---
title: 'Don''t Use AI for Sports Betting Until You Watch This'
source: 'https://youtube.com/watch?v=ykICNBiFubg'
video_id: 'ykICNBiFubg'
date: 2026-08-26
duration_sec: 1177
channel: 'Circles Off'
---

# Don't Use AI for Sports Betting Until You Watch This

> Source: [Don't Use AI for Sports Betting Until You Watch This](https://youtube.com/watch?v=ykICNBiFubg)

## Summary

Rob Pizzola, a professional bettor, explains why using AI to pick winners is a losing strategy, and instead advocates for using AI as a labor tool to build models, process data, and save time. He illustrates this with a real World Cup bet he made using an AI-assisted expected goals model, which lost but demonstrates the correct process.

### Key Points

- **The Wrong Way to Use AI** [00:02] — Asking AI who will win tonight's games is using the most powerful betting tool in history in the dumbest way possible. AI is an incredible tool but a terrible oracle.
- **Two Wrong Camps** [00:46] — Camp one thinks AI is useless for betting after getting garbage picks. Camp two thinks AI is a magic money printer. The truth is in the middle: AI does the labor, not the thinking.
- **The Market is Smarter Than AI** [01:59] — The betting line is a crowdsourced opinion of every sharp bettor. A general-purpose chatbot is dramatically less sophisticated than the market, which already prices in all available information.
- **AI's Real Role** [03:12] — AI helps at every layer underneath the actual bet: writing code, building datasets, structuring problems, and running math. The newer models are reliable enough for betting math, but sanity checks are still needed.
- **The Correct Division of Work** [04:06] — You bring the thesis and judgment; AI brings the horsepower. The second you flip that and ask AI to bring the thesis, you are 'cooked'.
- **No More Programming Barrier** [04:47] — Two years ago, building a betting model required statistics, coding, and betting market knowledge. Now, AI coding tools can build, test, and revise models from plain English descriptions.
- **Model vs. Edge** [05:31] — Building a model that predicts reality accurately is not the same as building an edge. The market already predicts fairly accurately. If your model agrees with the market, you have no edge.
- **Perceived Edge Trap** [06:24] — If your model says a player scores 7 points but the market says 5, that doesn't mean you have an edge. Your model might just be worse than the market. You don't automatically win the disagreement.
- **Time is the Biggest Win** [07:19] — The single biggest thing AI has given Rob is time. Automating data pulling, cleaning, and workflows saves hours, which is literally money. This time is spent on thinking and thesis generation.
- **AI as a Thinking Partner** [09:01] — AI can be used to work through vague ideas, like a smart colleague. It helps organize thinking, fill gaps, and catch blind spots, but doesn't hand you the answer.
- **World Cup Example: Building the Model** [11:04] — Rob built an expected goals model for the World Cup using AI. The tool did the labor of construction; he brought the thesis and ideas. His model projected Egypt at 1.0 xG and Australia at 0.8 xG.
- **World Cup Example: Finding the Edge** [11:45] — Instead of asking AI who to bet, he fed his projections and all goal-based market odds into the system. AI ran a Poisson distribution and compared probabilities to market prices, identifying the first half under 0.5 goals at +144 as the best edge, underpriced by ~4%.
- **The Bet Lost** [13:10] — The bet lost because there was a goal in the first half at 13 minutes. Rob emphasizes that a losing result doesn't invalidate the process; one bet tells you nothing either way.
- **Garbage In, Garbage Out** [14:33] — AI did not make the model good; it just did the math on whatever was fed to it. If inputs are bad, the output is bad. The quality of the projection still comes down to the bettor's work.
- **Timing and Market Liquidity** [15:14] — This process works better in less liquid markets like openers. At post in a massive efficient market, the number is razor sharp, and timing matters hugely.
- **The Whole Point** [16:21] — AI did not tell him who to bet; it had no opinion. He brought the projection, model, and thesis. AI ran the distribution and market comparison. The judgment came from him; the horsepower from the tool.
- **The Wrong Way Explicitly** [17:00] — Do not ask AI to find bets or predict winners. It is pattern matching language that sounds authoritative, with zero informational edge over the market. If a prompt reliably printed money, everyone would run it and the edge would vanish.
- **The Future of Winning Bettors** [18:37] — Winners will be those who use AI to do more and faster: build more models, test more theses, cover more markets, automate the grind, while still bringing human judgment. AI is an assistant, not a handicapper.

### Conclusion

AI is a powerful assistant that can handle the labor of betting—building models, running math, and saving time—but it cannot replace human judgment or beat the market. The bettors who win will be those who use AI to amplify their own thinking, not those who ask it for picks.

## Transcript

or whatever the most powerful AI model of the week happens to be and you're asking it who's going to win tonight's games, you are using the single most powerful betting tools in history in the single dumbest way possible. And I want
to show you what you should be doing instead because AI really can make you a much better better, just not the way that most people are trying to use it right now. Later on in this video, I'm going to walk you through an actual bet
that I made at the World Cup that I found using AI. Not by asking it who to bet, but by using it as a tool. And how that bet actually turned out makes it a would have been. So, stick around for that.
>> Here we go. >> There are two wrong ways to think about AI and betting and the truth sits somewhere right in the middle. So, you have camp one and these are people who think AI is useless for betting. They
opened up Chat GPT or Claude. They asked it for some picks. It gave them garbage and they wrote the whole thing off. And I understand why they got there, but the other side. These are people who think AI is this magic money printer.
They believe that if they just find the right prompt or the right wording, the perfect question, it's going to start spitting out winners. These people are about to donate a significant amount of money to the
betting market. And now here is the real truth. truth. AI is an incredible tool and it's also a terrible oracle. It will do the work of a data scientist, a coder, a research
assistant. It'll do it for you all at once for basically free. But what it will not do is outthink a market that already has every data scientist, every coder, and every research assistant in the world working against you. And this
is the core thing that I need you to understand. When you ask an AI, "Who wins tonight?" you are asking it to beat the line as it exists at that moment in And think about what that line actually is.
It is essentially a crowdsourced opinion of every sharp better who has fired a bet into the market up until that point. Every professional model, every betting group with real resources, every piece of information anyone has acted on, all
of that is already baked into the number you're looking at. A general-purpose chatbot working off of public data and whatever is in its training set is not more sophisticated than that. It is dramatically less sophisticated than
that. You are showing up to a fight with none of the information the other side none of the information the other side has and asking a language model to make up the difference, and it simply cannot do that. Now, if you're new to this
channel, I'm Rob Pizzola. I bet for a living, and Circles Off is where I break down how I actually think about this stuff. So, if AI is not an oracle, what framework before I get into some specific examples. AI helps you at every
layer underneath the actual bet. It writes code you don't know how to write. It builds and cleans data sets. It helps you structure a problem that you know is there, but you can't quite formalize. It runs math that is correct
something on that last one because it is important. For a long time, AI was really bad at math. If you were doing this a year or two ago, you ran into calculation errors all the time, and some of that still exists if you're not
careful, but the newer models have gotten really, really good at this to the point where for the most part, most of the math you could actually use it for betting. You should still sanity check the
important stuff, though, but the reliability has come a long way. And on top of all that, it speeds up research that used to take you hours. And here is the through-line for all of it. In every one of these cases, you are doing the
one of these cases, you are doing the thinking and AI is doing the labor. That is the correct division of work. You bring the thesis and the judgment. It brings the horsepower. The second you flip that, the second you ask it to
bring the thesis and the judgment, you are cooked. If this framing is landing for you, do me a favor really quickly, hit that like button down below. It really does help push the video to other people who want
level. I'd really appreciate that. Now, let me show you what this looks like in practice. The most accessible use case and the one that has changed the most is that you no longer need to be a programmer to build things.
Two years ago, if you wanted to build a betting model, you needed to know statistics and coding and betting markets. And if you didn't already have those skills, you were looking at months, if not years, of learning before
you could even start. Gaining a real foundation in all three of those things at once is really hard, and it's not something that most people can do quickly. That barrier is largely gone now. You can describe what you want in
plain English, and an AI coding tool will build it, test it, find its own errors, and even revise it. But here's the part that trips people up. Being able to build a model is not the same thing as building an edge.
AI will very happily build you a model that predicts reality accurately, but nothing to do with the book specifically. It is the market. The
market, meaning all the other people betting into it, already predicts fairly accurately. Remember, you're not competing against a sports book, you're competing against every other person that's putting money into that market.
So, think about what that means for your model. If your model says a player takes three shots a game and the market also thinks he takes three shots a game, well, you don't have an edge. You just have a model. Those are two very
different things. And here's the part that really gets some people sometimes. Let's say your model has a player scoring seven points, but the market is only at five. It's really tempting to look at that and go,
"Great, huge edge. The market is way off here." But that does not mean you have an edge either. What you have is a perceived edge based on your model. And your model might actually just be worse than the
market. The market might be at five because five is closer to the truth and your model is the thing that's actually wrong. You do not automatically win the disagreement just because it's your number. So, the tool removes the
friction of building. It does not hand you the edge. The edge still has to come from your thesis, from finding something that the market is not accounting for or something that you can price better than the market can. That's the proprietary
part. AI is not going to give you that. It can only build the machine once you've decided what the machine is for. Now, the next one does not get talked about enough. And honestly, it might be the biggest one for me personally. I use
AI basically non-stop every single day. And the single biggest thing it has given me is not some secret edge. It's time. And I want to make a point about time that I think people really undervalue.
Time is money in a very literal sense. Everyone who works gets paid a certain amount, whether it's salary or an hourly rate, but it all breaks down to some amount of money per hour of your life. So, when a tool gives you hours back,
it's not just convenient, it has real quantifiable value. It is the same as being a handed money. I have automated an enormous amount of my process. Data that used to require me to sit at a desk and manually pull and
to sit at a desk and manually pull and clean and organize now runs on its own. Workflows that used to eat up hours now happen in the background without me touching them, and the value of that is very hard to overstate. I can set
something up, trust that it's going to run seamlessly and correctly, and then go play a round of golf instead of grinding at a desk all day babysitting For a professional better, that is massive. Your time is your most limited
resource. Every hour you spend doing manual labor that a tool could do is an hour you're not spending on the actual thinking, on the thesis generation, the stuff that actually produces edges. AI taking the grunt work off of your
plate is not a small quality of life improvement. It fundamentally changes able to do. And you can talk through problems with it as well, like a thinking partner basically. When I have a vague sense
that there is an edge somewhere, but I can't quite formalize it, I will work through it with AI the same way I would with a smart colleague. What variables matter here? How would I test this? What am I not thinking of?
It does not hand me the answer, but it helps me organize my own thinking and fill in or even catch the gaps. Now, real quick, I want to talk about who's making this video possible, and it is a company I actually believe in, which
makes this super easy. This video is brought to you by Profit X. Now, everything we're talking about today comes back to one idea, which is finding the market's number. And when you find one of those spots,
where you actually get your money down matters just as much as the edge you found, because price is everything when you're betting on sports. Here's the thing about Profit X that I think a lot of people don't fully appreciate. When
you bet at a traditional sports book, you're betting against the house, and the house builds its margin into every single line that you see. Profit X is different. It's a prediction market. You are trading against other
people, not against a house that is taking a cut right off of the top. It is federally regulated, it is CFTC backed, and the price is set by the market itself, which more often than not means a better price on the exact same trade.
Profit X. I really like what they're building, and for a better better pricing on the same bet, it's not just a small thing. Over a large enough sample, not check them out yet, I'd really encourage you to give them a try. Sign
up to Profit X, use promo code circlesoff, all one word. You can trade $10 and get 20. The link for that is down in the description. Okay, now let me show you the example I promised you at the start, because I think it ties
all of this together in some way. This is the bet I mentioned in the opening, because it's the clearest illustration of everything that I've been saying so far. So, at the World Cup,
I built out a basic expected goals model. And here's the fun part. I actually built that model using AI wherever I could. So, the tool did not just help me place the bet, it helped me build the thing that generates the bet
in the first place. AI did the labor of construction. I brought the thesis, I brought the ideas of what to build and why to build it. Now, for a particular match, my model projected Egypt at 1.0 expected goals
and Australia at roughly 0.8 expected goals. Those are my numbers, my model. That's the part that is mine, even though it was AI-assisted. Now, here's
where AI came back in. I did not ask it who to bet. Instead, I fed my final projections and all of the odds into the system, every single market associated with that game that was based on modeling goals. Every total, every first
half line, team totals, all of it. And I asked it to do the work that would have taken me forever by hand. I said, "Given these full match expected goal numbers, work out approximately how much of that scoring is expected to happen in the
run a Poisson distribution across the range of possible outcomes, and then tell me where my numbers diverge the most from the actual market prices." And quickly, for anyone who does not know, a Poisson distribution is just a
way of taking an average, like expected goals, and turning it into the probability of specific outcomes. So, the probability of zero goals, the probability of one goal, two goals, and so on. It is exactly the kind of math
that is correct, but can also be tedious. So, it's the perfect thing to hand to a tool. So, AI took my projections, worked out the first half portion, ran the distribution, and compared those
probabilities against every one of those goal-based markets that I fed it. And after going through all of them, the spot that stood out the most as the best edge was the first half under half goals at plus 144. Based on my projection,
that number was underpriced by roughly 4%. So, that is the bet that I made. And it lost. There was a goal in the first half, 13 minutes in. It did not cash. Very frustrating because there were legitimately no good chances in
that first half. But, such is life. And does not matter for what I'm trying to show you here. I could have dug up a bet from the World Cup that won and use that instead, but it would have made the
exact same point and might have left you with the wrong idea that a winning result means more and that the process was not good. Doesn't work like that. One bet tells you nothing either way. The process is
what we are looking at and the process is identical whether this particular bet won or it lost. And I want to be really careful here as well cuz I don't want anyone walking away thinking, "Oh, Rob built a model, you know, pointed an AI
at it and now all of a sudden he can beat soccer." That That's not what I'm saying at all. In fact, it's the exact opposite of what I'm saying. Let me get ahead of something here cuz I I know a lot of you watch this content
already be typing it down in the chat below. Somebody's going to say, "Rob, that model might not even be good enough to beat soccer markets, especially when you consider the limits in place on some of these bets."
And they might be completely right. This is a basic expected goals model. It's very possible it's not sharp enough to beat those markets over the long run. And here is the bigger point buried in all of that. The AI did not make my
models good. The AI just did the math on whatever I fed it. If my inputs are bad, whatever I fed it. If my inputs are bad, the output is bad. Garbage in, garbage out. The quality of that projection still comes down to me and the work I
put into the model. The AI cannot save a bad number. It just processes the number I give it faster than I could. Now, the timing of the bets still matters, too. This kind of process, taking your projection, scanning a bunch of markets
for the for the biggest gap or the biggest edges, that might hold up when you're betting into, let's say, less liquid markets, something like openers, price might be softer. It's a very different story if you're trying to do
this right at post in a massive efficient market where the number is already razor sharp. Where you bet and when you bet is still a huge part of whether any of this actually works. So, no, I'm not telling you that I cracked
soccer. I'm just showing you a clean example of a repeatable process. When I take a stab at a new sport, I want that process to be as technically sound as I can make it. And I use AI to help me get there faster. Maybe this particular
model beats the market, maybe it doesn't. That's not the point. The point is the method. And for the record, I absolutely factor in limits and market size when I decide what to actually bet, which is something I went deep on in the
last video here on this channel. If you've not seen that one, go check it out. Link is in the description below because it pairs really well with this. happened there because this is the entire point of the video. AI did not
tell me who to bet. It did not have an opinion on the game. I brought the projection, I brought the model, I brought the thesis. AI did all the labor for me. It ran the distribution, it did the market
comparison across dozens of lines faster and more accurately than I ever could by hand. The judgment came from me. The horsepower came from the tool. And whether the bet won or lost has nothing to do with whether the process was
correct. It lost, and I'd run that exact same process again tomorrow. That's the whole relationship right there in one bet. So, now that you've seen the right way to do it, let me be really direct and explicit about the wrong way.
Do not ask AI to find you bets in tonight's games. Do not ask it who's going to win this game. Do not trust it when it gives you match-up. When it answers those questions, it is
not doing sophisticated analysis. It is pattern matching language that sounds authoritative and it has zero informational edge over a market that is already pricing in everything the AI knows and enormously more that it does
not know. And here's the logical tell that should settle this for you forever. If you could actually pick winners just by asking an AI the right way, that edge would be gone instantly because everyone has access to the same tools. The moment
a prompt reliably printed money, 10,000 people would run it. The market would adjust and the edge would just vanish. The only edge that survives is the one nobody else can copy. Your proprietary thesis, your unique data, your judgment
and then applying the tool to that. The AI is not the edge. The AI is what you yourself. I really want to know how you are using AI in your betting and I do not mean, you know, I asked it for picks kind of
stuff. I mean the real workflows. What have you built? What have you automated? What has actually saved you time or helped you find an edge? Drop it down in the comments below. I read all of these every single week. If there's some good
ones, I might actually feature them in a follow-up video here on Circles Off as home. The bettors who win over the next few years are not going to be the ones who found the magic prompt. There is no
magic prompt. They're going to be the ones who use these tools to do more and to do it faster. Build more models. Test more theses. Cover more markets. Automate more of the grind. All while still bringing the human judgment that
the tools do not have. AI is the best assistant you've ever had. It is not a handicapper. Use it like an assistant and you're going to be dangerous at betting. Ask it to be your handicapper and you're going to be broke.
These tools expand what you're capable of, but you still have to drive the car A while back, I had Kirk Evans in studio. We let ChatGPT pick 10 sports betting questions for us to answer on the spot. The back half of that
conversation goes deep on AI and betting. It's a great next watch if you want to keep going down this road here on Circles Off. Click down here for that on Circles Off. Click down here for that one.