Before you start
You need telo connected in Claude or ChatGPT and a recent sync from your iPhone. If you have not connected yet, follow the Claude or ChatGPT guide first.
Each prompt starts with “Using the telo connector” on purpose: it makes the assistant read your records instead of answering from general training advice. If an answer has no dates and no counts, it did not read anything. Ask again, or ask which records it used.
1. Am I ahead of my last build?
Comparing a build with your own last one is fairer than comparing it with a plan from the internet. Same window, same watch, same person.
A good answer puts both 8-week blocks side by side: hours per week, the longest session of each, and hours in each heart rate zone. In our test the recent block had twice the weekly hours, but almost all of the extra time was easy cycling. Hard time was unchanged at about 0.9 h a week, and the longest run had halved.
Apple Health has no “race” flag. Give the date when the assistant asks; otherwise it guesses from distance, start time and heart rate, and tells you it guessed.
Using the telo connector, ask me the date of my last race. Then compare the 8 weeks before it with my last 8 weeks: weekly hours, longest session and time in each heart rate zone. Skip walks and anything under 20 minutes.2. Is my easy actually easy?
Most endurance plans want the bulk of the week easy. The watch knows whether it was.
telo splits heart rate into six bands by percentage of your maximum, so zone 3 starts at 70%. In our test the share below zone 3 fell every week for a month, from 67% to 36%, and one 100 km ride set the lowest week on its own. That is the useful part: not a verdict, but which session moved the number.
Zones use the highest heart rate seen in your workouts, which is an estimate. If you know your tested maximum, add it to the prompt. The 20 minute filter keeps short commutes from counting as easy training.
Using the telo connector, take my runs and rides of 20 minutes or more from the last 6 full weeks and show the share of time below zone 3, week by week. Flag the lowest week and the sessions that pulled it down.3. What did my long session say?
Holding the same pace while heart rate climbs is what coaches call cardiac drift. Your longest session is where it shows.
In our test, on a 100 km ride, moving speed was the same in both halves (29.9 and 30.4 km/h) while average heart rate went from 130 to 143 bpm. Heat, terrain, fuelling and pacing all move that number, so read it next to how the day felt.
Second-by-second heart rate is kept for 30 days, so ask within a month of the session. The assistant has to average a long series, which works best where it can run code, like Claude or ChatGPT on the web.
Using the telo connector, open my longest workout of the last 2 weeks, by duration. Compare heart rate and pace or speed in the first half with the second half, leaving out stops. Show the splits you used.4. How do I sleep before races?
The night before a race is famously short. The two before it usually matter more, and those you can change.
In our test the night before a half marathon was 5 h 42 min, against a usual 7 h 56 min: a 03:29 alarm for an early start. Before another race, a midnight bedtime two nights out was followed by a longer night than usual.
Nights without the watch on simply do not exist in the data. A good answer says which nights are missing instead of averaging around them.
Using the telo connector, ask me the dates of my last three races. Then show my sleep in the three nights before each one next to the 30 nights before it: time asleep, time awake and bedtime.5. Am I tapering or just tired?
This one compares you with yourself, not with a population chart. In our test resting heart rate was 48.6 bpm against a 49.3 baseline and HRV 59.7 ms against 62.1, both inside the usual range, and the last 7 days were 61% of the biggest training week of the year.
“Without walks” is there because of a real trap: the biggest week of the year in our test was a hiking trip. “No advice” is there on purpose. The numbers say where you sit; what to do with them is your call, or your coach’s.
Using the telo connector, put my resting heart rate, HRV and sleep from the last 7 days against the 30 days before, one line each. Then compare my last 7 days of training, without walks, with my biggest week this year. No advice.What each question reads
| Question | What it reads | Where it stops |
|---|---|---|
| Ahead of my last build? | Workouts by week, duration and heart rate zones | No race flag in Apple Health: give the date |
| Easy actually easy? | Time in each heart rate zone per workout | Zones come from your highest observed heart rate |
| Long session | Splits, pauses and the heart rate series of one workout | Detailed series kept for 30 days |
| Sleep before races | Sleep nights: time asleep, awake and time in bed | Nights without the watch are missing |
| Tapering or tired? | Resting heart rate, HRV, sleep and weekly training | Describes where you sit, does not decide |
None of this is medical advice. telo makes your records readable; it does not diagnose anything or tell you how to train.
Sources and availability
Plans, regions and interfaces can change. Check the official sources before setting up.
Practical questions
Do I need an Apple Watch?
For these questions, yes, or another watch that writes workouts with heart rate into Apple Health. Without heart rate, zones, drift and the taper check have nothing to work with.
Does it work in both ChatGPT and Claude?
Yes, in both, with telo connected. The long session question has to average a long series and works best where the assistant can run code, such as the web versions.
Will the assistant tell me how to train?
The prompts ask for numbers on purpose, and the last one says “No advice”. If you ask for a recommendation the assistant may give one, but telo is not a coach and nothing here is medical advice.
How this guide was prepared
Every prompt on this page was run on 5 November 2026 against one person’s real Apple Watch history through telo, each in a fresh session with no other context, and the wording was tightened where the first answers were ambiguous. The examples come from those runs: one person’s data, so they show what an answer looks like, not what yours will say. Prepared with AI assistance.