16 October 2026 · Guy Turner
The chatbot said 39. It was wrong twice.
We asked a general chatbot a simple question: what's the highest number of disposals a St Kilda player has had against Hawthorn at Marvel Stadium? It answered in seconds. Seb Ross, 39 disposals. Confident, specific and wrong on both counts.
The right answer was Nasiah Wanganeen-Milera, 43 disposals, round 17, 2025, matching the official league statistics.
For a football fan, that's an annoyance. For a broadcaster, a club or a sports website, it's a liability. And it's the problem StatBreak was built to solve.
Fast isn't the same as right
General chatbots are remarkable at language. They're not built to do arithmetic on a database, and they don't know when they don't know. Ask them for a statistic and they'll produce something that looks like one. Sometimes it's correct. You can't tell which times.
People who work in sport can tell, though, and they stop trusting the tool the first time it's wrong. One bad number in a broadcast or a match preview costs more credibility than a hundred good ones earn.
Separate the reading from the maths
The fix sounds simple and took a great deal of work. StatBreak separates two jobs that general chatbots blur together.
The language model reads the question. It works out what you're asking for, which players, which venue, which time period, and what counts as a qualifying game. That's the part language models are genuinely good at.
Then ordinary, deterministic code computes every number from documented formulas. No statistic on screen is generated by the model. Every answer arrives with the table behind it, so anyone can check it.
Checkable answers build a business
Accuracy turned out to be the commercial case as well as the technical one. Sports media organisations already have audiences. What they need is a reason for those audiences to stay, ask another question and come back tomorrow. Verified answers, with tables and artwork a partner can publish, give them that.
So rather than spending a fortune buying its own audience, StatBreak puts its engine inside partners' own sites and apps. In a newsroom pilot, working journalists asked, narrowed and asked again, and every refinement showed what fans want next. The same engine also turns answers into games, such as Box Score Challenge, that people can play inside a partner's own site.
What this means for your business
Most businesses now have someone suggesting they put a chatbot on something. Fair enough. But ask two questions first.
What happens when it's wrong? If the honest answer is "someone gets embarrassed" or "we lose a customer", you need a design where mistakes are caught before anyone sees them.
Can the answer be checked? If people can see where a number came from, they'll trust it. If they can't, they'll eventually stop believing all of them.
Fast is easy now. Right, and provably right, is where the value is.
