I Used AI to bet the NBA Draft

I’m not a historical NBA‑draft bettor — certainly not one with an edge.

After a very profitable playoff run I gave myself permission to take a risk yesterday by experimenting with a sprinkle of draft wagering.

A related question I wanted to answer was: would an AI tool be valuable in handicapping the draft and choosing bets to make?!

As the Robin Hood of sports betting, taking from the books and giving to the bettors, I’m excited to share the process I used to bet the draft using AI and (spoiler alert) lose some money.

Win or lose, the exercise showed where AI adds real value and how to best leverage these tools.

Today I’ll walk through the exact steps, the results, and the lessons so you can decide if—and how—you use an AI model on the next niche market.

As always, we’ll close with our bets of the week.

Let’s get after it.

My Approach

1. The Tool

I used the premium version of ChatGPT with the “advanced reasoning” setting. You’ll get responses a bit slower than you would otherwise, but I think this is an important step for this type of analysis.

You could do the same with Anthropic Claude, Perplexity Pro, or Google Gemini—any model that can read a prompt, fetch public data, and do reasonable math.

2. The Betting Hypothesis

I was browsing draft betting lines to try and find an edge, and saw that DraftKings listed Tre Johnson to be picked in the Top 5 at –330 (about a 77 % implied chance of happening).

At the same time, I noticed an exchange had the same bet priced at +110 (47.6 % implied).

That gap looked like easy value! If you can bet something that’ll happen 77% of the time at a price that implies it’ll happen less than half the time — run, don’t walk!

When an opportunity feels too good to be true, however, it’s worth doing some quick research to understand the market’s informed opinion.

3. What the Model Returned

ChatGPT gathered:

• Sportsbook odds from DraftKings, FanDuel, BetMGM, and one offshore.

• Five mock drafts (ESPN, The Athletic, Yahoo, CBS, Bleacher Report).

• Team‑fit notes on Utah at pick 5.

Here’s an example of the sportsbook research:

Using a weight of—45 % sportsbooks, 40 % Tier‑1 mocks (ESPN, The Athletic, Yahoo), 10 % Tier‑2 mocks, 5 % team notes—the model produced a 63 % fair probability.

This confirmed the exchange price (+110) was about 13 cents of edge per dollar. Seemingly very solid!

Not wanting to miss the opportunity, at 2 p.m. ET I bet 1.5 units.

4. Monitoring Line Movement

To stay current through the day I saved and deployed a second prompt:

Every two hours I pasted Prompt #2. The model refreshed in under 30 seconds—much faster than pulling every input by hand.

What Happened?!

1. The ESPN Mock Draft Shift

At 3 p.m. ET ESPN released a live mock showing Tre Johnson at pick 8 and Jeremiah Fears at pick 5.

Within minutes FanDuel flipped Tre from +110 to +100; DraftKings paused the market.

Running Prompt #2 above again gave me the bad news:

• Tre Top‑5 probability fell from 63% to 52 %.

• Fears Top-5 probability rose to 57 %.

Here’s the update:

With much of my edge eroded I traded out of 25 % of the Tre position at that point but did not exit.

I considered hedging with Fears to be exactly Pick 5 at +175 and ultimately passed.

2. Draft‑Night Reality

The exchange prices drifted down through the afternoon and early evening.

I resisted the temptation to add to my position as I presumed the markets might be more current/knowledgeable than my earlier cursory research.

So what happened?! In actuality, Utah surprised the market and selected Ace Bailey at pick 5. OOF!

Tre Johnson went pick 6; Jeremiah Fears went pick 7.

My Tre bet lost, and I dodged a bullet by not using Fears as a partial hedge.

Reflections for Next Year/Next Draft

Here are my takeaways from this activity:

1. AI speeds up the grunt work. Gathering five books and five mocks (not to mention team context) manually would have taken an hour. The model did it in under a minute and removed vig automatically to provide me useful pricing estimates.

2. Mock drafts are not created equal. ESPN seemingly moves markets because it reflects front‑office intel. Inside the final six hours I should have raised ESPN’s weight from 40 % of the mock block to at least 60 % of the entire blend.

3. Prediction markets may have more current data than sportsbooks. Because exchanges allow betting on both “YES” and “NO”, they may be a more trusted proxy for fair pricing than sportsbooks. Books can protect themselves by only offering “YES” bets at exorbitant prices.

4. Limit your unit size if you’re hobbying in a new market. Betting more than one unit on an unhedged edge that was (apparently) illusory is more fun than responsible!

It was valuable to play around with using AI to handicap unique markets. I’ll be back next week to share another such attempt that I am optimistic about!

Bet of the Week $$

Last week we closed out the NBA season with another win (!!), as the Pacers cruised in Game 6 before succumbing in 7.

We won one full unit boost our running total in this section to +15.4 units, at a positive 19% ROI. We’ll update this regularly.

Based on my research, I am making the following bets this week:

• DRAW (Salzburg vs Real Madrid) @+550 on DraftKings for 0.5 units

Real Madrid can guarantee a spot in the final 16 with a draw, and because of this I like the value on a tie at +550 or better.

I’d add to this bet live in a tie game late in the second half!

Next week we’ll use AI to handicap and bet on Nathan’s Hot Dog Eating Contest – don’t miss it!