From The Conversation · W. Joseph Campbell · · 2m
Pollsters face a reckoning ahead of 2026 midterms
A decade of polling failures, fake surveys, and competition from prediction markets has left election forecasters scrambling to rebuild credibility.
In brief
A decade of polling failures, fake surveys, and competition from prediction markets has left election forecasters scrambling to rebuild credibility. Pollsters must solve fundamental problems with response rates and weighting methods while competing with prediction markets that claim superior accuracy. Originally…
A decade of polling disasters
Pollsters have misfired consistently since 2016, when they underestimated Trump's support. In 2020 they had their worst performance in 40 years, and in 2024 they underestimated Trump again for the third straight election.
Fake polls spread confusion
In August 2026, fabricated poll results appeared in competitive races across Wisconsin, Nevada, and Los Angeles. A 21-year-old created the fake data through a website called Median Strategies to show how false surveys attract attention.
“Should we ever trust polls again?”
Primary elections are harder to poll
Primary elections attract low voter turnout, making it difficult for pollsters to estimate who will actually vote. This challenge was compounded in 2026 when fabricated results muddied the waters further.
Prediction markets make a comeback
Betting platforms like Polymarket and Kalshi are reshaping election forecasting. Polymarket founder Shayne Coplan declared in 2024 that prediction markets are 'the most accurate thing we have as mankind right now.'
“the most accurate thing we have as mankind right now”
Markets beat polls in 2024
Polymarket and Kalshi both predicted Trump's victory in 2024, while polls gave Kamala Harris a slight edge. After the election, Coplan declared 'Trust the markets, not the polls.'
“Trust the markets, not the polls”
Markets stumbled in Wisconsin
Prediction markets gave Francesca Hong a 95 to 96 percent chance of winning Wisconsin's Democratic gubernatorial primary in 2026. She lost narrowly, showing prediction markets are not infallible.
The weighting problem persists
Low response rates force pollsters to adjust raw data through 'weighting,' aligning results with demographics and past voting patterns. But weighting choices vary by pollster and offer no guarantee of accuracy.
AI-generated polls enter the mix
Silicon sampling uses AI agents to simulate poll responses, much faster and cheaper than traditional surveys. Gallup is exploring the method, though traditionalists worry about ethical implications.
AI forecasts also failed in 2024
Startup Aaru used AI to predict Harris would win four of seven swing states, matching traditional polls. Harris lost all seven, showing AI simulation offers no advantage over conventional polling methods.
No easy fixes ahead
Pollsters face mounting pressure from prediction markets, low response rates, weighting challenges, and public skepticism. There are no guarantees these troubles will be resolved before the 2026 midterms.
The bottom line
Pollsters must solve fundamental problems with response rates and weighting methods while competing with prediction markets that claim superior accuracy.
