
Francesca Hong, the socialist darling of the Wisconsin gubernatorial primary, had a 20-point lead in the polls heading into election night. She lost to moderate David Crowley. Twenty points. Gone. Not shaved down, not narrowed in the final days — evaporated, like the number never existed.
That was three weeks ago. And somehow, we're still supposed to take the midterm polls seriously.
Michigan tells the same story from a different angle. Abdul El-Sayed, the radical who won the Democrat Senate primary, was polling at double-digit leads over his primary opponents for weeks. His actual margin of victory? One point. The polls didn't miss by a little. They missed by a universe. And now those same polling outfits are telling us El-Sayed is positioned to beat GOP nominee Mike Rogers in November.
The Real Clear Politics averages right now show Democrats leading by 8 points in North Carolina and 7.5 points in Georgia. Not New York. Not California. North Carolina and Georgia — states President Trump carried, states with Republican governors, states where the conservative infrastructure is deep and active. If you believe those numbers, you also believe the Wisconsin polls were just having a bad day.
The core problem isn't complicated: fewer people answer polls. The ones who do aren't representative. And pollsters fill the gaps with subjective adjustments that consistently — not occasionally, consistently — tilt the results toward Democrats.
The acceptable margin of error in polling used to be 1-4 points. What we're seeing in 2026 is closer to double digits.
This isn't a new phenomenon, either. In 2004, exit polls projected John Kerry winning 26 states. George W. Bush won the election. The polling industry promised reforms. Two decades later, the same systematic bias shows up in the same direction every cycle. At some point, the word for a consistent error in the same direction stops being "error" and starts being something else.
The races to watch tell you everything about where the polling industry is trying to steer the narrative. In Minnesota, Republican Michelle Tafoya is running against leftist Democrat Peggy Flanagan, and the polls conveniently show the race as uncompetitive. In New Hampshire, where the primary isn't even until September 8, polls already have Rep. Chris Pappas comfortably ahead of likely GOP nominee John Sununu — a Sununu, in New Hampshire, where that name has won statewide races for decades. They're even floating fantasy scenarios about Susan Collins's Senate seat in Maine being competitive.
The polling industry's defense is always the same: methodology is evolving, response rates are a challenge, we're doing our best. Which would be more convincing if the misses went in both directions. They don't. When polls are wrong about a Republican, the Republican outperforms. When polls are wrong about a Democrat, the Democrat underperforms. Every cycle. Every state. The thumb is always on the same side of the scale.
The purpose isn't prediction. It's suppression. A Republican voter in Ohio who sees polls showing his candidate down by 9 points might not bother driving to the polls. A donor in Texas who sees Kansas and Iowa listed as "toss-ups" might redirect money to defend seats that were never actually in danger. The polls don't reflect the electorate. They attempt to shape it.
The numbers are only as honest as the people building the models. And the models keep breaking in the same direction.


Comments are closed.