Let’s look at the recap of the infamous NFL Hall of Fame Game, as picked by AI’s!. I tasked Claude, ChatGpt, and Gemini to take a look at the game and give me a recap of what the outcome of the game would be. [AI Predicts Tonights NFL Game]
I’m not going to lie to you. It’s a bit like the box office numbers of SuperGirl. It’s not great.
The winner was ChatGPT, although “winner” is doing some heavy lifting. None of the three agents correctly predicted Carolina’s 33–30 victory. ChatGPT came closest to understanding the game’s shape, Claude produced the most impressive-looking hallucination, and Gemini apparently wagered heavily on punts and bird interference.
The real game
Carolina defeated Arizona 33–30 after Haynes King scored on a five-yard scramble as time expired. The game was tied 17–17 at halftime, Arizona led 20–17 after three quarters, and the teams combined for 26 fourth-quarter points.
The actual stars were:
- Haynes King: 21-of-34, 180 yards, two passing touchdowns, 39 rushing yards and the winning touchdown
- Carson Beck: 15-of-19, 188 yards, one touchdown and no interceptions
- Kenny Pickett: 8-of-11 for 62 yards
- Kedon Slovis: 9-of-10 for 106 yards and one touchdown
There were only 12 accepted penalties and three total punts—not quite the preseason landfill Gemini envisioned. Official NFL gamebook | NFL recap
Final standings
| Rank | Agent | Predicted result | Accuracy grade | Entertainment grade | |
|---|---|---|---|---|---|
| 1 | ChatGPT | Cardinals 20–17 | 5.5/10 | 8/10 | |
| 2 | Claude | 21–21 tie | 3.5/10 | 9/10 | |
| 3 | Gemini | Cardinals 13–9 | 1.5/10 | 8/10 |
1. ChatGPT: accidentally nailed three quarters
This was the best prediction in the original post, largely because it identified the players who actually shaped the game.
What it got right:
- Beck started, looked comfortable and avoided turnovers.
- Pickett started and was reasonably efficient.
- Haynes King provided mobility and led a late touchdown drive.
- Arizona’s reserve receivers contributed.
- Both teams treated the evening primarily as a roster evaluation.
- Its predicted score—Arizona 20, Carolina 17—was the exact score at the end of the third quarter. That is easily the most interesting hit of the experiment.
What it missed:
- It picked the wrong winner.
- King entered during the second quarter, not just the second half.
- King scored the winning touchdown himself. Trevor Etienne did not cap the drive; he finished with only 12 rushing yards.
- Beck’s touchdown went to Simi Fehoko, not Reggie Virgil. Virgil did contribute five catches for 66 yards, so the model identified a relevant player but assigned him the wrong highlight.
- Jimmy Horn Jr. did not catch a deep ball from Pickett.
- Mitchell Evans and Chris Brazzell II were inactive.
- Bam Knight’s longest run was 15 yards, not a dramatic breakaway.
- Cameron Robertson did not record the imaginary fourth-quarter sack.
Verdict: ChatGPT built a recap from the right collection of names and correctly sensed that Beck and King would matter. Then it shuffled their accomplishments like a deck of cards. Still, predicting the exact third-quarter score earns it the crown.
2. Claude: extremely specific and extremely fictional
Claude’s response sounded the most like a finished sports column. Unfortunately, the confidence frequently outran the football.
What it got right:
- Beck and Pickett were the starting quarterbacks.
- Beck looked composed and productive.
- Harrison Wallace III contributed with two catches for 34 yards and two kickoff returns.
- Reggie Virgil was active in the passing game, leading Arizona with five receptions.
- Jacoby Brissett and Marvin Harrison Jr. did not play.
- The game was extremely close and tied for a meaningful stretch. Claude predicted 21–21; the actual halftime score was 17–17.
- Its broader premise—that backups and roster-bubble players would control the evening—was correct.
What it missed:
- It predicted a tie instead of a Carolina victory.
- Beck’s invented 9-of-14 for 112 yards was well below his actual 15-of-19 for 188.
- Minshew did not replace Beck; Kedon Slovis did.
- Pickett’s invented 7-of-10 for 68 yards was close, but the actual line was 8-of-11 for 62.
- Walter Nolen, Budda Baker and Jeremiyah Love were inactive, despite Claude giving all three detailed playing time.
- Derrick Brown and Bobby Brown III were also inactive, not carefully rationed for 11 combined snaps.
- Carolina did not produce two takeaways. The Panthers had no takeaways, while Arizona recovered Carolina’s only lost fumble.
- Ryan Fitzgerald did not miss a potential winning field goal. He made both attempts.
- Most importantly, Claude completely missed Haynes King, the player who actually dominated the story.
Verdict: Claude was the best writer but not the best forecaster. It provided the highest volume of checkable claims, which also gave it the highest volume of ways to be wrong. It resembles a veteran columnist who filed from the hotel bar without attending the game.
3. Gemini: correct about the existence of football
Gemini clearly treated the assignment as comedy rather than prediction. For the first preseason NFL game, that’s about right.
What it got right:
- Pickett started.
- The game featured second- and third-string players.
- The finish was dramatic.
- Arizona did score a late fourth-quarter touchdown using deep reserves: Slovis found Bryson Green with 1:55 remaining to take a 30–27 lead.
What it missed:
- Wrong winner and wildly low score.
- It forecast 22 penalties; the teams had 12.
- It forecast 14 punts; the teams had three.
- The Cardinals’ decisive drive was not powered entirely by defensive holding penalties.
- No linebacker tackled his own safety.
- No pass ricocheted off a defender, then a bird, then into the hands of a tight end.
- Carolina, not Arizona, produced the last-second winner.
- It failed to anticipate strong performances from either Beck or King.
- The actual game produced 803 total yards, so this was almost the opposite of the punting exhibition Gemini described. Official team statistics
Verdict: Gemini submitted a funny preseason parody with a score attached. As football analysis, it was roadkill. Possibly the same bird from the fourth quarter.
The strongest follow-up angle for your site is: ChatGPT predicted the exact score—one quarter too early. That is genuinely funny and gives the experiment a better ending than simply declaring that all three were wrong. Claude won the writing contest, ChatGPT won the prediction contest, and Gemini won whatever contest involves airborne wildlife.