AI Sports Picks

What an AI sports pick actually is

An AI sports pick is a bet a model recommends because it thinks the sportsbook has priced something wrong.

That is the whole idea, and it is narrower than it sounds. The model does not try to guess who wins. It produces a probability for one specific market — this side covers, this total goes over — and turns that probability into its own fair price for the line on the board. Then it compares that price against what the book is offering. When our number and the market's number agree, there is no bet. The market is right most of the time.

A pick only exists in the gap between the two.

That is what separates this from a tipster with a hot take. A tout tells you who they like. A model tells you where the price is wrong, and how wrong, and then publishes what happened either way.

How the model prices a game

Every night, while the lines are still soft, a pipeline runs.

It ingests the slate. Who is playing, how they have been playing, the conditions they are playing in, and how the market is pricing it across multiple sportsbooks.

It prices every market. Each matchup is priced by a machine-learning model trained on more than a decade of completed games. It is trained separately for every league and every market we publish — the model that prices an NHL moneyline is not the model that prices an MLS total. It is validated only on games it has never seen, in chronological order, so a model is never scored on a season it was fitted to. It is retrained on a fixed weekly schedule.

It compares. Our price against the market price, with the book's margin still in it. Stripping that margin out first would make every edge look bigger than it is. We leave it in, so the bar a pick has to clear is the harder one.

It publishes, then grades. Picks that clear the threshold go to the public ledger with their confidence, their edge and their unit size, timestamped before the games start. After the final whistle each one is graded. Wins, losses, pushes, everything.

The full version of this, stage by stage, is on the how it works page. The claims behind it are on the methodology page.

What the model reads

The inputs are the things that measurably move a result: recent form, matchup context, injuries and lineup availability, rest and travel, weather where the game is played outdoors, and match officials. Market prices come from a commercial odds feed covering multiple sportsbooks. Results come from a commercial sports data feed.

What it does not read is just as important. No news wires. No social sentiment. No reported betting volume. If a story breaks after a pick is published, the model has not seen it.

Why most games produce no pick

Not every prediction becomes a pick.

Each one is checked against a minimum edge threshold configured for that market. Clear it and the pick is published. Fall short and it is held, and never shown to anyone. We also only consider a fixed list of leagues and a fixed list of markets — anything outside them is dropped rather than priced.

Most predictions do not become published picks. That is why a day's card is short, and why a service promising twenty locks a day is telling you something about itself.

Every pick is published. Including the bad ones.

A record that only shows winners is not a record.

Every pick that clears the threshold is written to a public ledger and graded when the game settles. Nothing is edited after the fact. Nothing is deleted. The losing weeks are in there, at the same size and in the same place as the winning ones, because a track record you can only see the good half of tells you nothing about the half you are betting on.

You can read the whole thing on the track record page before you give us an email address.

What the model gets wrong

Any picks service that will not tell you its failure modes is hiding them. Ours:

Late news. The model prices a game before the slate and does not read the news wire. A star scratched hours later can invalidate the math, and nothing downstream re-prices it. If the number has moved a long way from the price we published, treat the edge that justified the pick as gone.

Thin data. Season openers, rookie debuts, post-trade rosters. When the history is thin, the model is noisier, and it does not always know that it is.

Variance. This is the big one. A model with a real edge still loses plenty of individual bets, sometimes several in a row, sometimes for weeks. Nothing about a positive expected edge protects you from a bad month. If you are not prepared to sit through one, no model is going to help you.

Which sports

The model currently prices NFL, MLB, NBA, NHL, MLS and the World Cup.

Coverage tracks what has a trained model behind it, not what would be convenient to advertise. If a league is not on that list, we do not publish picks for it, and we do not pretend to.

How the stake is sized

Stake is expressed in units, not currency. A unit is one standard bet size — for a $20 bettor a unit is $20, for a $500 bettor it is $500 — so the same record means the same thing at any bankroll.

Unit size scales with both the size of the edge and the price on offer. A bigger mispricing is not automatically a bigger bet, and a large edge at a poor price can be sized below a smaller edge at a good one. What a unit is, in full.

Free AI picks, and what the paid card adds

One AI pick goes out free every morning, by email, with its confidence score. The full public track record is free too, and always will be — it is the only part of this that is worth anything before you have tested it.

Pro adds the rest of the card, typically six to twelve picks a day across every league the model covers, with the edge percentage and recommended unit size on each one, plus line-movement alerts before tip-off. It starts with a seven-day free trial.

There is also a guarantee that is tied to the ledger rather than to a claims process: if our full published card finishes a calendar month net-negative in units, your next month is credited automatically.

Where to start

Take the free pick for a couple of weeks and grade us yourself against the public ledger. That is the honest way to evaluate any model, and it costs you nothing but an email address.

Then read the methodology before you decide we are worth paying for. If a picks service cannot explain what its model reads, what it ignores, and where it breaks, the number on its homepage is decoration.

These are model outputs, not guarantees, and not financial advice. A positive expected edge does not mean an individual pick wins, and any betting strategy can lose over any finite sample. If betting has stopped being entertainment for you, our responsible gambling page lists free confidential help.