the answer arrives, the easier it is to forget the checking bit.
What the reported mistakes tell us
The reported errors are not just small wording slips. If an AI tool invents an expenses scandal, gives the wrong polling date, or says voters need ID when they do not, it can affect how people understand the election itself.
It is especially awkward because chatbots often present answers in a calm, polished voice. They do not always look uncertain when they should. They may cite links that are missing, broken, irrelevant or out of date. A reader can come away with a neat summary, but no reliable trail back to the original evidence.
The Guardian reported that the Electoral Commission wants stronger legal controls around AI misinformation. That policy debate will take time. Ordinary users need a simpler rule they can apply now: treat AI election answers as a starting point, never as the authority.
What UK voters should check directly
If you are asking an AI tool about voting, use it for plain-English context only. For anything that could change what you do, go to an official source before acting.
That means checking the Electoral Commission, your local council, your nation’s election office where relevant, or the official candidate and party material you can inspect directly. For deadlines, registration rules, postal voting, proxy voting, polling stations and voter ID, the official page should win over the chatbot every time.
For claims about candidates, be even more careful. AI tools can blur old stories, local rumours, similar names and unrelated constituencies. If the claim matters, look for named reporting from a reputable publisher, the candidate’s own statement, official records, or a fact-checking organisation. If you cannot trace the claim back to something solid, do not treat it as settled.
How to use AI without being led astray
AI can still be useful during an election. It can help explain terms such as proportional representation, manifesto, constituency, turnout or devolved powers. It can summarise a long official page after you have opened the real source. It can help you turn a confusing rule into a checklist of things to verify.
The key is to keep the jobs separate. Let the AI help you understand the question. Do not let it be the final source for the answer.
A good prompt might be: “Explain what I should check on the Electoral Commission or council website before voting in this election.” A risky prompt would be: “Tell me whether I need ID to vote tomorrow.” The first points you towards verification. The second invites the chatbot to sound certain about a detail that may depend on where you are and which election is happening.
Why this will keep coming up
AI search and chatbot tools are becoming part of everyday information habits. Google, OpenAI, Meta, Anthropic and other companies all want their assistants to be the place people ask first. That makes accuracy, sourcing and freshness more important, not less.
It also means users need a better instinct for when an answer is merely convenient. Election information, medical questions, money decisions, legal rights, safety issues and anything involving real-world deadlines should all trigger the same pause: where did this answer come from, and can I check it?
For most everyday AI use, that pause is not a rejection of the technology. It is just sensible use. A chatbot can help you get oriented. The final step still belongs with you and the source that is actually responsible for the facts.
Source: The Guardian.
