
- First you check the file: if the totals do not match your spreadsheet, nothing that follows is any use.
- Results by market mislead with few bets: always ask for the count next to every number.
- The audit ends in a single rule to change, measured over your next 50 bets.
Save this article: a full audit takes more than one sitting, so come back, copy the next prompt and carry on where you left off.
Your progress is saved in this browser.
Which prompts work for analyzing your betting history?
Your history is the only betting data that is one hundred percent yours.
The AI does not have to look for it on the internet and cannot make it up: you give it.
That is why it is one of the tasks where a chat performs best.
The risk is a different one: that it misreads the file and draws tidy conclusions from wrong numbers.
That is why this article splits the audit into five prompts.
The first one does not analyze anything: it checks that the file was read correctly.
All the prompts use parameters in [BRACKETS].
Replace each one with your own data before you paste.
They work in any AI that accepts pasted text or a file.
Golden rule: no conclusion is worth anything if the AI’s totals do not match those of your spreadsheet.
In every prompt we ask for the rows that prove each claim.
If something does not come from the file, the right answer is “I don’t know”.
If you do not keep a record yet, start with the Excel template to track your bets.
If you want a fixed workspace, see how to bet with Claude or how to bet with Copilot in Excel.
This article is something else: audit prompts, to paste wherever you like.

Before you paste: three rules so the AI reads your record well
- One bet per row: with clear headers and no merged cells.
- A single number format: decimal odds and the same separator throughout the file.
- No personal data: delete your name, email, account and payment methods before uploading it.
Uploading files and running calculations on them depends on each AI.
Features change: check in the app what your plan includes.
If your AI does not accept files, copy the rows and paste them as text.
Prompt 1: cleaning and checking the file
This is the starter prompt and the most important one.
The AI tells you what it read, and you compare it with what you know.
If it fails here, do not go on.
This is my betting record: [PASTE OR UPLOAD YOUR RECORD]. Do not analyze it yet. First tell me: 1) which columns you found and what you understand each one to mean, 2) how many rows there are and what the first and last dates are, 3) which rows have strange data: odds below 1.01, an empty or negative stake, a result not filled in, repeated dates or different formats, and 4) the totals: number of bets, total stake and net result. Show the row number of each problem. Do not correct or delete anything on your own, and if you do not understand a column write “I don’t know”.
What it returns
An inventory of the file and three totals.
Also a list of suspicious rows, with their number.
It is the quality check before any analysis.
How to verify it
Compare the three totals with those of your spreadsheet.
If they do not match, the AI misread a column or cut the file short.
Fix the odd rows yourself, in your spreadsheet, and upload it again.
A long file can arrive cut off.
If the number of rows does not match, split it into parts or paste fewer months.
Never accept an analysis done on half of your history.
Prompt 2: results by market, by odds range and by day
With the file checked, the breakdown arrives.
You are looking for where you win and where you lose.
The trap is believing a group with few bets.
With the record we already checked, build three tables: by market, by odds range (up to 1.50, from 1.51 to 2.00, from 2.01 to 3.00 and over 3.00) and by day of the week. In each row put: number of bets, total stake, net result and yield, calculated as net result divided by total stake. Mark with the words “small sample” every row with fewer than [MINIMUM NUMBER OF BETS] bets. Do not draw conclusions from the marked rows. At the end check that the sum of each table matches the totals from the first check. If it does not match, say so and do not go on.
What it returns
Three tables in the same format.
Each number comes with its number of bets next to it.
That way you see whether a “good” market is twelve lucky bets.
| Market (example) | Bets | Total stake | Net result | Yield |
|---|---|---|---|---|
| 1X2 result | 120 | 120 units | +6.0 units | +5.0% |
| Goals over/under | 60 | 60 units | −4.5 units | −7.5% |
| Corners | 15 | 15 units | +3.0 units | +20.0% (small sample) |
| Total | 195 | 195 units | +4.5 units | +2.3% |
The numbers in the table are an example to show the calculation.
The best yield is the corners one, and it is exactly the one that is worth the least: it comes from 15 bets.
One winner more or less changes it completely.
How to verify it
Add up a small row by hand and check the yield.
The sum of bets in each table has to match the total from prompt 1.
To know what count is enough, read how many bets you need to know whether you are winning.
Prompt 3: discipline errors
This prompt does not look at whether you won: it looks at whether you followed your rules.
Often the problem is not the prediction, it is the stake.
For it to work, your rules have to be written down.
My rules are: maximum stake of [MAXIMUM STAKE] per bet, a maximum of [BETS PER DAY] bets per day and allowed markets [MARKETS]. Review the record and list, with row numbers: 1) bets with a stake above the maximum, 2) days with more bets than the maximum, 3) bets placed after a lost bet on the same day and with a higher stake than the previous one, 4) bets outside my markets and 5) bets placed between [FROM TIME] and [TO TIME]. For each group tell me how many there are and what their net result was. If the record has no time, say so instead of assuming one. Do not judge me or give me advice yet.
What it returns
Five lists of rows, one for each type of breach.
The third is the one that hurts the most: raising the stake after losing.
Each list shows what that group of bets cost you or gave you.
How to verify it
Open three rows from each list in your spreadsheet and check that the breach exists.
If your record has no time, points 3 and 5 cannot be measured well.
The reason behind these mistakes is in the psychology of the bettor, and one way to stop them is the spreadsheet with limits.

Prompt 4: how much of your result can be chance
A positive result does not prove you have an edge.
A negative one does not prove the opposite either.
This prompt puts your number next to what luck alone would produce.
With my record, calculate: number of bets, hit rate, average odds, yield, the longest streak of consecutive lost bets and the longest of won ones. Then answer step by step: 1) what hit rate I need to break even with my average odds, 2) whether a bettor with no edge at all, with my number of bets and my average odds, could have my result by chance alone, and 3) whether my longest losing streak is normal for my hit rate. Show the calculations and state which assumptions you use. Do not say that I have or do not have an edge: tell me how many more bets would be needed to know better.
What it returns
Your basic numbers and an honest reading of chance.
Example calculation: with a 50% hit rate, losing five in a row starting from a given bet has a probability of 3.1%.
In a history of hundreds of bets, the strange thing would be never to see it.
How to verify it
The percentage to break even is 1 divided by your average odds: check it with the calculator.
Find the longest streak in your spreadsheet and count it by hand.
The references are in losing streaks: how many are normal.
The chance part is an estimate with simple assumptions.
If you bet at very different odds, the real calculation is more complex.
Take it as an order of magnitude, not as a verdict.
Prompt 5: one rule to change and how to measure it
This is the closing prompt.
An audit that ends in ten resolutions changes nothing.
This one ends in a single rule and a way to measure it.
This is the summary of my audit: [PASTE THE CONCLUSIONS FROM PROMPTS 2, 3 AND 4]. Propose three possible rules, each one written so that it can be followed or broken without argument. For each one tell me which piece of data in the record justifies it and whether that data was a small sample. Then help me choose a single one and define how I measure it over my next 50 bets: which column I add, which number I look at and which result would make me keep it or drop it. Do not propose rules that promise winnings. If none of them is supported by my data, say so.
What it returns
Three candidate rules and a measurement plan for one.
A well-written rule is “maximum stake of 1 unit”, not “bet more carefully”.
The measurement is how many times you followed it, before how much you won.
How to verify it
Make sure the chosen rule comes from a piece of data with a large enough sample.
Add the new column to your spreadsheet today.
At 50 bets, repeat prompts 1 and 3 and compare.
How to use the 5 prompts in order: step by step
Common mistakes when auditing your history with an AI
- Skipping the totals check and believing the first analysis.
- Uploading only the bets you remember, or only the winning ones.
- Giving up a market because of a loss over twenty bets.
- Asking the AI to tell you what to bet on based on your history.
- Changing five rules at once and not being able to tell which one worked.
The AI audits what you give it.
If the record is incomplete, so is the audit.
An honest history includes the bets you would rather forget.
What Gambeta does with its own history, no promises
Gambeta publishes predictions with its own model.
Each one is recorded with its odds before the match, and it is not deleted if it loses.
You can review it in the public record and in the daily summary of the Betting Journal.
If you want to see how it works, go to the guide to AI for betting with a verifiable record, with free predictions from Gambeta’s AI every day.
A prediction is not a promise.
No good record guarantees the next result.
For adults 18 and over only.
If the audit shows that you bet more than you intended, or to win back what you lost, stop and ask for help.
More at responsible betting.
What to read next
If you want to keep your record with comments and not just with numbers, continue with the betting journal with AI and the Jornal.
There every bet also keeps its reason.
To improve the bets to come, use the 5 prompts to spot value bets.
If your problem is odds that change before you bet, there are the prompts for reading odds movement.
And for a specific market, go back to the index of betting prompts by market.
Frequently asked questions
Can I ask ChatGPT to analyze my betting history?
Yes, if you paste or upload the record.
The AI cannot access your bookmaker account.
Before believing an analysis, check that its totals match yours.
Which columns does my record need?
At a minimum: date, market, odds, stake and result.
If you add the time and your estimated probability, the audit improves a lot.
One bet per row and a single number format.
How many bets do I need to draw conclusions?
More than it seems.
With fewer than 100 bets in a market, almost any result can be chance.
That is why every prompt asks for the number of bets next to each figure.
Can the AI tell me whether I am a winning bettor?
Not with certainty.
It can tell you whether your result is compatible with chance for your number of bets.
A good run over few bets proves nothing.
Is it safe to upload my history to an AI?
Upload only what is necessary.
Delete your name, email, account number and payment methods before pasting the file.
For the audit, date, market, odds, stake and result are enough.
How often should I audit my history?
Every 50 or 100 new bets, or once a month.
Auditing after every bad run leads you to change method all the time.
Choose a fixed date and stick to it.