Why AI-Generated Marathon Nutrition Plans Get It Wrong

You've typed the prompt. Something like: "Build me a marathon nutrition plan for a 4:15 finisher, 75 kg, running in October." And the answer came back fast, confident, and surprisingly detailed. Grams of carbs per hour. Gel timing. A pre-race breakfast. Maybe even a hydration schedule.

It looked perfect. The problem is that it was built for a person who doesn't exist.

The Prompt Sounds Specific. The Plan Isn't.

When you ask a general-purpose AI to build a fueling plan, it pulls from a vast pool of sports nutrition literature and distils it into a plausible-sounding recommendation. And in fairness, the broad strokes are usually fine. Aim for 60–90 grams of carbs per hour. Start fueling early. Don't try anything new on race day. That's textbook advice you'll find in any marathon guide.

But textbook advice is exactly the problem. Your body isn't a textbook.

The plan it gives a 75 kg runner targeting 4:15 will look nearly identical to the plan it gives every other 75 kg runner targeting 4:15. It has to — it doesn't know anything about you beyond what you typed into the prompt.

Here's what it's missing.

Your Sweat Rate Is Yours, Not an Average

Most AI-generated plans assume a "typical" sweat rate of around 1–1.2 litres per hour. That's a population average, and it's about as useful as saying the average shoe size is 9.

Real sweat rates among endurance athletes range from 0.5 to over 2.5 litres per hour. The difference between the low and high end of that range is the difference between sipping at aid stations and needing a deliberate fluid strategy every 15 minutes.

Your sweat rate depends on your fitness level, your body size, how well you're heat-acclimatised, and the conditions on race day. It changes across the season — if you tested it in April, it may be significantly different now after a summer of heat training. An AI that's never seen your sweat data is guessing. It has to be.

Gut Tolerance Isn't in the Training Data

Here's where AI plans do real damage. A chatbot might recommend 80 or 90 grams of carbs per hour because the latest research supports high-carb fueling strategies for marathon performance. That research is real. The recommendation isn't wrong in principle.

But your gut doesn't care about principles. It cares about what you've trained it to handle.

If you've been practising with 50 grams per hour on your long runs and your AI plan tells you to jump to 90, you're setting yourself up for nausea, bloating, or worse at mile 18. Gut tolerance is built over weeks of progressive training, not prescribed in a prompt response.

An AI has no way to know what your stomach can handle. It doesn't know that you switched from isotonic gels to hydrogels in June, or that fructose-heavy blends make you nauseous above tempo pace, or that you can only stomach solid food in the first half. Those details determine whether a plan works or sends you to the porta-potty.

Pace Changes Everything — and AI Uses One Number

When you tell an AI your target finishing time, it calculates a single average pace and bases the plan on that. A 4:15 marathon is roughly 6:02 per kilometre. Clean. Simple. Wrong.

Nobody runs a marathon at one pace. You'll start conservatively (or you should). You'll slow through hills. You'll push in the final 10K or fade — and the plan needs to account for both possibilities.

Here's why it matters for fueling: your carbohydrate burn rate at 5:40 per kilometre is meaningfully different from your burn rate at 6:20. A plan that assumes a flat pace under-fuels you when you're pushing and over-fuels you when you've slowed, which is exactly when your gut is least able to absorb extra carbs.

Your wearable tracks your actual pace, heart rate zones, and effort across every training run. That data tells you how your body burns fuel at different intensities. A fueling plan built from that data adjusts with you. A chatbot plan can't.

Race-Day Weather Isn't a Variable AI Plans For

You asked for a plan for "October." But October where? October in Chicago is different from October in Melbourne. A cool, overcast 12°C race and a warm, humid 24°C race need fundamentally different hydration strategies.

Heat increases your sweat rate, accelerates glycogen depletion, and diverts blood from your gut to your skin for cooling. That last point is critical — it means your ability to absorb gels decreases on hotter days, even if you take the same amount. A plan that ignores race-day conditions is a plan that breaks when conditions change.

Most athletes training through August right now are acclimatised to summer heat. If your fall race is cooler than your training, you'll actually sweat less and may need to pull back your fluid intake. If it's unexpectedly warm, you'll need more fluid and possibly fewer gels per hour to avoid GI distress. An AI-generated plan doesn't adjust for either scenario because it was built before the forecast existed.

The Real Problem: Confidence Without Calibration

The most dangerous thing about an AI-generated nutrition plan isn't that it's completely wrong. It's that it's close enough to feel right. The numbers are in the right ballpark. The timing sounds reasonable. The language is authoritative.

So you trust it. You follow it. And when things go sideways at kilometre 32, you don't know whether the plan failed or you failed. Was it the gel timing? The carb amount? The hydration volume? You can't debug a plan that was never calibrated to your body in the first place.

This is what separates generic AI advice from data-driven fueling. Generic AI gives you a plan that works for an imaginary average athlete. Data-driven fueling gives you a plan that works for you — because it's built from your training sessions, your sweat data, and your wearable metrics.

What a Personalised Plan Actually Looks Like

A fueling plan that accounts for your data doesn't just say "take a gel every 30 minutes." It says:

• Pre-race: Based on your body weight and glycogen targets, eat this amount of carbohydrate 3 hours before start. Drink this much fluid with this amount of sodium.
• First 10K: Your planned pace puts you at this heart rate zone. At this intensity, your burn rate is X grams of carbs per hour. Start fueling at kilometre 5 with a half gel.
• 10K–30K: Increase to your tested tolerance of Y grams per hour. Drink Z ml at each aid station based on your sweat rate in similar conditions.
• Final 12K: If your heart rate drifts above threshold, back off gel intake — your gut blood flow is compromised. Switch to smaller, more frequent sips of sports drink instead.
• Weather adjustment: Race forecast is 18°C and 65% humidity. Your training data from similar conditions shows a sweat rate of 1.4 L/hr. Fluid targets adjusted accordingly.

That's not a template. It's a plan built from your sessions, your data, and your race-day reality.

Your Data Already Exists. Use It.

If you own an Apple Watch, Garmin, or log your runs on Strava, you're already generating the data you need. Heart rate across different paces. Workout duration and intensity. Environmental conditions during training. Calorie estimates (imperfect, but useful as relative benchmarks).

The missing piece has never been the data. It's been a system that converts that data into a race-day plan you can actually follow — gel by gel, sip by sip, kilometre by kilometre.

Sweatr does exactly this. It pulls your wearable data, factors in your body weight and sweat rate, accounts for race-day conditions, and builds a fueling plan personalised to your body and your race. No prompt engineering required.

The Bottom Line

AI chatbots are extraordinary tools. They can help you write training plans, understand sports science concepts, and answer quick questions. But building a race-day fueling plan requires data they can't access: your sweat rate, your gut tolerance, your pace variability, and the weather forecast for your specific race.

A marathon is too long and too hard to fuel on guesswork — even when the guesswork sounds sophisticated. Use your data. Build a plan that's actually yours.