Skip to content
Export in ChatGPT and Claude

Why your Apple Health export will not upload to ChatGPT or Claude

For a question about steps, send ChatGPT or Claude a CSV with one row per day. Years of individual measurements in the Apple Health XML file can exceed the reading limit. To keep up with your week, connect telo once and ask about your sleep or workouts as your iPhone syncs.

By the telo team · Reviewed

Upload limit is not reading limit

Each assistant has two different ceilings. The first is the size of the file it accepts. The second is how much text it can read at once, measured in tokens. An Apple Health export usually passes the first and blows through the second.

In ChatGPT, OpenAI's help gives 512 MB per file and 2 million tokens per text document; spreadsheets are exempt from the token count, at about 50 MB. In Claude, Anthropic's help gives 500 MB per file and up to 20 files per chat, warns that token limits may apply based on the length of extracted text, and suggests splitting large documents. When Claude analyzes with code (file creation and data analysis), the ceiling is 30 MB per file.

OpenAI estimates 1 token per 4 characters of English text. By that count, 2 million tokens is about 8 MB of text, and XML, full of digits and quotes, gets fewer characters per token. A 220 MB export.xml, the size of the 1 million record sample file from our CSV guide, would be tens of millions of tokens: several times the ceiling. This is an estimate, not an upload test.

What fits, by the published limits (checked on 2026-11-15)
FileSizeFits the upload?Fits the reading?
export.xml with 1 million records (illustrative sample)220 MB220 MB is under ChatGPT’s 512 MB file limit. For Claude, convert to CSV.No: tens of millions of tokens, against 2 million in ChatGPT (estimate)
CSV of one data type, one row per measurementTens of MBDepends: may pass ChatGPT's roughly 50 MB for spreadsheets and Claude's 30 MB for analysisThrough data analysis, yes, if it fits the upload
CSV of one data type, one row per day, 10 yearsAbout 3,650 rows, under 200 KB (estimate)Yes, in bothYes, in both

From the export to a file the assistant can read

The path is to export once, pick the question and send only the data type it needs, totaled per day. The script below streams export.xml, the same way as the tested script in our CSV guide, and writes a CSV with date, number of samples, sum and mean per day. It needs only Python 3.

  1. On the iPhone, open the Health app, tap Summary, your picture and, at the bottom of your profile, Export All Health Data. Confirm with Export.
    Health app profile scrolled to the bottom, with the blue Export All Health Data button below the Privacy section
    Health app profile. iPhone simulator, iOS 27, 2026-11-14.
  2. On the share sheet, tap Save to Files and move the file to your computer. Do not send the ZIP straight to the assistant.
    Share sheet with the export ZIP Archive at the top and the Copy, Add Tags and Save to Files actions
    The share sheet. The 2 KB size is from an empty simulator; a real export is much larger. iOS 27, 2026-11-14.
  3. On the computer, unzip the file. The records are in apple_health_export/export.xml.
  4. Save the code below as per_day.py next to the apple_health_export folder and run in Terminal: python3 per_day.py apple_health_export/export.xml HKQuantityTypeIdentifierStepCount Watch
  5. The last argument is optional and filters the source by name (here, only what came from the watch). You get HKQuantityTypeIdentifierStepCount-per-day.csv with one row per day.
  6. Send that CSV to ChatGPT or Claude and ask your question. For heart rate, use HKQuantityTypeIdentifierHeartRate and read the mean column, not the sum.
python
import csv, sys
import xml.etree.ElementTree as ET
from collections import defaultdict

src, want = sys.argv[1], sys.argv[2]  # export.xml HKQuantityTypeIdentifierStepCount
only = sys.argv[3] if len(sys.argv) > 3 else None  # optional: part of the source name
days = defaultdict(lambda: [0, 0.0])
events = ET.iterparse(src, events=("start", "end"))
_, root = next(events)
for event, el in events:
    if event == "end" and el.tag == "Record":
        if el.get("type") == want and (only is None or only in el.get("sourceName", "")):
            day = days[el.get("startDate")[:10]]
            day[0] += 1
            day[1] += float(el.get("value"))
        root.clear()
name = want + "-per-day.csv"
with open(name, "w", newline="", encoding="utf-8-sig") as out:
    w = csv.writer(out)
    w.writerow(["date", "samples", "sum", "mean"])
    for d in sorted(days):
        n, total = days[d]
        w.writerow([d, n, round(total, 2), round(total / n, 2)])
print(len(days), "days in", name)

A common mistake: steps counted twice

When the iPhone and the Apple Watch record steps at the same time, the Health app does not add them up: it uses the source at the top of the priority list, a rule Apple documents. export.xml does not apply that rule. It carries each source's records, with the source's name in sourceName.

If you sum every step record in the export, or ask the assistant to, you get a total higher than the Health app's on days you carried both. The assistant cannot know this on its own: it sees valid rows and adds them.

The paths side by side

Ways to get Apple Health data to an assistant
PathPriceWhat it needsLimitSource, date
Upload the whole export to ChatGPTIncluded in ChatGPT; the Free plan has fewer uploads per dayComputer or the ChatGPT app512 MB per file and 2 million tokens per text document; export.xml goes past thatOpenAI Help
Upload the whole export to ClaudeIncluded in ClaudeComputer or the Claude app500 MB per file and 20 per chat; 30 MB per file for analysis with codeClaude Help Center
Export + per day script, then the CSV to the assistantFreeComputer with Python 3; export again for new dataOne metric per file; the file is a snapshot of the day you exportedApple Support
ChatGPT HealthIncluded in ChatGPTiPhone with the ChatGPT app; a US accountUS only; Apple Health connects through the iPhone app, not the webOpenAI Help
Health in the Claude appPro, Max or Ultra plansiPhone with the Claude appBeta, US only; reads Apple Health, does not writeClaude Help Center
teloLast 7 days in chat on Free. Full synced history on Pro, US$49 per year in the US store.iPhone with the telo appReads synced daily metrics, nights of sleep and workouts. For individual measurements, use Apple’s XML export.App Store

Ask about this week’s sleep with telo

In the telo iPhone app, choose which Apple Health data to share. Connect the same telo account to Claude or ChatGPT and ask “Which nights did I sleep the longest this week?” Next week, your assistant can look up the nights your iPhone has synced since.

Free gives your assistant access to the last 7 days. With Pro, you can compare your full synced history, such as the month before and after a shift change. If you prefer a spreadsheet, download the CSV files from your account on the web, including on Free.

FAQ

Practical questions

Why does ChatGPT reject or cut off my export.xml?

ChatGPT caps text documents at 2 million tokens. A large XML file can exceed that limit even if it is under 512 MB. For a question about steps, send a CSV with a daily total.

Does Claude accept the Apple Health export?

Anthropic lists CSV as a supported chat format. Convert the metric you want to explore, attach the CSV and ask about the dates you care about.

What is the upload limit in ChatGPT and Claude?

ChatGPT: 512 MB per file, 2 million tokens per text document and about 50 MB for spreadsheets. Claude: 500 MB per file and 20 per chat, 30 MB for analysis with code. Figures from the official help pages, checked on 2026-11-15.

Can I use ChatGPT data analysis instead of having it read the text?

Yes. Attach the CSV and ask for your average daily steps each week. Data analysis can read the rows with code. Ask how many days went into each average so you can check the comparison.

Why does the assistant's step total not match the Health app?

The export carries iPhone and Apple Watch steps separately, and the Health app uses only the priority source when both count at the same time. Filter to one source before summing.

Does this work outside the US?

Yes. telo connects Apple Health to your assistant where the telo app and the connector are available, including Brazil. You can also attach a CSV wherever your assistant account supports file uploads.

Sources

How this guide was prepared

We checked file formats, upload limits and health integration requirements in the OpenAI and Anthropic help centers on November 15, 2026. Apple Support documents the source priority rule. The table estimates tokens using the illustrative 220 MB XML file from our CSV guide and OpenAI’s approximation for English text. The screenshots show the Health app in the iOS 27 iPhone simulator on November 14, 2026.