Ninety Days Out: What August Polling Actually Predicts

Every August, one party celebrates its summer polling and the other explains it away. History has opinions about who’s right — and they’re more specific than either side wants to hear.

Right now, Democrats are reading summer polls showing leads in Texas, a coin-flip in Michigan, and a favorable generic-ballot trend, and permitting themselves optimism. Republicans are reciting the oldest catechism in campaign politics: the only poll that matters is in November. Both postures are annual rituals. Neither is analysis. The actual question — how much does August polling tell you about a midterm outcome ninety days later? — has a real answer, and it’s worth laying out plainly, because it should discipline how you read every number this site publishes between now and Election Day.

What August reliably gets right

The honest summary of decades of midterm cycles: August tells you the direction of the year with respectable reliability, and the magnitude poorly. By late summer of a midterm year, the fundamental architecture — presidential approval, the dominant issue set, which party’s voters are angrier — is usually locked in, and that architecture, more than any horse-race number, is what November expresses. The wave years telegraphed themselves: the out-party’s summer energy in 1994, 2010, and 2018 was visible in August polling, candidate recruitment, and special-election overperformance long before the size of the wave was knowable. When every indicator points the same way in August, November has rarely pointed the other way.

What August gets wrong, chronically, is amplitude and allocation. Summer polls sample registered voters and enthusiasm; November counts likely voters and turnout machinery. Individual race polling ninety days out is noisy almost to the point of decorative — undecideds break late and usually toward the fundamentals, October events reshape closing arguments, and candidates who are famous in August (both parties’ nominees in our marquee races very much included) can still redefine or be redefined. The 2022 cycle remains the canonical caution in the other direction: a loudly forecast “red wave” built partly on a late flood of partisan-adjacent polling simply failed to arrive, because the summer fundamentals — which had never actually deteriorated for Democrats as much as the autumn narrative claimed — reasserted themselves.

Applying the rules to 2026

Run this cycle through the framework and the picture sharpens. The inputs that history says to trust are aligned for Democrats: Trump’s approval is underwater in the states that matter, the issue environment is dominated by costs — tariffs, war-adjacent energy prices, housing — which cuts against the governing party, and the summer generic-ballot trend has moved one direction. By the reliable indicators, 2026 is shaping as a Democratic-leaning year of unknown magnitude.

The inputs that history says to discount are, naturally, the ones generating the most excitement: individual toplines like Talarico’s two-to-five-point Texas leads, and exotic crosstabs like his nine-point non-college edge in one survey. Those are August numbers of exactly the type that has evaporated before — O’Rourke also had encouraging summers. The disciplined read is not “Democrats lead Texas” but “Texas is behaving like a competitive state in a Democratic-leaning year,” which is itself remarkable and, unlike the topline, likely durable.

Our commitment

Between now and November, this site will publish a lot of polls. Here is the standing promise: every one gets weighted by the rules above — fundamentals over horse race, trends over levels, October over August. When our ratings move, we’ll show which input moved them. And when a number is exciting but historically unreliable, we’ll say so in the same sentence, because the alternative — narrating every outlier as a turning point — is how audiences end up shocked by elections that the fundamentals predicted all along.