For the past five weeks, Full Fact has prioritized fact-checking the May election. We’ve created over 15 fact-checks, crowdsourced claims, and used AI tools to monitor media and social media at scale. And now, as voters in Scotland, Wales and many parts of England prepare to head to the polls, they have some time to think. What did we learn?
Some of the things we saw felt really new. One clear lesson is that AI imagery is no longer a hypothetical element in British elections, but is being used in increasingly complex and sometimes surprising ways.
We saw this on social media. AI tools were used to create a fake image of Reform UK MP Sarah Pochin and a fake video purporting to show voters on the street saying they would vote for Reform and Recovery.
But we also saw it from politicians themselves. The independent candidate running for the Glasgow Southside constituency in the Scottish Parliament election has shared two AI-generated videos of himself meeting voters, addressing rallies and visiting schools, which he described as “illustrative” and said the events depicted in the videos were “not real”.
The SNP also did not directly confirm whether it used AI when editing John Swinney’s photo to sharpen the sign behind him, but distortions in the edited version caused confusion online. Some social media users questioned the authenticity of the image, even though it actually depicts a real event.
What’s worth noting about these examples is that they’re not all the same type of problem. Yes, in some cases, AI is being used to create completely fabricated images that are passed off as real. However, it is also used illustratively, raising questions about how prominent labeling should be. And minor photo editing, whether using AI or not, is creating confusion as to whether a genuine image is actually a complete fake. The challenge for fact checkers is no longer just identifying outright fakes. The line between what is authentic and what is not is becoming blurred.
At the same time, much of what we’ve covered in recent weeks has been reassuringly old.
Our biggest investigation into this campaign focused not on deepfakes or AI, but on a phenomenon we’ve been writing about for more than a decade: misleading bar graphs on local election flyers.
Our report, published on the front page of the Guardian newspaper, found more than a dozen examples of leaflets that misled people about their chances of voting, making claims that were unsourced or unsupported by reliable evidence. Of course, there is nothing wrong with political parties appealing to voters. Adequate data may not be available, especially for local elections. However, I was disappointed to learn that political parties too often make exaggerated or dubious claims in their leaflets, based on cherry-picked, misleading or unreliable data.
In the end, one of the takeaways from this campaign was simply the scale of the challenge.
With local elections taking place in much of England and parliamentary elections in Scotland and Wales, with thousands of candidates standing, there was a lot to monitor. So this campaign led us to experiment with new ways to monitor what voters are actually seeing.
This included using Full Fact’s AI tools to monitor claims made by many candidates on social media, based on accounts listed by Friends of the Democracy Club. This helped us find claims we wouldn’t have seen otherwise, such as incorrect statistics by the checkered Simul Senedd candidate on youth unemployment, which the party removed after fact-checking.
It also meant relying on readers to alert us through election bulletins to claims they saw locally. For example, many readers sent us examples of reform claims on council tax, which we included in our analysis of what the party did or did not promise on this topic.
Once voting begins, the buzz of the 2026 campaign will quickly fade away. But we will continue to grapple with some of the questions it raises, including AI, transparency, and the challenges of monitoring an ever-expanding information landscape.
