Over 5,000 council seats across 136 English authorities go to the polls on 7 May 2026, alongside all 32 London boroughs, the Scottish Parliament, and the Senedd. For pollsters, it’s the most complex set of elections they’ve had to deal with since the 2015 general election.
Labour is defending over 2,100 seats while its poll numbers slide. The Conservatives defend over 1,300 and are getting squeezed from both sides. Reform UK has turned up with unprecedented local election spending and a 20-point rise since 2022. The Greens have a new leader, five MPs, and a by-election win in the north of England. With over 25,000 candidates standing across 140 parties, anyone trying to model outcomes has a serious problem on their hands.
The UK’s two-party polling assumptions haven’t been fit for purpose for a while, but these elections will make that painfully obvious. Labour polls around 20%, a long way down from the 35% that won these seats in 2022. The Conservatives sit at roughly 18%. Reform has surged to 27% nationally. The Greens poll 12–14%. The Liberal Democrats mop up what falls between the cracks.
Polling models built on historical party performance fall apart when the landscape moves this fast, and it’s worth remembering what happened in 2015 under far simpler conditions. The Independent Inquiry into the 2015 British General Election Opinion Polls, chaired by Professor Patrick Sturgis and commissioned jointly by the British Polling Council (BPC) and the Market Research Society (MRS), concluded that the primary cause was unrepresentative samples. The methods in use systematically over-represented Labour supporters and under-represented Conservative supporters. Final polls had both parties level at around 34%, but the actual result came in at Conservatives 38% and Labour 31% – a 7-point error on the lead in what was essentially a two-horse race.
Now try doing that with five competing forces, each pulling from distinct voter bases with different turnout patterns. Stephen Fisher at Oxford University projects Labour losing around 1,900 seats and the Conservatives losing around 1,010, which means two parties collectively losing three-quarters of the seats they’re defending. The historical baselines that weighting models rely on are essentially gone at that point. Who turns out, where, and for whom – all of those assumptions need rebuilding.
All 32 London boroughs holding all-out elections at once is a particular nightmare for pollsters because each borough has its own demographics, turnout patterns, and local dynamics. Tower Hamlets, where three wards have 22 candidates each, has about as much in common with Richmond as Glasgow does with Bath. But most polling still treats London as one big bucket, which means the aggregate numbers can look perfectly reasonable while being wrong in every borough that actually matters.
Online panels are especially bad at this kind of granular geographic work. London samples from panels tend to over-index on younger, more engaged voters – the kind of people who sign up for survey panels in the first place. The Sturgis inquiry made this exact point. When your recruitment method systematically pulls in certain voter types over others, weighting can only do so much to fix it. That limitation produced a 7-point lead error in a two-party contest, so across 32 separate five-party fights where borough-level accuracy is what determines council control projections, those same biases get worse, not better.
Then there’s differential turnout. In 2022, London turnout ranged from 28% to 45% depending on the borough, and there’s no automated way to test whether someone who says they’ll “definitely vote” in Barking and Dagenham means it with the same conviction as someone saying the same thing in Westminster. When margins are this tight, that gap between stated intention and actual behaviour is where projections go wrong.
Reform UK is spending over five million pounds on these locals, and Nigel Farage has called them the most important electoral test before the next general election. A party going from near-zero local presence to 27% nationally between cycles is something standard demographic weights just can’t account for, because there’s nothing in the historical data to anchor the weighting.
The timing of that spending makes things worse for automated methods. A big, concentrated push in the final weeks can shift momentum, but panels recruited months earlier lack the flexibility to respond. They can’t oversample in areas where Reform is pouring money in, and they can’t pick up voters who are only just starting to think about backing the party. It’s the same pattern we saw with UKIP before 2015, when polls consistently underestimated insurgent party support, except that UKIP never had anything close to Reform’s money or ground operation for local elections.
Research published in the Proceedings of the National Academy of Sciences (PNAS) in November 2025 found that AI bots passed 99.8% of standard survey quality checks across 6,000 trials, getting past every existing detection method the researchers tested (Westwood, 2025). The study showed that as few as 10 to 52 fake responses could flip which candidate appeared to be leading in major national polls. This research examined general survey vulnerability rather than UK political polls specifically, but the takeaway is hard to ignore. Polling commissioners should be asking much tougher questions about the integrity of online panels.
Nobody is saying every online poll is compromised, but when errors show up after election day and methodology gets picked apart – as it always does – “we used an online panel” is a much harder line to hold than it was even a couple of years ago. And when 46 councils might change hands on narrow margins, the gap between the methodology you can defend and the methodology you can’t shows up pretty quickly in the numbers.
Telephone polling with live interviewers does things automated methods can’t – verifying respondents through conversation, probing uncertain answers, adjusting sampling on the fly, and producing audit trails that hold up during post-mortems. In an election this messy – five parties, regional variation, late spending surges – that kind of flexibility stops being a nice-to-have.
These elections will be the biggest test of polling methodology since 2015, and with projected margins in dozens of councils coming down to single percentage points, any weakness in approach will be visible almost immediately once results come in. Some polls will get it wrong, because some always do, but what matters is whether the firms behind them picked methods that can handle a level of complexity we genuinely haven’t seen before.
If you’re commissioning polling for these elections, methodology is the whole game. Results will face scrutiny, reputations will be on the line, and the landscape is moving week to week. In that environment, the only approaches that give you real confidence are the ones built on verification, and May 2026 will tell us a lot about whether the polling industry has caught up with what it’s actually being asked to measure.
For insights on polling methodology in complex electoral landscapes, contact our research team.