
- The audit starts outside the AI: without the complete record, with dates and odds, there is nothing to audit.
- The AI organizes and calculates sample size, average odds, yield, streaks and stakes, without accusing anyone or proving fraud.
- With few bets the result may be pure luck: the final prompt asks the AI to say so first.
There are seven steps and the spreadsheet is built only once. Bookmark the article: your ticks stay in this browser and you can carry on whenever you can.
Your progress is saved in this browser.
What can and what can’t the AI do when auditing a tipster?
Auditing is comparing what the tipster says with what the numbers allow you to claim.
The AI is useful for the boring part: adding up, averaging and looking for patterns across hundreds of rows.
The judgment is still yours.
There is a clear boundary.
The AI cannot prove that someone lied.
It can only show you whether the numbers you gave it are consistent with each other.
If you are still unsure whether to follow someone, start with how to spot a fake tipster.
To see how an AI compares with a person, see AI vs tipster.
This article is the next step: the method with numbers.
This guide is about signals, not accusations.
An odd result can be bad luck, a data-entry error or a badly built record.
The features of each AI change: check in the app what your plan includes.
Betting involves risk, 18 and over only.
What data to ask the tipster for before you start
Without data there is no audit.
The AI does not fill in what is missing: it invents it or ignores it.
That is why the first step is to gather material.
- The complete record of picks, from the first one, with no cut-off by date.
- The date and time each pick was posted, not just the date of the match.
- The published odds and the bookmaker where they were available.
- The stake of each pick, in units, and the reference bankroll.
- The result of each pick, including pushes and voided bets.
- A way to check the dates: screenshots of the channel or a record with a visible publication date.
If the tipster cannot or will not give you this, you have already learned something.
Note it down as a data point, without drawing conclusions yet.
Paid tipsters deserve the same audit: first see how to choose a paid tipster.
How to build the tipster’s record in a spreadsheet
One row per pick.
You fill in seven columns and three calculate themselves.
Row 2 is the first pick.
| Column | What you enter | Example formula |
|---|---|---|
| A · Date posted | Day it was posted | By hand |
| B · Match and market | Text of the pick | By hand |
| C · Published odds | Decimal | By hand |
| D · Closing odds | The ones at kick-off | By hand |
| E · Stake | Units | By hand |
| F · Result | W, L or V (void) | By hand |
| G · Profit | Calculates itself | =IF(F2="W",E2*(C2-1),IF(F2="L",-E2,0)) |
| H · Implied prob. | Calculates itself | =1/C2 |
| I · Profit at closing odds | Calculates itself | =IF(F2="W",E2*(D2-1),IF(F2="L",-E2,0)) |
| J · Source | Where you saw the pick | By hand |
Format column H as a percentage.
The formulas use commas as the argument separator; if a formula gives an error, try changing the commas to semicolons.
If your odds come in another format, convert them to decimal with the odds converter.
Want a ready-made structure?
The guide to betting control in Excel works as a base and you adapt it to these columns.
How to audit a tipster with AI in 7 steps
Steps 1 to 3 are the setup.
Steps 4 to 7 are the questions you ask the AI.
Tick them off as you go.

Which prompt to use to summarize the sample, the average odds and the yield
Paste the spreadsheet or attach it.
Ask it to show the calculations step by step so you can check them.
If the result does not match your spreadsheet, your spreadsheet wins.
I am giving you a tipster’s record in a spreadsheet. Columns: A publication date, B pick, C published odds, D closing odds, E stake in units, F result (W won, L lost, V void), G profit. Calculate using only this data: 1) number of picks, not counting voids, 2) win rate, 3) average published odds, 4) yield = total profit divided by total stake, in %, 5) the win rate needed to break even at those average odds (1 divided by the average odds). Show each calculation with its numbers. On the first line tell me whether the number of picks is too small to draw conclusions. If a data point is missing, write “no data”. Do not estimate it and do not assume the tipster is good or bad.
So that you have something to compare against, look at this example with made-up numbers.
The AI should give you back something similar, with its calculations visible.
| Data (example) | Value | How it is calculated |
|---|---|---|
| Picks | 200 | Without voids |
| Won | 104 | 52% win rate |
| Average odds | 2.00 | Average of C |
| Profit | +8 units | 104 × 1 − 96 × 1 |
| Yield | +4% | 8 ÷ 200 |
| Win rate to break even | 50% | 1 ÷ 2.00 |
That +4% looks good.
But with 200 bets at odds of 2.00, chance alone moves the win rate about 7 points up or down.
The 52% in this example fits a tipster who really wins 50%.
The full calculation of the margin of error is in how many bets you need to know whether you are winning or losing.
There you will see why a short sample misleads so much.
How to compare the published odds with the available odds
This is where most tipsters fall apart.
Odds of 2.10 that nobody offers anymore by the time you get in are not odds you could have taken.
That is why you always compare what was published against what was there.
With the same spreadsheet, you compare the published odds (C) with the closing odds (D) of each pick. Calculate: 1) in how many picks C was higher than D, equal or lower, 2) the average difference between C and D, 3) the yield using the published odds and the yield using the closing odds, 4) the differences of more than 0.10 between C and D, listed with their date. Do not conclude anything about the tipster. Just tell me what the numbers show and what data point I am missing to interpret it (for example, the time each pick was posted).
A simple example shows why it matters.
Take 100 picks, 50 won, all at published odds of 2.10 and a stake of 1 unit.
| Example | Odds | Profit | Yield |
|---|---|---|---|
| Published odds | 2.10 | +5 units | +5% |
| Odds available at closing | 1.95 | −2.5 units | −2.5% |
Same picks, same results.
With the odds that were there at closing, the tipster goes from winning to losing.
The implied probability goes from 47.6% (1 ÷ 2.10) to 51.3% (1 ÷ 1.95), and that is what the difference costs you.
One caveat: the closing odds are not always the ones you could have taken.
The tipster may have posted early and the odds dropped afterwards.
That is why the posting time matters as much as the odds.
To understand that number, see what implied probability is.
How to detect deleted picks, odd streaks and changing stakes
Here the AI saves hours.
It looks for jumps in the dates, missing results and stakes that go up right after a loss.
It does not tell you why it happens: it tells you where to look.
Review the record in this spreadsheet and look only for formal signals, without interpreting intentions: 1) gaps of more than [DAYS] days between one date and the next, 2) picks without a result entered, 3) the longest winning streak and the longest losing streak, 4) whether the average stake of the picks that follow a loss is higher than that of the rest, with both averages, 5) markets or leagues with very few picks. For each signal, list the rows involved. If there are none, say so. Do not use the words fraud or scam.
On streaks, keep in mind that a long streak can happen by itself.
The article on losing streaks: how many are normal gives you the reference.
An odd streak is a question, not a verdict.
The case of deleted picks deserves an example.
A tipster showed 100 picks with 50 won at odds of 2.10: +5 units, 5% yield.
But 120 had gone out in the channel, and the 20 deleted ones were all losses.
| Example | Picks | Won | Yield |
|---|---|---|---|
| What is shown | 100 | 50 | +5% |
| What was actually posted | 120 | 50 | −12.5% |
The bottom-line calculation: 50 × 1.10 − 70 = −15 units, on 120 of stake.
That is why dated screenshots are so valuable.
Only with them can you see what was deleted.

Which red flags to look for
No signal proves anything on its own.
Several together call for more care.
Note how many you find before you decide.
| Signal | What it may mean | What you do |
|---|---|---|
| Record without dates | The order cannot be verified | Ask for dates or screenshots |
| Published odds far above the closing odds | Unobtainable odds or posted late | Ask for the time of each pick |
| Picks that disappear | Edited record | Save screenshots from today |
| Stake that goes up after a loss | Loss chasing | Look at the risk, not just the yield |
| Only the best months are shown | Selective data | Ask for the whole period |
| Promises of a fixed result | Marketing, not a method | Distrust the tone |
This list is not a list of culprits.
It is a list of questions to ask the tipster.
If they answer with data, that is a good sign.
When the sample is not enough to draw conclusions
This is the part that gets skipped the most.
A short record can look brilliant or disastrous without the tipster having anything to do with it.
Chance dominates small samples.
With the results in this spreadsheet, answer: 1) given the number of picks and the win rate, what range of win rates is compatible with chance? Show the calculation (standard error and a range of roughly two standard errors), 2) does that range include the win rate needed to break even at these average odds? 3) what can be claimed with this sample and what cannot? Use cautious language. If the sample is small, the first line of your answer must be “the sample is not enough to conclude”.
With the earlier example: 104 wins in 200 picks gives 52%.
The standard error is 3.5 points, and a two-error range runs from about 45% to 59%.
The 50% falls inside, so that +4% proves nothing.
That calculation uses a simple approximation.
It is not a verdict, it is an order of magnitude.
If you want to go further, the article on calibrating probabilities teaches you to measure whether a tipster’s estimates come true in each band.
Which common mistakes to avoid
- Auditing with a summary instead of the complete record.
- Pasting data without dates into the AI and believing it will order it correctly.
- Accepting the AI’s calculation without recalculating a row by hand.
- Concluding that a tipster is good just because the yield is positive over a short sample.
- Concluding that a tipster is bad because of a losing streak.
- Asking the AI to decide whether to follow them: that decision is yours.
When to follow, when to wait and when to walk away
The audit does not end in “yes” or “no”.
It ends in one of three positions.
- Follow with a minimum stake: complete record, verifiable dates, obtainable odds and a large sample.
- Wait and keep recording: small sample or missing data.
- Walk away: edited record, unobtainable odds or promises of a fixed result.
And while you decide, you have an open reference to compare against.
The AI for betting from Gambeta publishes free predictions with their probability and its public record.
You can audit it with this same method: the complete record is in plain view.
Responsible gambling: auditing a tipster does not remove the risk of betting.
Bet only money you can afford to lose and read the guide to responsible gambling.
18 and over only.
What to read next
If you also want to audit your own record, the next step is prompts to audit your betting history.
To measure whether your probabilities were good, continue with calibrating probabilities.
And for the close of each week, there is the weekly summary with AI.
Understand how many bets you need
Frequently asked questions
Can the AI tell me whether a tipster is a scam?
No.
The AI does not verify identities or proof of money.
It only organizes and calculates with the data you paste, and that data may be incomplete or altered.
What data do I need to audit a tipster?
Publication date, match, market, published odds, stake and result of each pick.
If you can, add the time it was posted and a screenshot.
Without the date and the odds, the audit is useless.
How many bets are needed to draw a conclusion?
Many more than it seems.
With a few hundred bets the yield still has an enormous margin of chance.
The detail with calculations is in the article on how many bets you need.
What do I do if the tipster deletes picks or does not let me see the record?
It is a red flag, not proof.
Save screenshots of what they post from now on, with the date.
Without a complete record, any number they show you is not comparable.
Is auditing a free tipster the same as auditing a paid one?
Yes, the method is the same.
What changes is how much money you put at risk if you judge them wrongly.
With a paid tipster, audit before you subscribe, not after.
Can I use any AI?
Yes, as long as you can paste the spreadsheet into it or attach it.
Features change: check in the app what your plan includes.
Recalculate a row or two by hand to make sure it did not invent anything.
🌍 GLOBALFrom any country · Telegram with a translator into your language
📲 The AI's picks, free on your phone
Every new pick arrives the moment it's made, with a link to its analysis. Choose where to get it:
How each one works: Free AI betting channel on Telegram · Free WhatsApp betting tips channel · Betting channel: Telegram or WhatsApp?