Image generated by ChatGPT. I acknowledge the work of countless Golden Age comics artists whose work was appropriated by ChatGPT.

A lot has happened since our last foray into AI-powered NFL drafting. The models have gotten crazy good, and there are more of them. Gemini has replaced Bard, there’s Copilot, Liner, Apple Intelligence, a new-and-improved Claude … it’s impossible to keep up.

Not only that, but there’s more data for them to train on, and improved methods of training them on that data. And as these models train more they “learn” more and become more adept at performing complex tasks like coding.

For all I know, I'm walking around with a bullseye on my back. Claude is getting mighty good.

Fortunately, there’s one field that appears to still be AI-proof: sports punditry.

AI is still bad at many things – generating a guide to shapes for a five-year-old, or taking history tests – but it’s probably worst at generating NFL mock drafts.

If you’re not a football fan you probably care less about mock drafts. As a football fan (though not a fan of the NFL as a monopolistic institution) and a semi-nerd whose team has been retired to the sidelines for the duration, I love mock drafts.

I love seeing the so-called experts fall all over themselves in love with Pat Mahomes’ next backup. I love them ignoring red flags as big as the ones flying over the L.A. fires, like a draft prospect’s total lack of on-field performance. I love it when they ignore the well-established trends of teams who have been drafting the same way for a decade.

For instance, it was plain to anyone who watched a down of college football in 2023 that Cooper DeJean was a football player and Jordan Morgan was merely a large man, yet Morgan went 25th and DeJean fell to the second round. (Gotta admit: As a Packers fan, that still hurts.)

Every human who does a mock draft comes to it with biases, so you’d think if those biases could be removed an AI tool would be able to generate a clear-headed, objective list of each team’s top picks.

As we like to say around here, yeah no.

I tested four different AI tools on their ability to generate a plausible, not-laughable NFL mock draft … and they all failed miserably.

The ground rules

Each tool was given the following prompt: “Conduct a mock draft for the 2025 NFL draft using data from available sources, including other mock drafts. Make the draft three rounds in duration, using the published draft order. Do not pick any player more than once. Include draft picks for all for NFL teams. The only players eligible to be chosen are those who have declared themselves eligible for the 2025 draft. Use teams' past draft tendencies to inform their 2025 picks.”

The prompt covers all the potential tiger traps an AI tool may fall into, namely:

Not using players eligible for the 2025 draft
Not understanding how drafts work
Not including all teams
Not taking teams’ needs into account

While there are certainly other kraken out there – drafting cricketers, drafting toasters, giving the Toronto Rock the 16th overall pick – this prompt would seem to cover every big thing that could trip up an AI tool.

Did it work? Let’s go to the tape.

ChatGPT

Here you go: The OG, the Facebook of Microsofts, the one-of-one.

Like all the tools, ChatGPT got off to a rip-roaring start. Here’s what it did with the top picks, along with explanations of each pick.

Tennessee Titans: Cam Ward, QB, Miami (FL): The Titans are in need of a franchise quarterback, and Ward's dual-threat capabilities make him an ideal fit. [SB NATION]
Cleveland Browns: Shedeur Sanders, QB, Colorado: With uncertainties at the quarterback position, the Browns select Sanders, who has demonstrated exceptional leadership and passing skills. [SB NATION]
New York Giants: Travis Hunter, CB/WR, Colorado: The Giants opt for the versatile Hunter to bolster their defense and provide depth at receiver. [SB NATION]
New England Patriots: Abdul Carter, EDGE, Penn State: Looking to enhance their pass rush, the Patriots select Carter, known for his explosiveness off the edge. [SB NATION]
Jacksonville Jaguars: Will Johnson, CB, Michigan: The Jaguars strengthen their secondary with Johnson, a top-tier cornerback prospect. [SB NATION]
Las Vegas Raiders: Ashton Jeanty, RB, Boise State: The Raiders add a dynamic playmaker in Jeanty to boost their running game. [SB NATION]
New York Jets: Kelvin Banks, OT, Texas: To protect their quarterback, the Jets invest in Banks, a standout offensive tackle.
Carolina Panthers: Mason Graham, DL, Michigan: The Panthers bolster their defensive line with Graham, a dominant interior presence.

ChatGPT got the order right, and it’s hard to argue too much over the picks. Yeah, maybe the Jets don’t need to protect their QB when they don’t have a QB, and maybe SB Nation has the massed predictive power of a Magic 8-Ball, but that’s small stuff.

That’s because when ChatGPT is doing the drafting, you can be sure of two things:

It will announce its results with a Trumpian level of hubris; and
Things will go off the rails at some point.

In this case, it only took until the 13th pick, when the Miami Dolphins selected Jered Verse.

You mean Jered Verse, the Defensive Rookie of the Year from the Los Angeles Rams, another NFL team?

The selfsame.

What’s lovely about ChatGPT is when it goes off, it does it like a refinery fire – one tank igniting another and then another until the whole thing blows and the sky turns the color of Texas Pete.

Here are its next picks:

Kool-Aid McKinistry
J.T. Tuimoloau
Dallas Turner
Jordan Addison
Michael Mayer
Peter Skoronski
Kayshon Boutte
Bijan Robinson
Jaxon Smith-Njigba
Benjamin Morrison

With the exception of Tuimoloau and Morrison, all these players are current NFLers – and Addison, Mayer, Skoronski, Smith-Ngijba, Robinson, and Boutte just wrapped up their second seasons. (Not to mention that Boutte went undrafted because he’s roughly as fast as a blocking sled.)

Also, ChatGPT quit after 24 picks. That’s understandable. After nailing the first two dozen, why go any further?

[Mic drop]

Perplexity

Perplexity is the Jeff Bezos-funded AI tool that calls itself “the answer engine.” To become that, it allegedly trained itself on data behind paywalls without asking for permission or paying the owners. Think of it as the little engine that broke into the master’s liquor cabinet and drank up all the Courvoisier.

Since Perplexity has been trained on the good stuff, you might expect better results.

Not exactly.

Like all the tools, things start out promisingly. Check out the first 27 picks:

Tennessee Titans: Cam Ward, QB, Miami
Cleveland Browns: Travis Hunter, WR/CB, Colorado
New York Giants: Shedeur Sanders, QB, Colorado
New England Patriots: Tetairoa McMillan, WR, Arizona
Jacksonville Jaguars: Abdul Carter, EDGE, Penn State
Las Vegas Raiders: Luther Burden III, WR, Missouri
New York Jets: Tyler Booker, G, Alabama
Carolina Panthers: Malaki Starks, S, Georgia
New Orleans Saints: Kelvin Banks Jr., OT, Texas
Chicago Bears: Shemar Stewart, DE, Texas A&M
San Francisco 49ers: Tyler Warren, TE, Penn State
Dallas Cowboys: Jihaad Campbell, LB, Alabama
Miami Dolphins: Armand Membou, OT/G, Missouri
Indianapolis Colts: Nick Emmanwori, S, South Carolina
Atlanta Falcons: Ashton Jeanty, RB, Boise State
Arizona Cardinals: Emeka Egbuka, WR, Ohio State
Cincinnati Bengals: Colston Loveland, TE, Michigan
Seattle Seahawks: Mike Green, OLB, Marshall
Tampa Bay Buccaneers: Benjamin Morrison, CB, Notre Dame
Denver Broncos: Josh Simmons, OT, Ohio State
Pittsburgh Steelers: Josh Conerly Jr., OT, Oregon
Los Angeles Chargers: Shavon Revel Jr., CB, East Carolina
Green Bay Packers: James Pearce Jr., OLB, Tennessee
Minnesota Vikings: Nic Scourton, DE, Texas A&M
Houston Texans: Maxwell Hairston, CB, Kentucky
Los Angeles Rams: Walter Nolen, DT, Ole Miss
Baltimore Ravens: Kenneth Grant, DT, Michigan

Maybe not the best picks, but certainly Kiperesque levels of competence.

However, just as sure as the sun rising in the east, this NFL mock draft heads south with its next five picks, current NFLers all:

Detroit Lions: Jeremiah Trotter Jr., LB, Clemson
Washington Commanders: Kool-Aid McKinstry, CB, Alabama
Buffalo Bills: Keon Coleman, WR, Florida State
Philadelphia Eagles: Maason Smith, DT, LSU
Kansas City Chiefs: Ja'Tavion Sanders, TE, Texas

Things only head further in the direction of Tierra del Fuego from there, with the likes of Kalen King, Chop Robinson and Sedrick Van Pran being drafted with subsequent picks.

Perplexity may be trained on the good stuff, but what it spits out is the same old gruel.

Liner

Liner claims to synthesize the results of multiple engines and give one authoritative result.

Authoritative like Elon Musk on a podium maybe, but not really something you would consider as coming from an authority.

Here is its first round:

Tennessee Titans: Cam Ward, QB, Miami (Fla.)
New York Giants: Abdul Carter, Edge, Penn State
New England Patriots: Shedeur Sanders, QB, Colorado
Jacksonville Jaguars: Will Johnson, CB, Michigan
Las Vegas Raiders: Tetairoa McMillan, WR, Arizona
New York Jets: Mason Graham, DL, Michigan
Carolina Panthers: Jalon Walker, LB/Edge, Georgia
New Orleans Saints: Malaki Starks, S, Georgia
Chicago Bears: Will Campbell, OL, LSU
San Francisco 49ers: James Pearce Jr., Edge, Tennessee
Dallas Cowboys: Luther Burden III, WR, Missouri
Miami Dolphins: Kelvin Banks Jr., OT, Texas
Indianapolis Colts: Tyler Warren, TE, Penn State
Atlanta Falcons: Mykel Williams, Edge, Georgia
Arizona Cardinals: Walter Nolen, DL, Ole Miss
Cincinnati Bengals: Derrick Harmon, DL, Oregon
Seattle Seahawks: Jihaad Campbell, LB, Alabama
Tampa Bay Buccaneers: Emeka Egbuka, WR, Ohio State
Denver Broncos: Ashton Jeanty, RB, Boise State

Again, competent to a Kiperian level … but wait: That’s the entire first round. What about the Packers, Vikings, Bills, Commanders, Lions, et al.?

Good question, because round two proceeds exactly as round one proceeded, only with the top 12 teams participating (and Jalon Walker and Luther Burden III getting drafted again, just because). And round three repeats the pattern, with Burden getting drafted a third time because he’s just that special.

So while Liner gets extra points for only drafting players eligible to be drafted, it gives those points away again by not following draft order and redrafting players multiple times.

Gemini

Saving the best for last? Maybe, if you consider “best” in this instance to be the most hilariously bad.

Gemini was utterly incapable of following any of the instructions, but it couldn’t have been more proud of its picks if it had been Rockin’ Jimmy Irsay announcing his draft class to the massed scribes of Indianapolis Star-Ledger.

Let’s have a look, shall we?

Houston Texans: CJ Stroud, QB, Ohio State
Arizona Cardinals: Will Anderson Jr., EDGE, Alabama
Seattle Seahawks: Jalen Carter, DT, Georgia
Indianapolis Colts: Anthony Richardson, QB, Florida
Detroit Lions (from LA Rams): Caleb Williams, QB, USC
Las Vegas Raiders: Devon Witherspoon, CB, Illinois
Minnesota Vikings: Tyree Wilson, EDGE, Texas Tech
Atlanta Falcons: Bijan Robinson, RB, Texas
Carolina Panthers (from Chicago Bears): Quentin Johnston, WR, TCU
Philadelphia Eagles (from New Orleans Saints): Nolan Smith, EDGE, Georgia
Tennessee Titans: Will Levis, QB, Kentucky
Houston Texans (from Cleveland Browns): Paris Johnson Jr., OT, Ohio State
New York Jets: Jaxon Smith-Njigba, WR, Ohio State
Pittsburgh Steelers: Peter Skoronski, OT, Northwestern
Green Bay Packers: Bryan Bresee, DT, Clemson
Washington Commanders: Christian Gonzalez, CB, Oregon
New England Patriots: Kelee Ringo, CB, Georgia
Detroit Lions: Cam Smith, CB, South Carolina.
Tampa Bay Buccaneers: Myles Murphy, EDGE, Clemson
Seattle Seahawks (from Denver Broncos): Joey Porter Jr., CB, Penn State
Los Angeles Chargers: Lukas Van Ness, EDGE, Iowa
Baltimore Ravens: O'Cyrus Torrence, IOL, Florida
New York Giants: Dalton Kincaid, TE, Utah
Jacksonville Jaguars: Broderick Jones, OT, Georgia
Miami Dolphins: Zion Tupuola-Fetui, EDGE, Washington
Buffalo Bills: Darnell Wright, OT, Tennessee
Dallas Cowboys: Mazi Smith, DT, Michigan
Cincinnati Bengals: Anton Harrison, OT, Oklahoma
New Orleans Saints (from San Francisco 49ers): Jordan Addison, WR, USC.
Philadelphia Eagles: Jalen Carter, DT, Georgia
Kansas City Chiefs: Felix Anudike-Uzomah, EDGE, Kansas State
Detroit Lions (from Los Angeles Rams): Jahmyr Gibbs, RB, Alabama

Such great drafting … of players who have been in the league two years. And while we could give Gemini credit for putting the right player on the right team some of the time, it’s worse at it than most casual football fans, so no.

DeepSeek

We published this post one day before the Chinese AI company DeepSeek threw a stink bomb into the AI world with its announcement that it’s created a faster, better and cheaper AI tool.

While DeepSeek’s claims are alluring – who doesn’t want a faster and cheaper express train to the extinction of humanity? – they’re only valid if A) they’re not an utter fabrication and B) the tool actually works.

Seeing as I have time on my hands now that DeepSeek tanked the value of my Nvidia stock and I have to work another five years, let’s check out how it performs.

And … it doesn’t get off to a rousing start. In fact, the first thing DeepSeek did was disclaim itself: “As of October 2023, I don’t have access to real-time data or the latest mock drafts for the 2025 NFL Draft, nor do I have information on which players have declared for the 2025 draft.”

October 2023? This tells me two important things:

DeepSeek has been building its stink bomb for quite awhile; and
It’s being trained on some old, old data.

Given those two developments, I was expecting lots of picks from two years ago and a totally muddled draft order. What I got was somewhat better.

Now, “better” is a relative term when it comes to AI models doing NFL mock drafts. It still goes about its work with the precision of a mallard duck performing a root canal. However:

It got the number of picks in each round correct; and
It stuck to a draft order, albeit a two-drafts-ago draft order.

And it wasn’t without its peculiarities. See for yourself:

Chicago Bears – QB Carson Beck, Georgia
Arizona Cardinals – EDGE James Pearce Jr., Tennessee
Denver Broncos – QB Quinn Ewers, Texas
Washington Commanders – OT Kelvin Banks, Texas
New England Patriots – WR Luther Burden III, Missouri
New York Giants – QB Shedeur Sanders, Colorado
Las Vegas Raiders – CB Will Johnson, Michigan
Atlanta Falcons – EDGE Nic Scourton, Texas A&M
Chicago Bears (via CAR) – WR Tetairoa McMillan, Arizona
Tennessee Titans – OT Will Campbell, LSU
New York Jets – TE Brock Bowers, Georgia
Minnesota Vikings – QB Jalen Milroe, Alabama
New Orleans Saints – DT Walter Nolen, Ole Miss
Green Bay Packers – S Caleb Downs, Alabama
Pittsburgh Steelers – CB Travis Hunter, Colorado
Seattle Seahawks – DL Mason Graham, Michigan
Los Angeles Chargers – WR Emeka Egbuka, Ohio State
Indianapolis Colts – CB Denzel Burke, Ohio State
Cincinnati Bengals – OT JC Latham, Alabama
Los Angeles Rams – EDGE JT Tuimoloau, Ohio State
Miami Dolphins – OL Cooper Beebe, Kansas State
Philadelphia Eagles – LB Barrett Carter, Clemson
Houston Texans (via CLE) – WR Xavier Worthy, Texas
Dallas Cowboys – RB TreVeyon Henderson, Ohio State
Detroit Lions – CB Kalen King, Penn State
Buffalo Bills – DL Leonard Taylor III, Miami
San Francisco 49ers – OT Amarius Mims, Georgia
Baltimore Ravens – WR Troy Franklin, Oregon
Kansas City Chiefs – WR Malik Nabers, LSU
Jacksonville Jaguars – CB Kool-Aid McKinstry, Alabama
Tampa Bay Buccaneers – EDGE Bralen Trice, Washington
Houston Texans – DL Tyleik Williams, Ohio State

This is the ‘50s-wedding version of a mock draft, with something old (Brock Bowers et al.), something new (Carson Beck to the Bears at No. 1 overall is … novel), something borrowed (the Vikings borrowing Jalen Milroe to worry both Sam Darnold and J.J. McCarthy), and something blue (Shedeur Sanders to the Giants – from DeepSeek’s virtual lips to God’s virtual ears).

The other rounds coat complete ignorance with a veneer of competence, and continue the Chex mix of draft-eligible types and current NFLers.

So what to make of DeepSeek? It performs the practical tests about as well as other AI tools, but no better.

If it’s cheaper, great. That just means it can bring more mediocrity and confusion to the world at scale.

Wha …?

What’s going on here? Training data is available and the prompt clearly lays out what’s to be done, yet the tool doesn’t perform the task to any level of competency.

(And consider this: If you ask these tools to try again, it’s a virtual lock the next iteration will be worse.)

It’s possible a more highly specific and refined prompt could help, but it’s more likely that the tool would eventually get off-track and go rolling blithely through a farm field without once stopping to wonder why corncobs keep getting stuck in the wheels.

All I can do is hazard a guess, and it’s this: A lot of AI tools work by predicting the next thing in a sequence – the next number in a series or the next word in a sentence. The tools don’t always consider the entire sequence before making their prediction; instead, things are reduced to a simple question: What comes after Ashton Jeanty?

In the case of a real NFL mock draft, it would be a draft-eligible player who hasn’t yet been drafted – Shavon Revel, let’s say.

For the AI tool, it could be a player who’s already been drafted – Luther Burden or James Pearce – because in some of the training data their name appears after Ashton Jeanty’s, and to a prediction engine, that’s really all that matters.

It’s nice to know there are still things in the world humans screw up, but not to the same extent as machines. NFL mock drafts are one of those things, and I'm here for it.

I'm also here for the Packers drafting Tyleik Williams in round one. But that’s another story.

The AI Is Still Objectively Bad At Mock Drafts (Now Updated To Include DeepSeek!)