Nektar Insights
College football / The polls
College football · AP poll · 1989–2025

Not every ranked win counts.

Beat a team ranked in weeks 2–6 and it finishes the season ranked only 43.5% of the time. That's worse than a coin flip. When you play them matters, and so does who you beat.

By David Hyde · Nektar Insights · 20 November 2026
Live: 2026 résumés, graded · AP poll before game week 12

I've been saying this for ten years. September 2016: "Top 25 rankings is silly at this point. Pay no attention to them until week 5." 2019: "Early season rankings mean nothing. You cannot know with just two games under your belt."

In 2018 I wrote a whole post on it. Three seasons of polls. My conclusion: "the rankings mean little to nothing until well late in the season."

I learned it as a Utah fan. In 2015 Utah started unranked, beat Michigan, went to Oregon and won 62–20, and climbed to #3. Then it lost at USC and finished #17.

Smart people predict how the season will go. Forecasting wins and losses, strengths and weaknesses. Last year's team and recruiting hype set those rankings.

And I get it, we want to tune in on a Saturday night to a #25 v #7 matchup. It gets us fired up. Unfortunately, it's hard to predict. In reality, we don't know how good a team is until November.

The AP poll is flawed. Not because it's subjective, but it's a timing issue.

So I checked again, with 37 seasons instead of three. For every ranked win: did the team you beat finish the season ranked? In September, usually not.

Here's why it matters. In September 2025 I wrote that the polls are "just marketing ploys. But unfortunately they do matter when the CFP reviews the season and the résumé says, 'X number of wins over top 25 teams.'" That's the list everybody shares. So I re-counted it.

Ten years of my calls, graded

My posts vs 37 seasons of data

Before I re-count anyone else's wins, here's my own record. Every call below is something I posted. Every grade comes from the data on this page.

  1. Sep 2016
    "Pay no attention to them until week 5."
    Wins over a team ranked in weeks 2–6 hold up 43.5%. Worse than a coin flip.
    Held up
  2. Sep 2018
    "Week 10, early November, is when the rankings start to see the most significant correlation."
    Weeks 7–10: 53.8%, barely past a coin flip. Weeks 11+: 65.0%. Early November is where it turns.
    Close enough
  3. Sep 2018
    "Really, the rankings should only be for the top five teams."
    Preseason #1–5 finish unranked 8.6% of the time. But #6–10 only 25.4%. I drew the line too high. It's the top 10.
    Partly right
  4. Sep 2019
    "You cannot know with just two games under your belt."
    Wins over a team ranked in the preseason poll hold up 41.7%. The poll before anyone plays is the worst one.
    Held up
  5. Sep 2024
    "Used for market purposes."
    I can't prove motive. What I can show: at the same final record and the same power rating, a preseason #6–10 team is 4.6 points more likely to finish ranked than an unranked one. August still echoes in January.
    Can't prove it
  6. Sep 2025
    "They do matter when the CFP reviews the season."
    This one's still open. 39 programs have a 2026 ranked win so far. See how their résumés grade today →
    Live
01

The catch

Wins over an AP-ranked team, 1989–2025

Take every win over a team that was in the AP Top 25 at kickoff. Then check the final poll. Did the team you beat finish ranked? Early in the year, usually not.

It gets better every month. Weeks 7–10 are barely better than a coin flip. By bowl season, about four in five ranked wins hold up, because by then the poll has seen enough football.

Coin flip

How to read itBar = share of beaten teams that finished in the final AP Top 25. The tick marks the value; the lighter end of each bar is the 95% CI. Weeks 2–6: 950 wins, 40.1–47.0 range. Bowl and CFP wins hold up 80.4%.

02

Last season, graded

The 2025 preseason ballot

The first poll comes out before anyone plays. Here is how two of its teams traveled through 2025, poll by poll. The full ballot, marked up, is in the next section.

Penn State opened at #2. It finished unranked. Indiana opened at #20. It finished #1.
12 of 25preseason teams finished unranked
3 of 10from the top 10
03

Every year, not a fluke

37 preseason ballots

One column per season, one row per spot on the preseason ballot. A solid mark finished in the final AP Top 25. A red strike did not. Every season since 1989, between 5 and 15 of the 25 preseason teams finished unranked.

finished rankedfinished unrankedClick a season to pull its ballot below

By preseason rankFinished unranked: #1–5 8.6% · #6–10 25.4% · #11–25 52.6%. The top 10 mostly holds. Below it, a preseason ranking is close to a coin flip.

finished unranked
from the top 10
Pick a season
04

It depends who you beat

Weeks 2–6 wins

A top-10 team in September is usually good. A #11–25 team in September is a guess. Most early ranked wins come against the guesses: 691 of the 950 wins in weeks 2–6.

Coin flip
05

This season, graded live

2026 · AP poll before game week 12

Here's 2026 so far. Every win over a team that was ranked at kickoff, graded two ways. Is the team you beat still ranked today? And how many of those wins did history say would hold up? It's all in pencil until the final poll.

How to read itEach chip is one ranked win: the opponent's rank at kickoff → its rank today. Solid = still ranked. Struck = dropped out. "History expected" adds up how often wins like these held up, 1989–2025, by week and by whether the opponent was top 10. 39 programs have a ranked win so far.

The 2026 preseason ballot, in pencil

8 of the 25 teams on the 2026 preseason ballot are already out of the poll. That's fewer than most years: only 6 of 37 past seasons had fewer out by this point. Ask me again in January.

UpdatesThese numbers move every week until the final poll. Past seasons at this point: 37.4% of preseason teams out on average.

out of the poll today
from the top 10
06

Find your team

The re-count, 1989–2025

Same list everybody shares: wins over a team ranked at kickoff. Now count only the wins where that team finished the year ranked. Florida moves from 8th to 3rd. Virginia Tech drops from 22nd to 32nd.

ScopeThe 30 programs with the most ranked wins, 1989–2025 (the AP Top 25 era). Totals differ from the all-time chart, which starts in 1936. "Held up" means the team you beat finished in the final AP Top 25.

07

"They only fell out because they lost to you"

The obvious rebuttal
08

Is it getting worse?

Top 20 basis, 1976–2025

Maybe. Early ranked wins held up 45.7% of the time in 1976–88 and 34.3% in 2014–25. The AP only ranked 20 teams before 1989, so every era here uses a Top 20 rule.

CaveatThe range on that drop is wide: 3.1–19.5 points. More games and conference title games give an early-ranked team more chances to fall out, but I haven't tested why.

09

How we checked

Method + data quality

The games. Every ranked win on this page is a real result: every FBS game since 1976, from CollegeFootballData.com, matched to the AP poll published before kickoff.

Two truths. Every result is checked against the final AP Top 25 and against a poll-free power rating. The rating is a margin-of-victory model fit to each season's FBS games, with the game in question left out.

Uncertainty. The ranges are 95% intervals from a bootstrap that resamples whole seasons 1,000 times.

The preseason poll echoes. Compare teams with the same final record and the same power rating. A team ranked #6–10 in the preseason is still 4.6 points more likely to finish ranked than an unranked one (range 2.7–7.0). That is a prediction, not proof that voters anchor: a noisy power rating can produce the same pattern.

CheckResult
Final poll, every seasonPresent for all seasons, 1976–2025
Team names matchedEvery team name in the polls matched a team in the game results (0 unmatched)
Game results vs Wikipedia25 random ranked wins, 1989–2025 (seeded, 6–7 per era): all 25 final scores match. All 25 AP ranks at kickoff match. On 4, the team page shows the CFP or Coaches rank; the AP poll for that week confirms ours.
Polls vs Wikipedia1987 preseason + final, 2012 week 5 + final: exact, except the order of tied teams
Week alignmentThe source labels game weeks one week high in 11 seasons (1976–80, 1983–84, 1986–87, 1990, 1998). Test: a ranked team that loses must drop in the NEXT poll. Fixed per season; all 11 pass.
Missing polls2001 weeks 4, 5, 9 · 2003, 2004, 2020 week 2. A game uses the latest earlier poll.