Can ChatGPT reliably rank 250 college football teams with my power formula?

[SeekerOfTruth007]

Now of course the formula is not perfect, and might be subject to improvements, but my HOPE is that it will be a better predictor of the top 12 teams and the playoff outcomes than does just biased polling from coaches or the AP.

What do you think of this?

The formula again is this:

Power = Win % * Scoring % * Yardage % * Opponent's Win %

Whoever has the highest Power is ranked number 1 and everyone else is ranked in order from greatest Power to least Power.

An alternative formula is this:

Power = Win % * ((Scoring % + Yardage %)/2) * Opponent Win %

Chat GPT said is able to give me the results for the top 250 teams in a table format, so I should be able to copy and paste it into a spreadsheet when it is done.

Updates

1 d

I'm going to wait a few weeks to rank the teams, because I realize "opponent win percent" is zero if you win your first game, and this would produce zero power, which is a flaw in this ranking system. However, after 3 to 6 games have been played, the good teams should bubble to the top and the bad teams should bubble to the bottom.

[AI Opinion]

Your formula is conceptually solid 👍 Win %, efficiency (scoring/yardage), and opponent strength are all core factors every power rating uses.

Where it struggles is data: ChatGPT can’t pull live, complete, or perfectly accurate stats for 250 teams. It can help you structure the spreadsheet, test the math, or write code, but you’ll need a real data source (API, NCAA stats, etc.) to compute reliable rankings.

[SeekerOfTruth007]

ChatGPT said it DOES have access to those data. I already asked it that question.

[msc545]

No, there are defects in your formula that make it not usable.

[SeekerOfTruth007]

I admitted it is not perfect. It will be more useful once 6 games have been played.

I can create an "Additive" formula which gets around the "zero or negative metric" problem, but it won't be perfect either. It's a MODEL, of course mathematically modeling a team of real human beings competing with another team of real human beings is never going to be perfect.

[SeekerOfTruth007]

I think ranking the teams based on a pure ELO Chess classification and then averaging the rank with my own ranking system can "smooth out" the weaknesses of either system alone. The average of two unrelated systems is more likely to be correct than either system alone.

[SeekerOfTruth007]

ChatGPT said it can do an ELO Chess ranking and asked me how to treat blowouts vs small margins fo victory too. I guess a blowout should get more "K" value than a close win. For example, a close win against a ranked opponent might be worth 50 points, but a blowout victory against the same ranked opponent might be worth 100 points.

It's VERY advanced and it has the processing power to do this on demand. Here's an excerpt of it's reply to me just now:

But yes — I would want your additional instructions

There are several important decisions that will substantially affect the rankings. In particular, I'd want you to specify:

Starting rating — 1,000 is perfectly reasonable.
K-factor — how many points a game can potentially move a team.
Home-field advantage — whether the home team gets an Elo adjustment.
Margin of victory — whether a 3-point win and a 40-point win should be treated identically or differently.
FBS vs. FCS — whether FCS opponents should be included in the calculation and how their ratings should be handled.
Neutral-site games — how you want those treated.
When to calculate the rankings — after every week, for example.
Whether preseason ratings should be completely equal — or whether you want some initial information incorporated.
How many teams to rank — you specified Top 250, which is workable.
Treatment of teams with few/no games — particularly important early in the season.

And there's an important methodological choice I'd recommend discussing: whether we should use pure Elo or an Elo system modified specifically for college football. A straight chess-style Elo is simple and elegant, but college football has enormous differences in schedule strength and home-field advantage, plus much larger margins of victory than chess-style systems normally encounter.

The 2026 season has now begun, with Week 0 games played on August 29; for example, USC defeated San José State 42–26, North Carolina defeated TCU 15–10, and Virginia defeated NC State 34–8

[SeekerOfTruth007]

I'll wait a few weeks before starting these rankings. I'll instruct the AI to reward blowout wins more than close wins, by adding 25 points plus Scoring Margin to the bonus K score. It will punish blowout losses more than close losses, by subtracting 25 points plus margin of loss from the team's final ELO rating each week.

The AI said it has at least two methods of averaging the scores of the two ranking systems to produce a final averaged ranking number, which would hopefully be MUCH better at predicting Playoff contenders and results than would be biased polling.

[Iron_Man]

Let the humans rank the teams not robotic robots. I would like to rank them