The whole industry runs on one rule: they voted last time, they’ll vote again. On the low‑turnout races that decide most offices, it misses real voters and targets the wrong ones.

That “likely voter” every PAC, campaign, and consultant leans on — and the score the big data vendors ship — is really just a participation rate: of every general election on a voter’s record, the share they actually turned out for. It’s a rear‑view mirror. For a high‑turnout presidential general it’s mostly fine. But in the primaries, municipals, and off‑cycle races that decide most offices, last time’s voters and this time’s voters are barely the same people. The Civly Turnout Score predicts who will actually turn out in this election — a calibrated turnout probability on every voter, delivered as a single ready‑to‑use score.

Tested across

Nine states

Calls correct, 2024 general

78% → 84%

On a low‑turnout race

67% → 83%

New data needed

None

The bet everyone makes

The rule breaks exactly where it counts — and it costs you twice

Rank the “reliable” voters by how many recent generals they turned out for, then build your poll sample, your targeting, and your GOTV around the top of the list: that’s the standard playbook, shared across PACs, campaigns, and consultants alike. It holds up in a high‑turnout general, when nearly everyone votes. It falls apart on a low‑turnout primary or municipal, where who shows up is a different question every cycle — and when it’s wrong, it’s wrong in two expensive directions at once.

Miss the real voters

The people who’ll show up aren’t on your list

The motivated, sporadic, and newly registered voters who don’t match last cycle’s pattern get ranked too low to make the cut — so no one ever calls them, and the model quietly writes off people who were going to vote.

Target the wrong ones

62% actually vote

Much of the list you build stays home

On a real low‑turnout race, of the “likely voters” the rule flags, only 62% actually cast a ballot. Nearly four in ten — and the budget behind them — go to people who don’t turn out.

On that same race — a 2025 municipal — the rule got just 67% of its calls right, barely two‑thirds, on exactly the elections that decide the most offices.

Measured against real turnout

The Civly Turnout Score gets it right far more often — and holds across nine states

We score every voter before an election the model has never seen, then grade each call against who actually turned out — a forward forecast, not a backward tally. On that low‑turnout race it calls 83% right, against 67% for the rule. And on the 2024 general, across nine states, it is more accurate in every one.

67%rule of thumb → 83%Civly Score

Calls correct on a low‑turnout race

For each voter, does the call — will vote or will not — match what they actually did? On a 2025 municipal, right 83% of the time, against 67% for the participation‑rate rule in the file today.

And it isn’t one race. Here is how often each is right when it says a voter will — or won’t — turn out, state by state, on the 2024 general (every one a statewide election, so the whole electorate was on the ballot):

State	Rule of thumb	Civly Score
Kansas	71%	87%
Delaware	76%	86%
Missouri	77%	86%
North Carolina	78%	86%
Pennsylvania	79%	83%
Florida	78%	81%
District of Columbia	78%	80%
Rhode Island	80%	82%
Michigan	85%	87%

The Civly Score is more accurate in all nine states — 84% of calls correct on average, versus 78% for the rule of thumb. Every score is graded against official state turnout records, and the model only ever sees information available before each election, so nothing is contaminated by the outcome it predicts.

The whole system

Turnout is one stage of Civly’s closed loop

The Turnout Score is one piece of a single cycle that runs off one scored voter file — poll the electorate, test your message, target the voters who move, weight for who will actually show up, then act and re‑poll to measure the lift. See how the whole loop fits together, or talk to us about putting it to work on your voter outreach.

See the closed‑loop cycle Talk to us about your outreach

Measured against real recorded turnout the model had not seen — the 2024 general across nine states, and a 2025 municipal · the foil is the participation rate the field relies on — of every general election on a voter’s record, the share they turned out for · graded turnout-matched (as many predicted voters as truly turned out).