Answered from the file. Published 2026-08-28. Last checked 2026-08-28

What ChatGPT knows about you, and which file holds it

Two different things, kept in two different places. Your conversations arrive as a file, conversations.json, in the export zip. The memory it keeps about you is a short list you read on a settings screen. Most guides collapse those two, and the short one is the one shaping your answers.

The Count runs entirely on your device. No account, no email field, no upload. If you do not have the export yet, the request takes about a minute and the email is reported to arrive within minutes.

The distinction the question turns on

Chat history is the transcript of what was said. Memory is the short set of conclusions kept about you and applied to later answers. The transcript is long and the memory is short, and the short one is the one you have never read.

People ask which file holds ChatGPT's memory of them because they assume the memory is buried somewhere in the export. It is a reasonable assumption and, as far as we can confirm, it is wrong. The export is built around the transcript. The memory is a separate, much shorter thing, and it is legible in one place that is not a file at all.

What is actually in the export

FileWhat it holds
conversations.jsonEvery conversation, with per-message timestamps. Each conversation is written as a mapping object of message nodes, with a create_time and a title. This is the file our free Count reads. in our parser
chat.htmlA browsable copy of the same conversations, built for a person rather than a tool. reported
user.jsonBasic account details. reported
message_feedback.jsonThe thumbs up and thumbs down you gave on individual answers. reported
model_comparisons.jsonThe A/B comparisons you took part in, where two answers were shown and you picked one. reported

conversations.json is labeled in our parser because that is the file our free Count opens and reads, and you can check what it does at pupulcorp.com/count.js before dropping anything on the page. The rest are labeled reported: they come from public accounts of the export, checked 2026-08-28, and the contents of this zip have changed before.

So which file holds the memory

The honest answer, which is not a file name

We cannot name one, and the pages that confidently do are contradicting each other. Some guides say the saved memories are folded into the export somewhere. Others say they are not in it at all and tell you to copy them by hand. We have not opened an export in which a memory list appears under a name we could point at, and OpenAI's own export article describes the zip as your chat history plus account data without naming a memories file. reported, checked 2026-08-28

So the answer we will stand behind is the one you can act on today: read the memory on screen, and keep your own copy. If that changes, this page changes, and it will say when.

How to read what it has saved about you

  1. Open Settings Click your name or profile icon, then Settings.
  2. Open Personalization Memory lives under Personalization rather than under Data controls, which is where most people look first.
  3. Open Memory, then Manage This is the list. Every saved memory, in its own words, each one deletable on its own.
  4. Copy it somewhere you own There is no export button on this screen. Selecting the list and pasting it into a note is currently the reliable way to hold a copy of it.

Reported as of 2026-08-28, from OpenAI's Memory FAQ. These screens move. If a step here has gone stale, tell us and it gets fixed.

Why the short list is the one that matters

Your transcript is a record of what you said. The memory is a record of what was concluded, and it is applied to answers you have not asked for yet. That is the same shape as every other platform in the inference file index: the activity log is long and freely handed over, and the conclusions are short, consequential, and harder to get at.

It is worth actually reading yours. People are routinely surprised by two things: how confidently a guess was written down, and how long a detail from one passing conversation has been quietly steering everything since.

What no file will tell you

The limits, stated plainly

Count your own conversations, free

The export answers a different question well: how much of your thinking has gone through it. Drop conversations.json on the free ChatGPT Count and it prints your conversation total, your message total, how many of those messages were yours, the date range, and your longest single conversation by name. It runs in your browser, nothing is uploaded, and there is no account, no email field and no card.

Run the free ChatGPT Count How to request the export

The Count gives you the numbers. The Reading is the next question: what the pattern across your files says the algorithms decided you are, written out in plain language you can argue with and mark wrong.

The full Reading is 6.99 a week or 69.99 a year. No free trial. There is a Refund button in your dashboard for 7 days, one tap, no reason required. The Count stays free forever either way.

Unlock my Reading

Where each claim on this page comes from

ClaimSource, checked 2026-08-28
The shape of conversations.json, and that Claude reuses the file name with a different shape insideOur own parser, readable at pupulcorp.com/count.js. It looks for a mapping object for ChatGPT and a chat_messages array for Claude, and reports which it found.
The export path, the wait, and the 24 hour linkOpenAI's help article, Exporting your ChatGPT history and data. reported
Where saved memories are viewed and managedOpenAI's Memory FAQ. reported
The other files in the zipConsistent public accounts of the export. We have not read them with our own code, so they stay labeled reported.
How this compares with other platformsOur own export file index and inference file index, both free as CSV under CC BY 4.0.

Questions people actually ask

Which file holds ChatGPT's memory about me?

We cannot name one, and pages that do are guessing. The export's documented core is conversations.json, which holds your chats and not a memory list. The saved memories themselves are read on screen under Settings, then Personalization, then Memory. Checked 2026-08-28.

What is in a ChatGPT data export?

conversations.json holds every conversation with per-message timestamps, structured as a mapping object per conversation with a create_time and a title. Public reports also describe chat.html, user.json, message_feedback.json and model_comparisons.json in the same zip.

Is ChatGPT's memory the same as my chat history?

No. Chat history is the transcript of what was said. Memory is a short written set of conclusions kept about you and applied to later answers. The transcript is long and the memory is short, and the short one is the one shaping your answers.

How do I read what ChatGPT has saved about me?

Open Settings, then Personalization, then Memory, then Manage. The saved memories are listed there and can be deleted individually. Reported as of 2026-08-28. Copying that list is currently the reliable way to hold your own copy.

How long does a ChatGPT export take and how long does the link last?

Reported as of 2026-08-28: the email usually arrives within minutes, OpenAI's help page describes up to 7 days, and the download link expires about 24 hours after the email. Miss it and you request again.

Keep pulling this thread

The free ChatGPT Countconversations, messages, hours, longest chatThe ChatGPT export guidethe exact taps, and the 24 hour linkCan you export ChatGPT memorythe portability question, answered separatelyThe export file indexwhich file holds what, on every platformThe inference file indexwhere each app writes what it decided you areMy export is missing the good partthe tier trap, platform by platformWhat TikTok knows about youanswered from the file, not the policyEvery export wait, one tableall 11 platforms, reported windows