After Two Brutal Weeks Picking College Football Games, ChatGPT Finally Turns A Profit

ChatGPT has been downright horrible picking college football games against the spread over the last two weeks. Still, given what we understand about Artificial Intelligence, it should come as no surprise that the computer has started to figure things out in its third week of making picks.

In the first go-around, ChatGPT went 14-21 with its picks against the spread. A week ago, it picked up the slack by posting a 17–16 record, but this week, the results were actually profitable.

This week, I selected 39 games from the CFB slate. I kept the prompt straightforward — telling ChatGPT to make picks against the spread (which I provided), to consider key injuries, and to "use analysis" when making the selections. The bot interpreted that as factoring in home-field advantage and look-ahead spots on the schedule.

Ty Simpson of Alabama. (Jake Crandall/ Advertiser / USA TODAY NETWORK via Imagn Images)

As in previous weeks, ChatGPT ranked each game with a confidence level ranging from 1 to 10. There wasn’t a single game the computer liked enough to give a 10, but it did pull three games with a level 9 and another six with a level 8.

Level 9

• Utah -11.5 @ Kansas
• Texas Tech -24.5 @ West Virginia
• SMU -13.5 @ Cal
• Record: 1-2

Level 8

• Notre Dame -32.5 @ Stanford
• Vanderbilt +2.5 @ Tennessee
• BYU -17.5  @ UCF
• Colorado @ Kansas State -16.5
• Iowa State -14.5 @ Oklahoma State
• Indiana -28.5 @ Purdue
• Record: 3-3

While ChatGPT did not fare particularly well with its most confident picks, it took care of business with the rest of the slate — posting an overall record of 22–17 against the spread.

For the first time in this three-week experiment, ChatGPT beat me, as I went 20–19 for the week.

The key takeaway this week is the same as the last: it is very difficult to pick winners against the spread.