See what Google, Meta or Apple kept on you TAKEOUT

The export is designed to be technically complete and practically unreadable

Ask Google, Meta or Apple for your data and what arrives is a zip of folders named after products, with the interesting parts buried in JSON files nobody is expected to open. The obligation is to give you the data; there is no obligation to make it legible, and it is not. A Records.json with three million location points satisfies the law and tells you nothing.

This reads that zip (or the folder you unpacked it into) and lays it out by product: where you have been, what you searched for and watched, what the advertising side has written down about you, who wrote to you and how often, what you posted and what you liked. Every table is searchable, exports as CSV or JSON, and the location tracks export as GPX so TRAIL can cut your home out of them before you share anything.

How it opens a file bigger than the memory in the machine

These exports run to gigabytes, and the obvious approach fails on the first one. A browser cannot hold a one-gigabyte string, let alone parse it as JSON: the engine's own limit stops it before your memory does.

So none of the reading happens on the page. It happens in a Web Worker, and the zip is never unpacked into memory at all. A streaming unzip hands each entry over as a run of chunks, and each chunk goes straight into a streaming parser for its format. The big JSON files are not parsed as one document: a byte-level splitter finds the elements of the top-level array and parses each element on its own, so three million location points cost the memory of three million small rows rather than the memory of the file. The mbox is read the same way, line by line, keeping only the From, Date, Subject and label headers.

The page itself never holds the data. It asks the worker for one screen of one table at a time. That is why a multi-gigabyte export opens here while the tab stays responsive, and it is the same reason nothing can leave: there is no point in the pipeline where a copy exists to send.

What to look at first

Location history is usually the one that lands. Google has kept it in at least five different shapes over the years (the flat Records.json, the Semantic Location History months, Timeline Edits, the on-device Timeline from Android, the location-history file from an iPhone), and all of them are read here, because which one you have depends on when you asked and what you carried. Seeing several years of your own movements plotted is a different experience from knowing abstractly that they were recorded.

Then the advertising folder, which is the one people find genuinely unsettling: the inferred interests, the categories you have been put in, the things a company concluded about you that you never told it. It is written in plain language because it was written for advertisers rather than for you.

Then search and watch history, which is the longest and most personal record most people have of their own attention, kept in full, in order, for as long as the account has existed.

When you have looked, the useful next step is to reduce what gets collected from here on: turn the histories off in the account settings, and use EXPUNGE for the brokers who bought the parts that were sold. DATASCAN finds personal data hiding in any spreadsheet you export from this, before you send it to anybody.