Google Takeout
Request a copy of data Google holds for one Google Account. There is no API for this. The owner submits the form.
Parent argument: data-exports.
Request URL
Official help: How to download your Google data.
Submit
Sign in to the Google Account that owns the data. Stop. The owner signs in.
Open https://takeout.google.com/. Google pre-selects every product that holds data.
Click Deselect all, then tick only the products you want. A full dump of Photos plus Drive plus Mail is huge and slow. Prefer one product per request when the library is large.
Some products expose All data included. Open it and drop folders or data types you do not need.
Click Next step.
Delivery method: choose Send download link via email unless you have a reason to land the archive in Drive, Dropbox, OneDrive, or Box. Cloud destinations count against that service's quota, and Google stops being responsible for the files once they arrive.
Frequency. Three options, verified on the live page:
Export once(1 export),Export every month for 1 year(12 exports),Export every 2 months for 1 year(6 exports).Prefer monthly. It is the only mechanism any provider offers that turns an export into a recurring source, which is the whole difficulty with export-based ingestion. Confirmation reads
Export 1 of 12 will start on <date>.File type:
.zipunless you already know you want.tgz.Archive size: pick a maximum part size. Google splits anything larger into multiple files. 50 GB is the largest option and reduces the number of parts.
Click Create export.
Wait for the email. Open Download archive. Re-enter the Google password if asked.
Advanced Protection accounts: the first archive is scheduled two days out, and scheduled exports are not available.
Workspace accounts: an admin can disable Takeout. If the page refuses, that is why.
YouTube on a Brand Account: switch to that Brand Account before you export, or the videos will not be in the zip.
Which products to actually take
A full Takeout can reach 180 GB, almost all of it Google Photos originals and Drive files. Do not take that whole for a personal record.
- Photos originals are pixels, not searchable text. Their metadata — date, geo, filename — is small and worth having, but the full-resolution files bury storage and add nothing a query can use. Photos also already live in Google Photos, which is their backup. Skip them.
- Drive is files, not life-signal. Skip unless a specific document is the goal.
- Gmail, Contacts, Calendar have live APIs. If a CLI already syncs them, Takeout only duplicates them. Skip.
For a queryable personal record, Deselect all, then take the small, text-based products with no good API:
| Take | Why |
|---|---|
| Location History / Timeline | Where you were, over years. No API. |
| My Activity (Search, YouTube) | What you searched and watched. High signal. |
| Maps (your places, reviews) | Saved and rated places. |
| Fit / Google Health | Only if not already pulled another way. |
This turns a 180 GB request into a few GB that downloads in one pass and indexes cleanly. The rule matches the pipeline's principle: fetch what answers a question, not the rawest thing available.
What it contains
Whatever you ticked, if Google still holds it. Official examples: email, documents, Calendar, Photos, YouTube videos, registration and account activity.
Fitbit / Google Health is a Takeout product. After the Fitbit Web API turn-down in September 2026, this is the remaining official dump for a Google-signed-in Fitbit account. If the account still uses a legacy Fitbit login (not migrated to Google), use Fitbit Settings → Data Export instead. See Google Health export help.
Common formats inside the zip:
| Product | Format |
|---|---|
| Gmail | .mbox, labels in an X-Gmail-Labels header |
| Contacts | vCard |
| Calendar | .ics |
| Photos / videos | original file plus a sidecar .json (comments, album membership, extra metadata) |
| YouTube | original format, or MP4 (H.264 + AAC) |
| Drive | exported Google formats plus uploaded originals |
The zip also has archive_browser.html. Open that on a desktop to see how to read each folder.
What it does not
- Data already deleted from Google's storage. Items still in the deletion pipeline are omitted.
- Changes made between the request and the moment Google builds the archive (Drive sharing, resolved comments, Photos added or removed).
- A chosen date range. Takeout does not filter by time. You get all remaining data for the ticked products.
- Another person's content that you merely have access to (a shared Drive you do not own, a Google Group you do not own).
- Brand-Account YouTube videos from the wrong signed-in identity.
Delivery
Google emails a link. Official range: a few minutes to a few days. Most people get it the same day.
Click the link. You land on the Takeout downloads page, not on a raw file in the mail.
Reading the delivery email by API
An agent can find the mail and extract the link. Two traps make this harder than it looks.
A Gmail CLI that only does search and modify cannot read a body. Check the
subcommands before you plan around one. If the tool offers search and
modify and nothing that returns message text, the delivery mail is findable
and unreadable. A thin wrapper over the raw API returns it:
<cli> gmail users messages get --params '{"userId":"me","id":"<id>","format":"full"}'
Use both tools. Search with the one that has a good query syntax to get the message id, then read that id with the one that reaches the raw API.
The body is base64url and nested. A single top-level read reports
size: 0 and looks like an empty message. The text lives in payload.parts,
sometimes several levels down, so walk the tree and decode each body.data.
Treating the top level as the body is the most common way this fails silently.
Extracting the link does not remove the browser step. The URL in the mail is signed and still redirects to a login, so it cannot be fetched with a plain HTTP client using the token alone. Expect to hand it to a signed-in browser.
Calendar does not need Takeout
Calendar has an ordinary API. A CLI can page it directly, which is faster than requesting an archive and gives a live source rather than a snapshot.
Two things to get right when paging it: a --max style cap applies per
request, so walk the range in chunks rather than asking for everything at
once; and recurring events repeat across chunk boundaries, so dedupe on the
event id afterwards. On one real account that turned an apparent few hundred
events into 1,698 spanning eleven years.
Check the same question for any product before requesting an archive. Takeout is for data with no API, not for data you have not looked for an API for.
Format
.zip or .tgz, possibly split into numbered parts. One folder per product.
Gotchas
- The archive expires in about 7 days. Request again after that. Expiry does not delete the live Google data.
- Each archive can be downloaded 5 times. After that, request a new one.
- Split archives. Anything over the size you picked arrives as multiple files. You need every part.
.tgzplus unicode filenames. Google warns that tgz parts can fail on unicode names. Prefer zip.- Password again at download. Expected. 2-Step Verification may add a second check.
- Failed export. Retry. If it still fails, export one product at a time, or drop the part size.
- Do not download on a phone. Official advice: use a desktop. Photos libraries are large.
- Drive delivery plus account deletion. Move the archive off Drive before you delete the Google Account.
- No polling. There is no Takeout API for "is it ready". Watch the inbox. Scheduled exports still arrive by email — but a monthly schedule plus an inbox watcher is as close to a pollable export as this ecosystem gets.
- Deselect all first. 67 products are selected by default, and Gmail,
Photos and Drive together can run to hundreds of GB.
Deselect allthen pick what is actually missing; the counter showsN of 67 selected. - Defaults on the delivery step: link by email,
.zip, 2 GB parts.