A benchmark poll is a long, one-time survey a campaign runs near the start of a race to learn where it stands and which messages might work. A tracking poll is a short survey repeated night after night in the final weeks, pooled into a rolling sample so the campaign can spot movement fast. The benchmark tells a campaign what to say. The tracking poll tells it whether that’s working.
Most of both never reach the public. What you see in the news is mostly media and university polling, plus the occasional campaign poll released on purpose. Knowing how the two campaign tools work helps you judge those releases, and it explains why daily trackers jump around so much.
What is a benchmark poll? #
A benchmark is usually the first big piece of research a campaign buys. It’s fielded months before Election Day, often more than a year out for a Senate or governor race, and it’s built to diagnose rather than predict.
That means a long questionnaire, commonly 20 to 30 minutes, and a sample big enough to break into subgroups: often 600 to 1,000 or more respondents statewide and fewer in a House district. A typical benchmark covers:
- Name recognition and favorability for the candidate and likely opponents.
- Issue priorities, so the campaign knows whether voters care more about prices, schools or something else.
- Message tests, such as “If you knew Candidate A cut property taxes, would you be more or less likely to support her?”
- Vulnerability tests, running the strongest attacks on both sides to see which ones land.
- An initial ballot test, the head-to-head number, which matters least at this stage.
Because it’s long and large, a benchmark is expensive, and a campaign may run only one or two per cycle.
What is a tracking poll? #
Tracking polls run in the last few weeks, when ads are on the air, debates happen and undecided voters start choosing. Instead of one big survey, the pollster interviews a small group every night and combines the most recent nights.
Here’s how a three-night rolling sample works:
- Night 1: interview 150 likely voters.
- Night 2: interview 150 different likely voters.
- Night 3: interview 150 more, then report the combined 450.
- Night 4: interview 150 new people, drop Night 1, and report the new combined 450.
The questionnaire is short, often just the ballot test, favorability, a few issue questions and demographics. Campaigns use trackers to see whether a new ad moved anyone, whether a debate stumble did lasting damage, and where to send the candidate in the final week.
Some public pollsters publish daily trackers too, most often for presidential approval, using the same rolling design.
Tracking polls vs. benchmark polls side by side #
| Benchmark poll | Tracking poll | |
|---|---|---|
| When | Early, months before the election | Final 3 to 6 weeks |
| Sample | Large, often 600 to 1,000+ | Small nightly samples (100 to 200) pooled over 2 to 4 nights |
| Length | Long, 20 to 30 minutes | Short, a few minutes |
| Purpose | Strategy, messages, target groups | Momentum, reaction to events |
| Stability | Stable snapshot | Noisy from day to day |
| Cost | High per poll, run rarely | High in total because it runs every night |
Why do tracking polls bounce around so much? #
The margin of error for a sample of 450 is about ±4.6 points, already wider than a typical 1,000-person poll. The day-to-day change is noisier still. When a three-night tracker moves from Tuesday to Wednesday, two-thirds of the people in both samples are the same. The entire change comes from swapping one night of about 150 interviews for another, and a sample of 150 has a margin of error near ±8 points.
So a “3-point overnight surge” in a tracker is usually noise. Wait for a shift to hold for several days, then check it against other pollsters.
Trackers also pick up a subtler problem. After a bad news week, one candidate’s supporters may simply stop answering surveys while the other side’s supporters answer eagerly. The poll moves even though few minds changed. Our explainer on non-response bias covers how pollsters try to correct for it.
Where do brushfire polls fit? #
Between the benchmark and the final tracking stretch, many campaigns run mid-sized check-ins called brushfire polls, usually 400 to 600 interviews. A realistic sequence for a 2026 Senate campaign:
- Benchmark (2025): the campaign learns the candidate is little known and that voters rank the cost of living first. The core message comes out of that.
- Brushfire (spring and summer 2026): a check on whether name recognition is rising, plus a test of how to answer the opponent’s first attack ad.
- Tracking (October 2026): nightly interviews show whether late ads and debates are moving the undecided voters the campaign needs.
How to read campaign polls that go public #
Campaign polls are private by default. When one appears in the news, someone chose to release it, usually because the numbers help. Before you believe it:
- Find out who paid. A poll paid for by a candidate or allied group is an internal, even if a reputable firm ran it. Our piece on who pays for political polls explains the difference.
- Ask for the full questionnaire. A memo with only the ballot test may hide message-test questions that pushed respondents one way before the vote question.
- Check the population. Late-race numbers should come from likely voters. A registered-voter result in October is a softer measure.
- Compare it with the average. A single internal showing a tie in a race where public polls show a 6-point gap is a data point, not a trend.
That last check is the fastest. In Election Tracker, each Senate and governor race shows a 30-day average of published polls, with the pollster, sample size, population and field dates on every card, so you can see at a glance whether a newly released poll sits near the others or far from them. Its approval screen works a lot like a tracker built from everyone’s polls: a 7-day average of every published approval poll, which keeps one pollster’s loud week from moving the headline.
Frequently asked questions #
Are tracking polls more accurate than benchmark polls? #
They answer different questions. A benchmark gives a stable, detailed picture with enough interviews to study subgroups. A tracking poll is better at showing direction in the final weeks, but its daily numbers carry wider margins of error.
How many people are in a tracking poll? #
Nightly samples are usually 100 to 200 interviews, combined over two to four nights into a reported sample of roughly 300 to 800. The pooled number is what gets a margin of error, and the nightly pieces are much noisier.
Why do campaigns keep their polls private? #
Benchmark and tracking polls contain the campaign’s strategy: which messages work, which attacks hurt and which voters are persuadable. Releasing them would hand that to the opponent. Campaigns release internals selectively, usually to reassure donors or change a news story.
Do public pollsters run tracking polls? #
Some do, particularly for presidential approval and in the final weeks of presidential races. Most public media and university polls are single surveys fielded over a few days. A few use panels that re-interview the same people over time, a different design explained in our guide to panel vs. cross-sectional polls.