PaceTrack AI Assistant
This is a detailed guide to the AI chat in PaceTrack. If you're looking for an overview of everything else PaceTrack does — training plans, payment control, .fit file analysis and more — see the general PaceTrack guide.
The AI assistant is a chat inside PaceTrack where a coach or an athlete can ask, in plain language, to do something with the system's data: pull up a plan, compare weeks, copy workouts, draft a message to an athlete, break down an interval structure. The assistant doesn't follow a fixed script — it decides for itself which data it needs, requests it through PaceTrack's internal tools (athlete profile, calendar, zones, macro-plan, chat, notifications, and around forty others), and forms its answer from real numbers.
The difference from a generic chatbot is fundamental: the assistant doesn't invent numbers or work "from memory." Every fact in its answer is the result of an actual query to your database, made right there in front of you, within that conversation. It works the same way for coaches and athletes — same engine, different set of permissions.
The core principle: the assistant advises, the human decides
Everything the assistant does follows one rule: the AI is a helper, the coach or athlete is the one responsible. This isn't a slogan — it's a set of concrete technical constraints:
- ✓ any plan edit is saved as a draft — publishing a workout in the same action as creating it is physically impossible, they're two separate steps;
- ✓ publishing a plan, messaging an athlete, rating a workout, and bulk-marking notifications all require a separate, explicit "yes" in the conversation;
- ✓ the assistant's toolset has no way to delete athletes, payments, or entire macro-plans — destructive operations simply don't exist;
- ✓ every tool call and every wallet charge is logged — a coach can always check the super-admin panel to see exactly what the assistant did and when.
In practice: you ask the assistant to copy one athlete's workouts to another's plan, it copies them as a draft and says outright — "to let the athlete see it, it needs to be published, please confirm." Nothing changes for the athlete until you do.
What the assistant can do, technically
| Area | Example capabilities |
|---|---|
| Athletes & profiles | athlete list, profile, intake questionnaire, coach's notes, time zone and weather at the athlete's location |
| Zones & physiology | heart rate/pace/power zones, anchors, recovery data (WHOOP, Garmin, Polar) |
| Planning | macro-plan (goal, phases, weeks), planned vs. actual calendar, step-by-step workout structure, templates, semantic search for similar workouts |
| Writing to the plan | creating and editing plan days and weeks (always as drafts), publishing a week, notifying the athlete the plan is ready |
| Feedback & diary | coach's workout ratings, athlete's diary |
| Communication | reading and sending messages in the coach-athlete chat, system notifications |
The athlete's toolset is a subset: self-scoped only, no access to other athletes, no publishing other people's plans.
⏱️ Saves time
Tasks that take minutes or tens of minutes by hand get solved by the assistant in a single exchange.
Copy a plan from one athlete to another — rescaled to their own level
Alex and Maria are training for the same marathon, but at different levels. The week's structure can look similar — both are on the same week of the same build-up cycle for the same race — but the actual paces need to be different, or Alex's workout will be all-out for Maria, or too easy, if you just shift the dates without rescaling.
The assistant reads Alex's week together with his zones to work out which relative intensity each pace represented (not the raw number, but the zone — "easy," "tempo," "threshold"), reads Maria's zones, and picks her own pace in the same zone, keeping the day's structure unchanged. The answer includes a table: "Alex (pace/zone) → Maria (pace/zone, recalculated time)," and the result is saved as a draft — publishing is a separate confirmation.
Find out who has no plan for tomorrow
A coach has forty-plus active athletes. In the evening, they need to quickly see who still needs a plan for tomorrow.
The assistant checks each athlete's calendar for tomorrow's date and returns a list of the ones with nothing scheduled — instead of the coach opening forty profiles one by one.
Bulk-processing notifications, broken down by date
This isn't as trivial as it sounds: first you need to know how many unread notifications there are and from which dates. The assistant pulls the list, groups it by date, shows the breakdown ("yesterday — 8, the day before — 7…"), and only after confirming the count marks the right ones as read, leaving today's untouched.
🛡️ Prevents mistakes
The assistant double-checks its own data, requires unambiguous confirmation for sensitive actions, and is upfront about side effects of edits.
The assistant didn't trust its own data and double-checked
A coach sees 49 unread notifications on screen. After the first query, the assistant only saw two.
Rather than silently agreeing with 49 or silently insisting on its own count of two, the assistant acknowledged the discrepancy, re-queried the data a different way, and only then gave a final breakdown by date. This is exactly the behavior that makes its numbers trustworthy: it doesn't bend them to match expectations, it verifies.
Publishing a plan needs a real confirmation, not just "yes"
A coach spent half the day sorting out workouts, asked to remove the draft flag — and got a clarifying question instead of a result.
This isn't pedantry for its own sake: the system deliberately requires an unambiguous phrase for publishing and sending messages, so a casual "yeah probably, go for it" buried in a long conversation doesn't turn into a real action the athlete will see.
The assistant caught its own time-zone mistake
An athlete is reviewing yesterday's workout together with the assistant.
A few steps later, while putting together a weekly report for the same athlete's club chat, the assistant circled back on its own, without being prompted: "By the way, you were right — the start was at 9:54, not 6:54, I mixed up UTC with local time." The key point isn't that the mistake happened — it's that the assistant doesn't argue with a fact the athlete knows better than it does, double-checks, and corrects itself instead of carrying the wrong number forward.
A heads-up about a side effect of an edit
A coach made a targeted edit to an interval workout's structure — adding calculated split times to the step comments.
Without this warning, a coach might assume that since the change is visible in PaceTrack, it's already on the athlete's watch. The assistant knows this integration quirk and mentions it unprompted, instead of waiting to be asked "will the athlete actually see this?"
🧭 Advises and guides
The assistant doesn't just execute commands — it analyzes data, clarifies a request it misread, and gives substantive recommendations.
Trend analysis without a manual Excel export
This is a classic fitness-tracking task: if heart rate per kilometer drops at the same pace, the athlete's endurance is improving. By hand, you'd export three different weeks of workouts, calculate heart-rate/distance for each, and put it all in a table. The assistant builds all three tables in one conversation and lets you see the trend directly, no spreadsheet required.
Clarifying the task on the fly instead of giving up
A coach's first attempt at asking about "the fastest kilometer" was answered with the average pace of the whole workout — not quite what was meant.
Given the clarification, the assistant didn't stay at the same, insufficiently detailed level — it found a deeper data source that actually gives a per-kilometer picture, and produced a final table of "best kilometer by week" for each workout individually.
A workout recommendation for a day with no plan yet
An athlete asks what to do tomorrow, but the coach hasn't set a plan yet.
The assistant doesn't just state that the calendar is empty — it gathers context (weekly volume, the last hard session, fresh recovery data from the watch) and gives a substantive recommendation, clearly framing it as its own advice, with the final call still belonging to the coach.
A draft message for review before sending
A coach noticed a gap in the calendar and wants to gently check in with an athlete, without writing the text from scratch.
The assistant pulls up the message history, profile, and calendar and drafts a message with the context in mind — but doesn't send it, because it wasn't asked to, only to show the text. The coach read it, could have edited it, said "send it" — and only then, after one more confirmation, did the message go out.
Working out the interval math inside a structured workout
This is math nested inside the workout structure: find the right repeat among all the steps, take the pace range, work out the average pace, apply it to the right fraction of the distance, convert seconds to mm:ss — and do it for two different segments inside one step. The assistant walks through the calculation right in the chat and writes the result into the structure's step comments.
For athletes: same logic, smaller scope
For an athlete, the assistant is the same engine, scoped down to a single person: their own plan, their own zones, their own recovery data, their own chat with their coach. It's not a stripped-down afterthought — athletes get a genuinely useful assistant:
- → break down yesterday's workout, plan vs. actual, with RPE and how it felt;
- → understand why the numbers in the system don't match what they remember (see the time-zone example above);
- → get advice for a day the coach hasn't set a plan for yet;
- → quickly find the right chat among several (personal with the coach, club) and send a weekly report there — also with confirmation before sending, just like on the coach side.
An athlete can't publish someone else's plan, can't see other athletes in the club, and can't act "as the coach" — the assistant is specifically instructed not to confuse roles, even if asked.
What the assistant will never do
This isn't about the model being cautious — these are missing tools and hard rules at the system level:
- • it won't delete an athlete, a payment, or an entire macro-plan;
- • it won't publish a plan or send a message without a separate, explicit confirmation of that exact action;
- • it won't rate a workout on the coach's behalf without asking;
- • it won't show a coach data about an athlete they don't have access to, or show an athlete another athlete's data;
- • it won't charge the wallet if there isn't enough balance — a new chat request simply won't start until the balance is topped up.
When the assistant honestly says "I don't know" or "I can't"
A good assistant isn't one that always answers confidently — it's one that recognizes the edges of its own data:
- → if an athlete simply has no macro-plan in the system, the assistant doesn't invent one — it says the data isn't there and offers to clarify or build one from scratch;
- → if data physically doesn't exist in the system (like per-kilometer pace where only workout averages are stored), it explains the limitation and suggests the closest available alternative instead of silently substituting an approximation;
- → on an off-topic question, it will gently remind you that its main job is training and athletes in PaceTrack, not a general-purpose calculator.
How to phrase requests to get more out of it
- ✓ no need to learn "commands" — write like you would to a colleague: "copy X's plan to Y," "who has no plan for tomorrow";
- ✓ for sensitive actions (publishing, sending a message, rating), be unambiguous: "publish it," "send it," "confirmed" — not "sure, I guess, go ahead";
- ✓ if the assistant misread the request, clarify what you meant — it can switch data sources and approach, not just repeat the same wrong answer more confidently;
- ✓ ask for a draft to review if you want control over the final wording of a message to an athlete — say "show me the text first" before saying "send it";
- ✓ name the period and the athlete explicitly if there's more than one in the conversation — it saves one clarifying question.
Billing: credits
Every message to the assistant spends balance from a wallet measured in credits — coaches and athletes each have their own separate balance. When the feature rolled out, everyone got a welcome grant, and the balance can be topped up from there. If the balance runs out, a new chat request simply won't start — you'll see a clear message with a top-up call to action; a response the assistant already started will always be completed rather than cut off midway. Daily spend is visible right in your account, as a table and a chart.
Tools themselves aren't billed — you only pay for the actual model call, based on the volume of the request and response, not for how many internal functions get called within a single message of yours.
Frequently asked questions
❓ Frequently asked questions
Can the assistant accidentally publish a plan or message an athlete without me?▾
No. Publishing a plan, sending messages, rating workouts and bulk-marking notifications all require a separate, explicit confirmation of that exact action in the conversation. A casual "yeah sure" is not strict enough — it needs an unambiguous "publish it" or "confirmed".
Does the assistant see coach-athlete chat messages?▾
Yes, within the rights of whoever it's talking to: a coach — the chats with their own athletes, an athlete — their own chat with their coach. Neither side sees anyone else's conversations through the assistant.
What happens if the assistant gets the data wrong?▾
Every number in a response comes from a real-time query to the system, not a guess. If a number looks off, the best move is to say so directly — the assistant will re-check the source instead of insisting it's right.
Can the assistant delete something important?▾
No — its toolset physically has no operation for deleting athletes, payments, or entire macro-plans. Destructive actions don't exist at the system level.
How is this different from the "AI weekly plan builder"?▾
These are two different mechanisms. The weekly and macro-plan builders are fixed, one-shot scenarios with no back-and-forth. The AI chat is a free-form conversation with access to dozens of tools, paid in credits, with history and the ability to clarify, push back, and ask for a recalculation.
Is the AI assistant available to athletes, not just coaches?▾
Yes. Athletes get the same engine, scoped down to a single person: their own plan, their own zones, their own recovery data, and their own chat with their coach. An athlete can't see other athletes or act on the coach's behalf.
Ready to try it?
The AI chat is available from the side menu right after you log in to PaceTrack.
Get startedAlready have an account? Log in