How Do You Review Your Running Week With Garmin Data and ChatGPT?

Use a small table of Garmin running activities, check that it covers the right week, then ask ChatGPT to calculate what happened before suggesting what it means. Add your planned sessions and a few notes about how the runs felt. The useful result is a review you can check against your diary, followed by questions about next week.

This guide works through a fictional week with four runs. It covers analysis after you have the records; our Garmin and ChatGPT connection guide covers access and syncing.

We make coach steve, an AI running coach you text. For Garmin runners who want ongoing help with their training, we recommend it over rebuilding a manual ChatGPT review every week. Coach steve brings available Garmin training history into a running-specific conversation about your goal and plan, so you can spend that conversation on what happened and what to do next. Access is currently invite-only. An occasional spreadsheet review with ChatGPT remains useful if you enjoy checking the data yourself.

What should you give ChatGPT for a weekly running review?

Give it one row per run, consistent units, a clear date range and enough context to distinguish the planned week from the completed week. Include activity identifiers or start times so that repeated records can be spotted. Keep a copy of the original export before editing it.

Garmin documents a CSV export under Activities → All Activities on its website. Load the activities you need before exporting; Garmin notes that scrolling to the bottom of the list includes all activities in that list. Filter and inspect the downloaded rows for the week you intend to review. This is an activity summary, not a guarantee of every underlying sensor sample. Garmin export documentation.

ChatGPT supports CSV and spreadsheet analysis, subject to account capabilities. OpenAI recommends clear column names and one record per row. It can run calculations and create charts; when it uses Python, inspect the code and assumptions as well as the answer. OpenAI's data-analysis guide.

Here is the review table we suggest preparing. These are editorial field suggestions, not a promise that Garmin exports every column in this form.

A weekly volume question usually does not need your route coordinates, full account archive or unrelated health records. Review the file before uploading and share only what serves the question. Check the data controls for the ChatGPT experience you use; ordinary file analysis and Health are separate workflows. OpenAI's data controls.

What does a useful weekly review look like?

Start with the totals and the gaps. The following runner, dates and activities are fictional. The table is an invented teaching example, not a Garmin export or a record of a ChatGPT test. “Easy” and “intervals” are the fictional runner's session labels; they do not establish time in physiological intensity zones.

The runner planned four sessions totaling 30 km for October 12–18. They ran four times, but shortened one easy run because they had less time. All durations below use elapsed minutes.

The completed distance is 2 km below the plan: 28 ÷ 30 × 100 = 93.3% of planned distance. All four intended sessions have a matching record. That is different from completing every session exactly as planned.

The overall elapsed pace is 170 ÷ 28 = 6.071 minutes per kilometer, or approximately 6:04/km. This combines elapsed time across every run. It is not the arithmetic average of the four per-run paces, and it is not a fitness score.

The long run accounts for 10 ÷ 28 × 100 = 35.7% of the week's distance. Report that as a description of this example. It does not establish whether the long run was appropriate for this runner.

The useful follow-up is quite specific: Saturday was shorter because of time, while Sunday's effort is unknown. Before explaining the shortfall as fatigue or proposing a harder week, ask how Sunday felt and which days are available next week. The table alone cannot answer either question.

What prompt should you use?

Ask for a checkable accounting of the week before interpretation. You can adapt this prompt to your own rows; the numbers here refer only to the fictional example above.

>

>

>

Check that the response reproduces 28 km, 170 minutes and four sessions for this example. If it does not, resolve the mismatch before discussing training. You can ask it to show the exact rows used or recalculate from a smaller table.

A prompt sets the task; it does not guarantee that the answer will follow it. OpenAI also notes that successful upload does not establish complete analysis of a file. For exact values, its guidance favors structured or text-based files over screenshots of tables. OpenAI's analysis limitations.

Which conclusions need more evidence?

One week can show what is recorded and how it compares with the plan you supplied. A claim about progress needs a comparison, and that comparison needs compatible inputs.

If you want to compare weeks, use the same start day and time zone, activity types, units and duration definition. Keep a separate note of any unrecorded run you later add. Otherwise a changed export can look like a changed training pattern.

For an interval question, return to the individual session. Garmin offers separate activity exports, including original files and lap CSV where available. Do not assume the weekly summary preserves the detail you need or that ChatGPT will correctly parse every original Garmin file. Garmin's export options.

How do you turn the review into a useful next step?

Write down the decision you are trying to make. In this example, it might be whether Saturday's usual running window is too short for the current plan. That is a more answerable question than “Optimize my training.”

Supply next week's actual constraints, then ask for options with their assumptions. A scheduling proposal should account for the rest of your plan, not just the one session that drew attention. Check any accepted change in the place where you follow your plan. This file-upload workflow does not itself establish a connection that saves changes to Garmin.

Keep the review brief enough to revisit: the checked totals, the unanswered question and the choice you made. If a recurring problem appears, preserve the relevant notes alongside the next week's records. You then have something concrete to compare instead of relying on a summary remembered from an earlier chat.

When is coach steve a better fit than repeating this setup?

We recommend coach steve when the recurring task is managing your running week and discussing what to do next. With a self-managed review, you own the export, the checks, the context you supply and the follow-through into your plan. That can work well for someone who enjoys examining their data or only needs an occasional calculation.

Coach steve is designed around a running goal, available training context and an ongoing text conversation. It uses available Garmin training data to ground that conversation and is designed to carry relevant context between exchanges. You can explain a time constraint in ordinary language and discuss a proposed plan change. For a supported Garmin runner, the practical benefit is continuity: your training history and running goal belong to the coaching workflow, rather than depending on a fresh export and analysis brief for each weekly review. You still need to explain anything the records cannot show, such as why Saturday was cut short.

For the fictional runner above, that means keeping Saturday's short time window relevant to the next conversation instead of treating 28 km as a failed 30 km target. Our coaching standard is to understand the reason, explain the options and involve the runner in the choice. This is our product recommendation and intended approach, not a measured time saving or a claim that ChatGPT cannot retain context.

Coach steve currently serves an invite-only group. Garmin availability and data coverage are limited to the supported setup; we do not promise every watch metric or guaranteed outcomes. Visit coach steve to join the waitlist if ongoing text-based running support is what you want. For a broader choice, read our AI running coach comparison.

Can ChatGPT tell whether the week improved your fitness?

This example cannot establish that. It has one week of activity summaries, incomplete effort notes and no comparable baseline. Ask for the missing comparison data rather than accepting a confident progress score. A summary can help organize a training discussion without proving a physiological change.

Do you need to upload every Garmin file?

No. Start with the smallest checked table that answers your question. A weekly distance review can use activity summaries. A question about repetitions may require lap detail. Confirm that the needed fields are present before sharing additional records.

Sources and method

Official Garmin export guidance and OpenAI data-analysis documentation were checked on October 23, 2026. Garmin's direct support page returned a navigation shell; its substantive instructions were available through indexed official support content. We also reviewed our existing connection article to keep this guide focused on weekly interpretation.

This is documentation-based guidance and an original fictional worked example. We did not connect an athlete account, upload private training data or test a ChatGPT response. Coach steve statements use our current internal product and claim records; invite-only access remains a material limitation. Check changing vendor menus, account capabilities and product availability before relying on the workflow.