Open the archive on a computer
Unzip the download — every split part, if Google split it. Open archive_browser.html in a browser.
Look for .json files whose names look like walking data: steps, distance, floors, or active minutes. Emberstep keys off the file name, not a required folder title, so the same JSON can sit under Physical Activity or another product folder.
A matching name is not enough. Each file must be a JSON array of objects with dateTime and value. A GPS workout export (TCX from the Google Health app) is not this format and is not imported as a route.
What a usable steps file looks like
Fictional sample — two records on one day:
[
{ "dateTime": "06/04/24 00:00:00", "value": "3120" },
{ "dateTime": "06/04/24 12:00:00", "value": "4890" }
]Those two values add to 8,010 steps for 4 June 2024. Dates also appear as 2024-06-04…. Impossible dates (31 February) are rejected.
Sleep, weight, and heart files are not walking history. A file named like exercise is treated as exercise minutes: if the objects are not dateTime / value rows, that file can show as damaged rather than quietly skipped. Detailed GPS workouts are not imported as routes.
If two steps files cover the same day (for example a daily file and an intra-day file), Emberstep adds both. It does not pick “the official daily total” for you.
A fictional two-day folder
| Date | Steps | Distance | Floors | Exercise minutes |
|---|---|---|---|---|
| 4 June 2024 | 8,010 | 6.1 km | 4 | 25 |
| 5 June 2024 | missing | missing | missing | missing |
| 6 June 2024 | 5,000 | missing | missing | missing |
4 June also had a sleep JSON. That file was ignored and did not become a warning. 5 June is a gap — unknown coverage, not zero steps. 6 June has steps only; the other metrics are missing, not zero.
For your own archive, pick one day you remember. Add the value fields in complete steps files for that date. Compare with Google Health if the account still shows the day. If the figures differ, check the date, duplicate or overlapping files, and whether a file is missing. Emberstep cannot fetch missing records from the account.
Count distinct calendar days that produced at least one accepted record. That is the recovered-day count on the import receipt. Three days of files in 2019 is not a complete year.
Damaged files versus bad rows
A broken file (truncated JSON, not an array) contributes no records from that file. Other complete files for the same day can still supply their metric. In a check folder, a truncated steps file plus a complete distance file left steps missing and distance present.
A bad row inside an otherwise valid array (missing value, unreadable date) is skipped; neighboring rows in that file can still count.
Size, duplicates, and what will not come back
Emberstep reads one ZIP or one unpacked folder per import. Uncompressed total 2 GiB, 64 MiB per file. Combining split parts into one giant folder does not raise the cap. If the archive is larger, keep the original and send Support the folder names and error text, not the data.
Re-importing the same archive changes nothing: days are not written twice. Two different exports that overlap are not fully de-duplicated for you — check the overlapping days yourself.
Emberstep cannot invent days Google never exported, or restore data already deleted. It computes its own walking records. It does not restore Fitbit badges.
If you cannot find archive_browser.html or any steps JSON, you may have another Google product, a .tgz you have not unpacked, or a split part without Health files. Request a new archive with Google Health selected.
When you import on the phone, use Choose file. This site does not take the file.
Export reference
Google’s current export instructions identify Google Health as the product to select in Takeout. Checked September 2026.