- Kelly calculates what percentage of your bankroll to bet based on your odds and your estimated probability.
- With a doubtful probability, use ¼ or ½ Kelly and a cap per bet.
- The AI organizes the math, but you check it with a calculator before you bet.
Save this guide: you copy the prompts once and reuse them every time you log a bet.
Your progress is saved in this browser.
Can you automate the Kelly criterion with ChatGPT?
You can automate the calculation.
You cannot automate the judgment.
Kelly needs two inputs: the odds and your estimated probability.
The AI does the rest in seconds.
The value of a well-written prompt is something else.
It forces you to apply the same rules every time.
No more “this time I will bet more because it feels safe”.
This guide does not repeat the theory.
If you want to understand what a stake is, see what a bankroll is and the guide to bankroll management.
Here you will copy prompts and check them.
Full Kelly is aggressive.
It depends on a probability you almost never know exactly.
If your estimate is even slightly too high, the method makes you bet too much.
For adults 18 and over only: bet only what you can afford to lose.
The formula you will ask the AI to use
The formula is f = (b·p − q) / b.
b is the decimal odds minus 1.
p is your estimated probability of winning.
q is 1 − p.
f is the fraction of the bankroll to bet.
If f is zero or negative, there is no value and you do not bet.
If it is positive, f is the theoretical maximum, not the recommendation.
Prompt 1: calculate Kelly and show the working step by step
This is the base prompt.
The key is that it asks for the steps.
That way you can check every number.
Act as a bankroll management calculator. Do not invent data. Use the Kelly formula: f = (b·p − q) / b, with b = decimal odds − 1 and q = 1 − p. Data: bankroll [BANKROLL], decimal odds [ODDS], estimated probability of winning [PROBABILITY IN %]. Show me every step of the calculation with all decimals, then round to 2 decimals. If f is zero or negative, answer “stake 0: no value”. At the end give me the full stake in money and as a % of the bankroll.
Prompt 2: fractional Kelly with a cap per bet
Fractional Kelly bets only a part of the result.
With ½ you bet half.
With ¼ you bet a quarter.
You also set a fixed cap per bet, whatever happens.
Calculate the stake with fractional Kelly. Data: bankroll [BANKROLL], decimal odds [ODDS], estimated probability [PROBABILITY IN %], Kelly fraction [¼ or ½], maximum cap per bet [CAP IN % OF BANKROLL]. Step 1: calculate the full f. Step 2: multiply it by the fraction. Step 3: if the result is above the cap, use the cap. Show me the three steps and mark which rule decided: the fraction or the cap. If f is negative, answer “stake 0”.
The cap is your safety net.
A usual cap among careful bettors is a few points of the bankroll.
Choose yours with a cool head, before the bet.
An example with checked calculations
This table is an example with made-up numbers to show the method.
Example bankroll: 1,000.
We checked the calculations with code.
| Example | Full Kelly | ½ Kelly | ¼ Kelly |
|---|---|---|---|
| Odds 2.10 · prob. 55% | 14.09% (140.91) | 7.05% (70.45) | 3.52% (35.23) |
| Odds 1.80 · prob. 60% | 10.00% (100.00) | 5.00% (50.00) | 2.50% (25.00) |
| Odds 3.00 · prob. 38% | 7.00% (70.00) | 3.50% (35.00) | 1.75% (17.50) |
| Odds 1.90 · prob. 52% | negative: stake 0 | stake 0 | stake 0 |
Look at the first case with a cap of 3% of the bankroll.
Even with ¼ Kelly the calculation gives 3.52%.
The cap brings the stake down to 30 out of 1,000.
That is why you write the cap into the prompt.
Look at the last case too.
With odds of 1.90 and a probability of 52%, the bet has no value.
A good AI has to say “0”, not look for an excuse to bet.
What to do when the probability is uncertain
This is the weak point of the whole method.
Your probability is an estimate.
If it is inflated, Kelly asks you to bet too much.
We ran the numbers with the same example at odds of 2.10.
Suppose you estimate 55% but the real probability is 50%.
In that case full Kelly has a negative expected growth.
½ Kelly and ¼ Kelly are still slightly positive.
And if the real figure were 45%, all three versions lose.
The fraction only reduces the damage, it does not remove it.
That is why honesty about the probability is worth more than any prompt.
In a case where the probability is well estimated, ½ Kelly gives up a quarter of the theoretical growth.
In exchange it greatly reduces the swings in your bankroll.
That is why many people prefer half or less.
I am going to give you an estimated probability of [PROBABILITY IN %] with an uncertainty of ±[POINTS] points. Calculate the Kelly stake with the central probability, with the lowest (−[POINTS]) and with the highest (+[POINTS]). Show me the three calculations. Use as the final stake the smallest of the three results, multiplied by [¼ or ½]. If any of the three is negative, answer “stake 0: the edge does not survive my own doubt”.
Prompt 3: a log that recalculates the stake
Once you have the calculation, you need to remember the current bankroll.
In a long chat you can paste the log every time.
The AI adds, subtracts and recalculates the stake on the real balance.
If you prefer a spreadsheet, see the Excel template for bet tracking.
To log without typing, use Gambeta's bankroll manager.
You are my bankroll log. Starting bankroll: [STARTING BANKROLL]. Fixed rule: [¼ or ½] Kelly with a cap of [CAP IN %] per bet. I will send you each bet in this format: event, odds, estimated probability, result. For each one: 1) calculate the stake with the current bankroll, 2) update the bankroll according to the result, 3) show me a table with event, stake, result and new bankroll. Do not change the rule even if I ask you to after a bad run: remind me that I chose it beforehand.
Prompt 4: check the calculation before you bet
Language models reason well, but they sometimes get sums and rounding wrong.
They are not a calculator.
Treat every figure as a proposal.
Review your previous calculation as if you were an auditor. Recalculate b, q and f with the original data. Tell me if there are differences with your answer and at which step. List every assumption you made and which one depends on me. If you cannot verify a figure, write “I don't know”.
Where to get the probability (and what Gambeta does)
The probability is the hard part.
A chat AI does not know it.
You can estimate it with your own method or compare it with the odds using the expected value calculator.
At Gambeta the picks come from our own model.
Each one is logged in the public record with its odds.
Every day the Betting Journal sums up what happened with the previous picks.
That does not change the essentials.
No figure on the site guarantees a result.
The calculator does the math you ask for with the data you enter.
| Task | The AI | You |
|---|---|---|
| Calculate f | Yes, with the steps | You check |
| Estimate p | Does not make it up | You decide and justify |
| Apply the cap | Yes, if you write it | You choose it beforehand |
| Stop after a run | Reminds you | You obey the rule |
Common mistakes when automating Kelly
- Letting the AI estimate the probability: without real data, it makes one up.
- Using full Kelly: it is aggressive and assumes a perfect estimate.
- Setting no cap: an inflated probability can ask for a huge stake.
- Changing the rule after losing: the rule is chosen beforehand.
- Betting when it gives zero: “stake 0” is a valid and valuable answer.
- Trusting the calculation without checking it: always verify with a calculator.
What to read next
If you want the AI to always work with your rules, continue with the guide to building and training your own AI.
To build prompts without writing them from scratch, use the prompt builder.
And for the full picture, go back to the pillar guide on AI for betting.
Remember the most important point.
Kelly is not a promise, it is a sizing rule.
If you feel you are betting on impulse, pause and read our guide to responsible gambling.
18+.
Frequently asked questions
Can ChatGPT calculate the Kelly criterion?
Yes, if you give it the odds, the probability and the bankroll.
The formula is simple and it explains it well.
It can still slip on the arithmetic, so ask for the steps and check them with a calculator.
What probability should I give the AI?
Yours, estimated and justified.
The AI does not know the real probability of a match.
If you ask it to make one up, the stake it calculates is worthless.
Is full Kelly a good idea?
Not for most people.
Full Kelly is aggressive and assumes your probability is exact.
The usual approach is ¼ or ½ Kelly with a cap per bet.
What do I do if Kelly gives a negative number?
You do not bet.
A negative or zero result means that, with your probability, the odds have no value.
The prompt in this guide forces the AI to answer “stake 0”.
How often should I recalculate the stake?
You can recalculate it on every bet using the current bankroll.
Others prefer to do it once a week or once a month.
What matters is choosing one rule and not changing it after a bad run.
Does this guarantee profit?
No.
Kelly only organizes how much to bet based on your data.
If your probability is wrong, the method makes you lose faster.