Every poll has Brad Raffensperger in third. The most recent, InsiderAdvantage (May 16–17), puts him at 16%, behind Rick Jackson at 31% and Burt Jones at 27%. Thomas Marketing analyzed the early vote file and thinks those numbers are wrong. Not because the pollsters are bad at their jobs. Because they're measuring the wrong people.
Georgia has an open primary. Anyone can walk in and pull a Republican ballot, Democrat, independent, recent party switcher, whoever. But every likely voter screen in this race is built on one thing: past Republican primary participation. If you've never pulled a GOP ballot before, you don't exist in these models. Even if you're about to.
The early vote file says roughly 32,000 of those people already showed up.
What the Early Vote Data Actually Shows
Thomas Marketing cross-referenced all 348,715 Republican ballots cast through May 15 against the L2 voter file, the same file both parties use, to find out who actually pulled them.
Nearly 32,000 voters who L2 doesn't model as Republicans pulled a GOP ballot in early voting. Every pollster missed them. They have no history of doing this, so the models don't see them coming.
Why Those Voters Are Going to Brad
The 19,380 modeled Democrats who crossed over aren't MAGA voters. They watched Raffensperger's feud with Trump play out over the 2020 election, and they're not going to walk into a Republican primary to vote for Trump's endorsed candidate. They're there for Brad.
The 12,373 Non-Partisan crossovers tell the same story. Non-Partisans don't typically show up in a GOP primary, especially one saturated with MAGA messaging from Jackson and Jones. The ones who did aren't voting for either of them.
The Polling Blind Spot, Pollster by Pollster
The May polling averages, Quantus at 14%, Remington at 14%, Cygnal at 12%, are all measuring Raffensperger inside a universe that structurally cannot see crossover intent. Here's how each firm's screen works against finding these voters:
University of Georgia / AJC (April 30, 1,000 LV ±3.1%): The gold standard in Georgia polling, and the most rigorous at excluding crossovers. UGA requires respondents to have voted in a recent GOP primary and to be planning to vote again. They also weight for party identification. A Democrat who has never pulled a GOP ballot can't qualify, by design. It's the right screen for a closed primary. Georgia doesn't have one.
Remington Research (May 4, 815 LV ±3.3%): Samples from “likely Republican primary voters selected and weighted to reflect the expected turnout demographics of the 2026 Republican primary electorate.” Expected turnout demographics means past primary participation. The crossover universe doesn't exist in that model.
Quantus Insights (May 5, 1,677 LV ±2.7%): Largest sample in the field. Publishes nothing about its likely voter screen methodology. No evidence they're capturing voters L2 codes as Democratic or Non-Partisan who intend to pull a GOP ballot Tuesday.
InsiderAdvantage (May 16–17): The most recent poll, showing Raffensperger at 16%. That's up from 8% in February, modest but consistent, movement entirely within the core GOP universe. The final poll also shows undecideds collapsing from 25–30% in earlier surveys to just 12%, with late-breakers distributing roughly proportionally across all candidates. Nobody had a dramatic surge. The undecided pool broke evenly. Brad's path to the runoff isn't late momentum. It's 31,753 voters who have already voted and whom no poll is counting.
The Adjusted Projection
Thomas Marketing applied the early vote crossover universe to the final InsiderAdvantage baseline. We used a conservative 65% Democrat capture rate. Carr's late surge to 10% suggests some moderate crossover voters are splitting rather than consolidating entirely behind Brad:
Undecideds broke roughly proportionally. Carr is pulling some of the moderate crossover vote that might otherwise have gone to Brad. Neither of those things changes the fundamental picture.
Burt's spending backed off a bit in early voting to save powder for the runoff. Raff and Rick Jackson didn't. Rick leads and should make the runoff, but the final spot is up for grabs. The race for second is a lot closer than the polls suggest.