What a prediction actually is
Ask ChatGPT or Claude to predict your marathon time and it will do arithmetic. Most commonly it applies Riegel's formula, which takes one known result and scales it to another distance using an exponent for how much you slow down as the distance grows. There is no mystery in it and but more importantly no model of YOU in it.
That matters because the output inherits everything about the input. A prediction from a sharp, recent, correctly-measured race is worth having. The same formula run on a parkrun you jogged in February produces a number with exactly the same confident formatting and none of the meaning.
Which result you feed it decides the answer
The single biggest source of nonsense here is a result that was never really a result. Four things break a prediction without being obvious about it:
- A split rather than a standalone effort. A 10K inside a half marathon is a pace you held while pacing something else, not a 10K.
- A course that flattered you. Downhill, point-to-point, or a strong tailwind all produce times you cannot reproduce on a normal day.
- A treadmill time, which is a different physical task and usually a different number.
- Age. A personal best from four years ago describes a runner who no longer exists, and the formula has no way to know that.
Feed any of those in and the prediction is not wrong by a little. It is answering a question about a different athlete.
Why ChatGPT doesn’t ask
A coach handed a suspiciously fast 10K time would ask where it came from. An AI assistant will not, because it has no basis for suspicion: it sees a number you typed and treats it as a fact you are asserting. Nothing in the conversation distinguishes a road race from a downhill effort, and the confident tone of the answer is unrelated to the quality of the input.
This is where a dated record earns its place, and the reason Athmex stores results with their context rather than as bare times. If the result carries what it actually was and when, the prediction can be sanity-checked before you plan a season around it. What to record after a race, and how to tell a genuine best from a fast split, is covered separately.
What a prediction is good for
Used honestly it is a useful reference point rather than a target. It tells you roughly what your current fitness is worth over a distance you have not raced recently, which is genuinely helpful when setting a goal or choosing a pacing band.
What it cannot do is account for the things that decide a somethjing like a marathon: heat, the specific course, whether you have done the long runs, fuelling, and whether you pace it sensibly in the first ten kilometres. A prediction is a statement about your current speed, not a promise about race day.
Where Athmex fits
Athmex stores your personal bests for running distances with the date each was set, alongside your threshold values, so ChatGPT asking about your current fitness is working from results you confirmed rather than whatever you half-remember mid-conversation, or worse, what ChatGPT has selectively remembered!
Every value in Athmex carries the date it became relevant, allowing the AI to understand how old each one is and allowing it to avoid treating a two-year-old personal best as current. Athmex does not calculate the prediction and does not tell you what to run. It supplies the inputs; the ChatGPT does the arithmetic.
Common questions
Is Riegel accurate for the marathon?
It is a reasonable approximation that tends to flatter marathon predictions from short-distance results, because the marathon punishes things a 10K does not, such as fuelling and durability. Treat a marathon number from a 5K as optimistic.
Which result should I give ChatGPT?
The most recent standalone race at the longest distance you have properly raced. That combination is usually the most honest single description of your current fitness.
Does Athmex predict my race time?
No. Athmex stores no activity history and performs no analysis. It supplies your confirmed personal bests, thresholds, goals and races to Claude or ChatGPT, and the assistant does the reasoning.