ChatGPT and other AI chatbots can help with the arithmetic and paperwork around betting: explaining a bet type, checking a sum you give them, summarising your own records. They cannot tell you who will win. A chatbot predicts likely words, not outcomes, so it has no calibrated chance you can test, no scored record of past picks and, unless a tool fetches a page for it, no live prices, team lists or scratchings.
ChatGPT is one chatbot among several, and what follows describes chatbots in general, not any one product's features or settings. Used for the jobs below, with a check after each answer, a chatbot can save time. Asked for picks, it produces confident text about games it knows little about.
Why a chatbot cannot predict sports results
A chatbot is built on a language model: it writes words that are likely to follow your question, learned from a large body of text. Its confident tone is a property of the writing, not of any forecast underneath. Ask for a percentage and it will give one, but a 62% that comes out of a sentence has never been scored against results.
A prediction model is a different thing. It outputs a chance for each outcome, built from data, and its record can be measured over hundreds of games; AI sports betting shows how such a prediction is tested against the price. A chatbot has no scored record of its picks, and asking the same question twice can produce a different answer. So ChatGPT sports betting picks are text about a game, not a forecast of it.
Products that bolt a chatbot onto software that places bets are covered in AI betting bots. Custom chatbots sold as racing or sports tipsters still write text from a language model, whatever data is added. Judge their picks as you would any tipster's, on every tip with its time and price, as how to check a tipster record sets out.
No live data: prices, team lists and scratchings
A chatbot answers from text it learned in training, which stops at a date, unless a browsing or search tool fetches a page while it answers. A fetched page brings its own problems. It can be a preview written before the final team was named, yesterday's cached copy or plain wrong. It can also be an overseas site quoting an operator that may not lawfully take your bet.
What it is most likely to miss is exactly what moves prices on the day:
- in racing, scratchings up to the start, the track rating and rider changes
- in sport, team lists that change on game day, and the weather
- in both, the current price, which can move between the page being written and your bet
A chatbot that quotes $2.10 may be repeating a number from old text. Treat any price it gives as unchecked, compare the market on an odds comparison board and confirm the price in your bookmaker account before you bet. That is also why ChatGPT horse racing tips fall short: a racing tip without the day's field, track rating and prices is a tip about a different race. AI horse racing tips covers the rating models that do produce testable chances.
Invented statistics and where to check them
Chatbots can produce form lines, head to head records and injury news that do not exist, stated as confidently as true ones. With illustrative claims:
| What a chatbot says (illustrative) | Where to check it |
|---|---|
| "Team A have won 6 of their last 7 at this ground" | The competition's own match records, from its governing body |
| "Runner 4 is 3 from 4 on heavy tracks" | The runner's form on Racing Australia's site or the state racing body's |
| "Their full-back is out with a hamstring injury" | The official team list the competition publishes |
| "The track has been upgraded to Good 4" | The current rating on Racing Australia's track conditions pages |
| "This bookmaker pays three places in this race" | The bookmaker's own terms |
A made-up figure costs money when it moves your price. With illustrative numbers, say invented injury news lifts your estimate of a side from 45% to 55%. A $2.00 price then goes from 0.45 x 2.00 - 1 = -10% per dollar to 0.55 x 2.00 - 1 = +10% on paper, and you bet on a number that never existed.
Betting jobs a chatbot does well, and the check for each
| Job | A prompt that works | The check |
|---|---|---|
| Explaining a bet type | "Explain how an each way bet is settled in Australia, and list what I should check in a bookmaker's terms" | Read the bookmaker's terms, because place terms depend on whose rules apply |
| Converting odds | "Convert +175 and -140 to decimal odds and show the formula" | 1 + 175 / 100 = $2.75 and 1 + 100 / 140 = $1.71; confirm with the odds converter |
| Checking a sum on prices you supply | "Work out the implied chance of each price and the total, step by step" | Recompute one line yourself |
| Summarising your own records | "Total the stakes, returns and profit in this list of my bets" | Match the totals to your bookmaker's monthly activity statement |
Place terms show why the first check matters. Racing NSW's Rules of Betting for NSW on-course bookmakers pay a quarter of the win odds for the first three places when there are 8 or more starters (BR 17), while online bookmakers set their own terms. A chatbot trained on overseas text can describe another country's terms as if they were Australia's.
Example: Illustrative prices for a soccer match: home $2.30, draw $3.25, away $3.40. The implied chances are 1 / 2.30 = 43.48%, 1 / 3.25 = 30.77% and 1 / 3.40 = 29.41%, a total of 103.66%. If a chatbot says the three add to about 98%, the book looks as if it sits under 100%: an arbitrage, where the market pays out more than it takes in. In fact the overround is 3.66%, and a book balanced at that total keeps about 1 - 1 / 1.0366 = 3.5% of what is staked.
Implied probability explains each step of that sum.
For records, remove names and account numbers first. With six illustrative bets, staked at $25, $25, $40, $25, $30 and $25, where three won and returned $52.50, $72.00 and $105.00, the totals are $170 staked and $229.50 returned. Profit is 229.50 - 170 = $59.50, a yield of 59.50 / 170 = 35.0%.
A summary that calls the $229.50 profit has mixed up return and profit; under the National Consumer Protection Framework your monthly activity statement shows the net result, so match it there.
Prompts that keep a chatbot honest
| Instead of | Ask |
|---|---|
| Who will win on Saturday? | Here are both teams' prices. What chance does each imply, and what is the margin? Show each step |
| Give me a lock for today's races | Explain how a quinella pays, then list what I should check in the operator's rules |
| What is Team A's record against Team B? | Here are the last 10 results, pasted from the competition's official site. Summarise them and add nothing that is not in the text |
| Is this a good bet? | If I think the chance is 40%, what price do I need to break even? Show the formula |
The last one has a one-line check: break-even price = 1 / 0.40 = $2.50. Every good ChatGPT betting prompt follows the same pattern:
- Supply the prices and the data yourself.
- Ask for the working, then recompute one line.
- Ask it to list what it is unsure of.
- Never ask who will win.
Mistakes that cost money when betting with a chatbot
With illustrative figures:
| Mistake | What it costs |
|---|---|
| Betting a chatbot's pick with no price attached | Nothing tells you if the bet is worth having: at $2.00 a side has to win 1 / 2.00 = 50% of the time just to break even, and more than that to show a profit |
| Trusting an invented statistic | The 45% to 55% shift above turns a -10% bet into a +10% one on paper only |
| Copying a price it quotes | A $2.10 from old text may be $1.95 at your bookmaker, so a $50 winner returns $97.50, not $105 |
| Pasting a full account export | Names, account numbers and balances go to a provider whose terms may let it keep what you type |
| Asking it for a plan to win back losses | Any plan it writes is still chasing, which raises the amount at risk without improving the odds |
Never paste logins, ID or bank details
Keep these out of every chatbot: bookmaker usernames and passwords, one-time codes, identity documents such as a driver's licence, passport or Medicare card, and bank or card numbers.
In guidance published on 21 October 2024, the OAIC, Australia's privacy regulator, recommends as best practice that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools. It also notes that some commercial AI products have terms or settings that let the provider collect what customers type to train its systems.
The same caution applies to any betting software that asks for your login, and giving betting software your login covers what to ask first.
A chatbot is not a counsellor
If betting is causing you harm, talk to a person. Gambling Help Online offers free, confidential counselling by phone and online, 24 hours a day, 7 days a week, for people who gamble and for their family and friends: call 1800 858 858.
BetStop, the free Australian Government self-exclusion register launched on 21 August 2023, excludes you from every online and phone wagering provider licensed in Australia with one registration, for 3 months up to a lifetime. If you are in crisis, call Lifeline on 13 11 14, or 000 in an emergency.
A chatbot can sound supportive, but it cannot set a limit or close an account. Your bookmaker can: online bookmakers must offer deposit limits, where a lower limit applies at once and a higher one waits 7 days, and a simple way to close your account.
Where B337 fits
B337 is not AI, and no chatbot or model picks its bets. It is one of the betting tools in Australia that compare prices: the Terminal lines up prices from 40+ bookmakers across racing and sports, side by side.
Those prices are read on a repeating cycle, not tick by tick, so a price on screen can trail the bookmaker's own: confirm it in your bookmaker account before you bet. A free account opens a limited view of the Terminal with live odds. ChatGPT is a trade mark of its owner. B337 is not affiliated with it.
Risk: Betting involves risk. A chatbot's answer is text, not a forecast, and a price you have checked can still lose. See responsible gambling for limits and support.