Why not just use ChatGPT for a training plan?

You can, and the plan it writes in week one will look fine. The problem is the other fifteen weeks: it does not know what you ran, what you skipped, or how you slept unless you type all of it in every session, and almost nobody does that for four months.

Is Shed just ChatGPT with a running prompt?

It runs on a language model, so the family resemblance is real. The difference is everything around it: your training history in a database, paces computed from your actual runs, a stored plan that persists between conversations, rules that constrain what the model may prescribe, and triggers that start the conversation when your data changes.

Can I just paste my Strava data into ChatGPT?

You can, and for a one-off analysis it works. The problem is doing it every day for sixteen weeks. The value of automatic ingestion is not that it is clever, it is that it still happens in week eleven when you are tired.

Will ChatGPT give me dangerous training advice?

Usually not dangerous, but often ungrounded. The common failure is prescribing volume or intensity that does not match what you have actually been doing, because it does not know what you have been doing. That is how people get hurt in week five.

Compare

Shed vs ChatGPT

ChatGPT will write you a sixteen week marathon plan right now, and it will look reasonable. Then it will forget you exist. Shed holds one plan, reads every run you do, notices what you skipped, and starts the conversation when something changes.

Head to head

Bold marks the decisive difference on that row, whichever side it falls.

	Shed	ChatGPT
Knows what you ran	Every run, split and heart rate, automatically.	Only what you paste in, every time.
Notices you missed a session	Yes, and rewrites the week.	No. It will not know unless you say so.
Reaches out first	Yes, when your data says something changed.	Never. You start every conversation.
Keeps the plan consistent	One stored block, enforced against a fixed set of training rules.	Regenerates from scratch each time. Ask twice, get two plans.
Where the paces come from	Your own runs, re-anchored as your fitness moves.	A plausible-sounding table.
Price	$19.99 a month.	Free, or about $20 a month for the tier you probably already have.

The plan is the easy part

A generic sixteen week marathon plan is a solved problem. It has been in books since the 1970s, the structure is well known, and a language model reproduces it competently. If a static plan is what you want, you do not need to pay anyone for it.

What a chat window cannot do is the other 111 days. It does not know you ran 8 miles at 8:42 this morning with your heart rate 9 beats above where it usually sits at that pace. It does not know you have slept badly three nights running. It does not know you skipped Thursday. Every one of those facts has to be typed in, by you, every time, or the advice is being given blind.

In practice nobody does that for four months. The plan gets generated in week one and never updated, which makes it a PDF that took longer to produce.

Consistency is a feature

Ask ChatGPT the same training question twice and you can get two different answers, both confident. That is fine for a recipe and a real problem for a training block, where week nine only makes sense in the context of decisions made in week three.

Shed holds one plan, stored, and every change is made against it and recorded. The paces come from your own runs rather than from a plausible table. And the sessions are checked against a fixed set of rules about how much hard running a week can carry, so the plan cannot quietly drift into something a coach would not sign off.

Frequently asked

Why not just use ChatGPT for a training plan?: You can, and the plan it writes in week one will look fine. The problem is the other fifteen weeks: it does not know what you ran, what you skipped, or how you slept unless you type all of it in every session, and almost nobody does that for four months.
Is Shed just ChatGPT with a running prompt?: It runs on a language model, so the family resemblance is real. The difference is everything around it: your training history in a database, paces computed from your actual runs, a stored plan that persists between conversations, rules that constrain what the model may prescribe, and triggers that start the conversation when your data changes.
Can I just paste my Strava data into ChatGPT?: You can, and for a one-off analysis it works. The problem is doing it every day for sixteen weeks. The value of automatic ingestion is not that it is clever, it is that it still happens in week eleven when you are tired.
Will ChatGPT give me dangerous training advice?: Usually not dangerous, but often ungrounded. The common failure is prescribing volume or intensity that does not match what you have actually been doing, because it does not know what you have been doing. That is how people get hurt in week five.

The short version

Shed writes the block. Every session, every pace.

You give it a race and a date. It builds the weeks around the days you can actually run, reads every run off your Apple Watch the same day, and rebuilds the week when the week falls apart. $19.99 a month. The block is written and on your phone before you decide.

Get your plan on iPhone