stats.training

What to ask your AI about your Garmin data.

Most people connect their Garmin, ask "how did I sleep last night", get a number they already saw in the app, and quietly stop. That question was never the point. Here are the ones that are.

The short version

Garmin already tells you what happened. Your AI is for the questions the app can't answer: comparisons across months, two metrics held against each other, and "given all of this, what should I do today?" Ask for those, not for numbers you can read off a screen.

The mistake everyone makes first

The first question is almost always a lookup. Last night's sleep score. Today's Body Battery. Yesterday's mileage.

Those work, and they're a fine way to check the wire is live. But the Garmin app already shows you all of it, faster, on your wrist. If that's the whole conversation, you've paid ten dollars a month for a slower version of a screen you already own.

The gap is this. Garmin is very good at one number at a time. It's poor at two numbers held against each other, and it has no opinion about what you should do next. That's the whole opening.

Sleep and recovery

Recovery data is where single readings mislead the most. One bad night means nothing. The shape of thirty nights means quite a lot.

"What's the trend in my resting heart rate over the last 60 days? Line it up against my weekly training volume and tell me if they move together." A resting heart rate creeping up while volume climbs is the earliest honest signal that you're absorbing less than you're adding.
"Compare my sleep on nights after hard sessions against nights after easy days or rest days. Over the last two months." Almost nobody knows this about themselves. Some people sleep worse after intensity for two nights running, which changes when you'd schedule it.
"My HRV status has been unbalanced for a week. Look at my sleep, stress, training load, and alcohol-free nights if you can see them, and rank the likely causes." Garmin tells you the status. It won't tell you which of four inputs is dragging it down.
"Do I sleep worse when I train in the evening? Split my nights by whether the session finished after 6pm." A question with an actual decision attached to it. If the answer is yes, you move your hard days to the morning.

Training load

The useful questions here are about direction, not position. Not "what's my load" but "is it going anywhere".

"Look at the last 12 weeks. Is my training load genuinely progressing, or have I been running the same week over and over?" The most common plateau isn't overtraining. It's ten weeks of identical volume at identical effort.
"Break down my time in each heart rate zone for the last month. What percentage was genuinely easy?" Most people run their easy days too hard and their hard days too easy. This is the question that shows it.
"Take my last six long runs. Show pace, average heart rate, and how much my heart rate drifted from the first half to the second. Is my aerobic base improving?" Cardiac drift shrinking at the same pace is fitness. It's hard to see one activity at a time, which is the only way the app will show it to you.
"Where's my biggest weakness right now: endurance, threshold, or hills? Use my endurance score, hill score, and lactate threshold." Garmin computes all three and shows them on separate screens with no ranking. Asking forces the comparison.

The questions only work with everything at once

This is the category with no equivalent anywhere in the Garmin app, because it needs several data types in the same answer.

"Is there a relationship between my Body Battery at bedtime and my sleep score that night?" Two metrics Garmin keeps in separate tabs and never compares for you.
"Look back six months. In the weeks before I got sick or took unplanned rest days, was there anything unusual in my HRV, resting heart rate, or sleep?" A personal early-warning system built from your own history rather than a population average.
"What day of the week am I consistently most recovered? Build my ideal training week around that." Most training plans assume your life is uniform across seven days. Yours isn't.

Decisions, which is the real point

The questions worth paying for are the ones that end in an action.

"Given today's training readiness, my Body Battery, and what I did yesterday: should I do the hard session today or move it to tomorrow? Give me a straight answer." Ask for a straight answer or you'll get a paragraph of hedging. It has the numbers to commit.
"I have a 10k in three weeks. Look at my race predictions and my recent sessions, then give me a target pace and the next three weeks of training." Better than a generic plan because it starts from what you've been doing. Push back on it and it adjusts.
"I felt terrible on today's run. Check my data and tell me whether that's in the numbers or just in my head." Sometimes it's there in the resting heart rate. Sometimes it isn't, and knowing that is worth something too.

The one that makes it stick

If you only take one thing from this page, take this one. Ask it every Sunday.

"Give me my week in review. What I did, how my body responded, what changed from last week, and the single thing you'd change next week." A standing weekly appointment beats twenty clever one-off questions. This is the habit that makes the connector worth keeping.

In Claude you can save that as a project instruction so it runs the same way every time. In ChatGPT, keep it in a note and paste it. Either way the value is the repetition, not the wording.

What it won't do

Worth knowing before you're disappointed by it.

  • It can't see what your watch never recorded. How you felt, what you ate, how work went. Tell it, and it will factor that in. It can't guess.
  • It will fill gaps confidently. If a number looks too neat, ask it to show the underlying values. It will.
  • It can't change anything. The connection is read-only, so it can't push a workout to your watch or edit an activity.
  • It's not a doctor. A resting heart rate trending the wrong way is a reason to see one, not a diagnosis.

Start here

Pick three from this page. The weekly review, one correlation question, and one decision. Ask them this week rather than saving the page for later.

The data has been sitting in your account for years. The only thing that changed is that you can finally ask it something.

Connect Garmin to your AI in a minute

33 read-only Garmin tools, from sleep and HRV to race predictions. $10/month, cancel any time. Works with Claude, ChatGPT, and any MCP-compatible AI.

‹ back to blog More on the Garmin MCP