Apple Health Data Anonymizer

Why anonymize Apple Health data before sharing it with AI?

Pasting your Apple Health export into ChatGPT or Claude is one of the fastest ways to get plain-English insight into your heart rate, sleep, and activity. But the raw export.xml that Apple produces is not just numbers — it's a dossier. Buried inside are your exact date of birth, blood type, skin type, the names of your devices (which usually include your first name, like "Keith's Apple Watch"), hardware serial numbers, and even ClinicalRecord entries that point to medical documents named after you. Once that text is pasted into a chatbot, you no longer control where it's stored, whether it trains a future model, or who can subpoena it.

This tool solves that in one step. It reads your export, removes or masks the identity fields, and hands you back a cleaned copy that keeps every health value intact — so the analysis still works, but the file no longer screams your name. Because everything happens in your browser, the sensitive original is never uploaded to us or anyone else.

What personal information is hiding in export.xml

• The <Me> element — your date of birth, biological sex, blood type, Fitzpatrick skin type, and a flag for cardio-fitness medication use. This is the single most identifying block in the file.
• sourceName attributes — the display name of each device or app that recorded data. Apple defaults these to things like "Keith's iPhone", leaking your first name across tens of thousands of records.
• device attributes — hardware model, manufacturer, and identifiers that can fingerprint your specific watch or phone.
• ClinicalRecord elements — links to external FHIR JSON files that are frequently named after you and contain lab results, medications, and diagnoses.

How the anonymizer works (and why it's trustworthy)

Apple Health exports are commonly 100 MB to over a gigabyte, so this tool never loads the whole file into a parser — that would freeze or crash your browser tab. Instead it reads the file in small chunks, applies targeted regular-expression replacements to each chunk, and streams the cleaned text straight into a downloadable file. Your profile data is anonymized, possessive names like Keith's Apple Watch are collapsed to Apple Watch, device attributes are dropped, and ClinicalRecord blocks are deleted. Because the transform is line-oriented and never re-orders anything, the output stays a valid Apple Health export.

Before you download, the tool shows you a before/after preview of the first changed lines plus a summary report counting exactly how many device attributes were removed and how many source names were masked. Nothing is hidden — you can see precisely what changed.

Date shifting: preserving trends while breaking the calendar link

Even after names are stripped, the exact timestamps in your file can be identifying — the day you flew, the night you didn't sleep, the week you were in hospital. Optional date shifting adds the same random number of days to every single date in the file. Because the offset is constant, the gap between any two readings is unchanged, so trends, streaks, and interval-based analysis stay perfectly accurate. What changes is that the data no longer aligns with the real calendar, decoupling it from your actual timeline. The tradeoff: day-of-week effects and seasonality shift by the same amount, so if your analysis depends on "weekends" or "summer," leave date shifting off.