It did not ignore your week. It never had it.
The training plan that arrives with six sessions when you have four evenings is not a judgement about your commitment. ChatGPT and Claude both respect constraints well when they are given them, and neither has any way to discover yours. Told nothing, an assistant writes a training plan for a notional athlete with open evenings, a bike that lives on a trainer, and a pool down the road.
That athlete does not exist, and the plan built for them fails in the first week for reasons that have nothing to do with fitness.
Availability is a fact, not a preference
The useful distinction is between things that change with your fitness and things that change with your life. A threshold moves as you train. Tuesday does not become free because you got fitter.
Three kinds of constraint decide most plans, and none of them are visible to an assistant:
- The hours you actually have, which is usually fewer than the hours you would like, and the number worth giving is the honest one.
- Which days can carry hard work. This is rarely a training decision. It is childcare, shift patterns, a standing five-a-side game, or the fact that Thursday follows a late finish.
- Where the session can happen. A turbo in the garage and a road ride are different sessions; a lane swim depends on when the pool is open. Equipment and venue decide what is even possible before anything else does.
Give an assistant those three and the plan changes shape immediately. It will move intensity to the days that can hold it, size the week to the hours that exist, and stop prescribing sessions you have no way to do.
Why telling it again next week does not fix it
Most advice on this stops at writing a better prompt, and a better prompt does work, once. The problem is that it works once. Every new conversation starts from nothing, and the constraint you explained carefully in March is not present in the chat you open in June.
Memory helps less than people expect here. It stores short paraphrased notes rather than structured facts, so an availability pattern can survive as an impression and still be wrong in the details that matter. The mechanics of that are covered separately.
The result is a slow drift. You supply your constraints, get a plan that fits, then over following weeks get plans that fit slightly less well, until you are editing every output by hand and wondering why it got worse. It did not get worse. It forgot, which is a storage problem rather than a prompting one, and the gap Athmex was built to close.
What a constraint looks like written down
Vague constraints produce vague accommodation. "I am quite busy" changes almost nothing. "Six to eight hours a week, hard days Tuesday and Saturday only, everything else easy, turbo on weekdays and outdoors at weekends" changes the entire structure of the block.
The same applies to the temporary ones. A work trip in three weeks with no bike, a fortnight of early starts, a pool closed for maintenance: these are not complaints, they are planning inputs, and an assistant will build around them accurately if it knows about them.
Where Athmex fits
Athmex stores availability as part of your athlete context rather than as something you retype. Your weekly training hours, the days you can actually train, and a free-text constraints note are held with your thresholds, goals and races, and supplied to Claude or ChatGPT when you ask them something.
Shorter-lived constraints are recorded as dated notes, categorised as schedule or injury, so a fortnight of early starts is visible with the date it applied rather than remembered vaguely or not at all.
Athmex does not decide anything about your training. It stores no sessions and writes no plans. It supplies the facts you entered and confirmed, and the assistant does the reasoning against them.
Common questions
Why does the plan keep giving me more sessions than I asked for?
Usually because the number of hours was mentioned once earlier in a long conversation, or in a previous one, and is no longer in front of the model. Restating it in the current message fixes that instance.
Is this a prompt problem or a memory problem?
Both, and they need different fixes. A weak prompt produces a bad plan now. A missing durable record produces a bad plan every time you start fresh, which is the harder one to notice.
Does the assistant need my exact schedule?
No. It needs the shape: how many hours, which days can take intensity, and what equipment or venue limits you. Precise calendars are not the point, and an assistant given the shape will produce a week you can actually complete.
Does Athmex build the plan around my availability?
No. Athmex stores your availability and constraints; Claude or ChatGPT does the planning. Athmex holds no activity history and gives no training advice.