How to Review Your Training Log with AI

Months of logged sessions hold a lot of answers. Has your training volume crept up or drifted down? Have you been as consistent as you think? Did that four-week block actually look different from the one before it? Scrolling back through a log rarely answers those questions. An AI assistant connected to your data can.

How good the answer is depends on two things you control: what you logged, and how carefully you read what comes back. This guide covers both. It walks through what to log, how to set up a review, how to check an answer before you trust it, and a simple routine you can repeat every week or month.

What to log so the answers are useful

An assistant can only summarize what's in your log. A few habits make the difference between a vague answer and a useful one.

• Log every session, the same way. Consistency matters more than detail. Ten sessions logged carefully and three skipped will read as a drop in training, even if you trained just as hard on the days you didn't log.
• Finish your sessions. Tracked's summaries count completed sessions. A workout left open or abandoned halfway won't show up the way you expect.
• Log effort, not just weight and reps. "100 kg × 8" says what happened, not how hard it was. Recording RIR or RPE on your working sets keeps that context in your log. Pick one scale and stick with it: RIR and RPE run in opposite directions and aren't interchangeable. Tracked's training summary tells the assistant which scale you use, but it reports totals rather than per-set effort. When a question depends on how hard the work felt, add that to your prompt yourself.
• Weigh in on a steady cadence. Bodyweight swings day to day. Weighing in under similar conditions several times a week gives a trend something to work with. One weigh-in a fortnight doesn't.
• Log whole days of food, or none. Tracked averages nutrition over the days you logged. A half-logged day pulls the average down and looks like a low-calorie day.

Set up the review: pick a period, compare like with like

"How's my training going?" is a fine opening question, but it invites a fuzzy answer. A good review names the period and the comparison.

Name the period. Say "the last 28 days" rather than "lately". Tracked's summaries always cover a period ending today, so running your review on the same day each week or month keeps the periods lined up.

Compare equal lengths. Ask for the last four weeks against the four weeks before, not this week against last month. The connector can read the equal-length period immediately before the one you asked about, so a like-for-like comparison is one question away.

Compare like with like. A deload week against a normal week, or a holiday month against a training block, will show a drop that tells you nothing new. If your program changed in between, say so in the prompt, so the assistant doesn't treat a planned change as a surprise.

Start broad, then narrow. Begin with an overview of the last few weeks, then follow up on the one area that looks interesting: training, nutrition, bodyweight, check-ins, or sleep and recovery.

How to read the answer critically

AI answers sound confident whatever the data underneath looks like. Before you act on one, run through four checks.

Check the period, units and coverage

Every Tracked summary states its period, its units and how much of that period you actually logged. Look for those details in the answer. If the assistant doesn't mention them, ask. "You averaged 2,400 kcal" means something very different over 26 logged days than over 6. Check the units too: Tracked reports load, bodyweight and energy in the units you use and never converts them. If a number appears in a unit you don't use, the assistant converted it, and conversions can go wrong.

A missing day isn't a zero

A day with no logged session might be a rest day, or a session you didn't log. A day without a weigh-in is a gap in the line, not a data point, and an unlogged food day isn't a day you didn't eat. Tracked keeps "nothing was logged" separate from "zero" so an assistant can tell them apart. Watch for answers that blur the two, such as an average that quietly treats gaps as zeros, or a "streak broken" that is really a logging gap.

Trends need enough history

Two weeks of data is a snapshot, not a trend. Bodyweight in particular needs enough weigh-ins over a long enough span before a weekly rate means anything. When there isn't enough history yet, Tracked withholds the rate instead of guessing. That isn't the same as "no change". If an assistant gives you a confident trend from a short or patchy period, treat it as a hypothesis and keep logging.

Correlation across domains isn't causation

Ask about sleep and training together and you may see that your shorter-sleep weeks were also your lower-volume weeks. That's a pattern, not an explanation. A busy month at work could explain both. A good answer shows the two series side by side, with their coverage, and stops there. If an answer jumps to "your training is hurting your sleep" or the reverse, push back.

What AI is good at, and what it isn't

The useful mindset: the assistant is a fast, tireless analyst of your log, not a coach or a clinician. It's good at "what happened". You still own "why" and "what next".

A simple weekly or monthly routine

You don't need a long session with an assistant. A short, repeatable review beats an occasional deep dive.

Weekly (about five minutes)

1. Ask for an overview of the last 7 days compared with the 7 days before.
2. Check coverage first: how many days have training, nutrition and weigh-in data.
3. Note one thing that changed and one gap in your logging to fix next week.

Monthly (about fifteen minutes)

1. Compare the last 28 days of training with the 28 days before: sessions, sets, volume, and which muscle groups got more or less work.
2. Look at the bodyweight trend over the last few months, and accept "not enough data yet" as an answer.
3. Review nutrition averages alongside the number of days you actually logged.
4. Decide on one adjustment for the next month, and write it down somewhere you'll see it.

A few prompts that work well for these reviews:

• "Compare my training over the last 28 days with the 28 days before. Say how many days each period has data for."
• "Summarize my nutrition for the last four weeks, including how many days I logged. Average only the days I logged."
• "Show my bodyweight trend over the last 12 weeks. If there isn't enough data for a weekly rate, say so instead of estimating one."
• "Put my training and my sleep for the last four weeks side by side, with coverage for each. Describe them, but don't claim one caused the other."

Reviewing your Tracked log with Claude or ChatGPT

Tracked's AI connector lets Claude or ChatGPT answer questions from your own logged data. It reads seven summaries:

• Profile and units: your timezone, units and which of your data is available
• Recent overview: headlines across your data over up to 30 days
• Training: completed sessions, sets, reps, volume and duration, by exercise and muscle group, over up to 90 days
• Nutrition: calories and macros averaged over the days you logged, compared with your targets, over up to 90 days
• Bodyweight: weigh-ins, overall change and a smoothed weekly rate once there's enough history, over up to 180 days
• Check-in scores: scores from numeric check-in questions, over up to 90 days
• Sleep and recovery: sleep and, from connected devices, resting heart rate and heart-rate variability, over up to 90 days

Getting connected. Tracked is live in Claude's connector directory. Open the listing, connect it, then sign in to Tracked. Tracked's ChatGPT app is in review for the app directory. Until it's approved, you can add the connector as a custom connector in ChatGPT's developer mode. The setup guide has the connector address and step-by-step instructions for both.

Privacy and control.

• Read-only. The connector can't create, edit or delete anything in your account, send messages or act on your behalf, and it can only read your own account.
• Your call on what it reads. You sign in on Tracked's own page, and your password is never shared with the assistant. Tracked's consent screen lists the data categories being requested before you approve. If you're not comfortable with them, decline.
• Some things never leave. Photos, videos, messages, coach notes, free-text check-in answers, contact details, billing and menstrual-cycle data are never exposed.
• Disconnect anytime. Settings → AI connectors in the Tracked dashboard shows your connections and recent access history. Disconnecting stops new requests immediately.

Summaries are informational, not medical or nutrition advice.

Try it

The review is only as good as the log behind it. Download Tracked for free, log your sessions with effort, weigh in regularly, and once you have a few weeks of data, connect it to Claude or ChatGPT and run your first review.