---
title: ChatGPT memory not working? A checklist to find the cause
description: ChatGPT memory not working or not saving? Run seven checks in order, from the two switches to Projects, custom GPTs and admin settings, with a fix for each.
date: 2026-11-18
updated: 2026-11-18
author: The Esment Team
category: Memory fixes
tags: chatgpt, memory, troubleshooting
canonical: https://esment.notas.ai/blog/chatgpt-memory-not-working
---

# ChatGPT memory not working? A checklist to find the cause

*The Esment Team · 2026-11-18 · [HTML version](https://esment.notas.ai/blog/chatgpt-memory-not-working)*

## In short

- Most "ChatGPT memory not working" cases are a setting or a chat type, not a bug. Check the two memory switches first.
- Turning off Reference saved memories also turns off Reference chat history. The reverse does not happen.
- Temporary Chats never create memories, and an older OpenAI help article says custom GPTs do not use memory at all.
- Projects and workspace admins can narrow or disable memory, and "Chat memory full" is a different problem from saved memories.

If ChatGPT memory is not working, the cause is usually a switch, a chat type or a permission rather than a bug. **Work through the checks below in order**, from the cheapest to the least likely. Each one says how to recognize the problem and what to change.

This is a troubleshooting list, not an explanation of how memory works. For the concepts, read [how to make ChatGPT remember previous conversations](https://esment.notas.ai/learn/how-to-make-chatgpt-remember). Product details here are as of November 2026, drawn from OpenAI's help center, and menu names change often, so match them against what you see in your own app.

## Why ChatGPT memory is not working: which kind is failing?

"Memory" means three different things in ChatGPT, and the fix depends on which one broke.

- **Saved memories:** specific facts ChatGPT keeps on their own, such as "Alex prefers metric units." You can see them in a list.
- **Reference chat history:** ChatGPT drawing on what you said in earlier conversations. OpenAI's [memory FAQ](https://help.openai.com/en/articles/8590148-memory-in-chatgpt) says there is no separate storage limit for this one.
- **The context window:** how much of the current conversation the model can hold. It resets in every new chat and has nothing to do with the other two. If ChatGPT loses details inside one long chat, that is this one, and [why ChatGPT forgets](https://esment.notas.ai/blog/why-chatgpt-forgets) is the post to read instead.

**Figure.** Memory has to be allowed at every layer. The checks below go from the bottom layer up.

- Workspace admin (if any)
- Your plan and region
- Your two memory switches
- Chat type and Project scope
## Check that both memory switches are on

Open **Settings › Personalization** and look for **Reference saved memories** and **Reference chat history**. The labels have moved before, so look for those two ideas if the wording differs.

The two are linked. Per OpenAI's memory FAQ, turning off Reference saved memories also turns off Reference chat history, while turning off chat history leaves saved memories intact. So if a past update or a tidy-up pass switched off the first one, both stopped.

> [!STEPS]
> 1. Open **Settings › Personalization**.
> 2. Turn on **Reference saved memories**, then check that **Reference chat history** is on too.
> 3. Start a new normal chat and say "Remember that I prefer metric units."
> 4. Look for a **Memory updated** note under the reply, or check your saved memories list if your app no longer shows one.

Turning chat history off is not free to undo. The same FAQ says information remembered from past chats is scheduled for deletion from OpenAI's systems within 30 days.

## Rule out Temporary Chat and custom GPTs

These two chat types are the most common reason a fact never appears in your memories, and both are by design.

**Temporary Chat.** OpenAI's [Temporary Chat article](https://help.openai.com/en/articles/8914046-temporary-chat-in-chatgpt) says these chats do not create or update memories. Recognize it by the Temporary label at the top of the conversation. Fix: start a regular chat. Release notes from August 2026 mention a setting for whether a Temporary Chat can read existing memories, but we could not confirm its exact name, so check it in your app.

**Custom GPTs.** OpenAI's article on [memory with GPTs](https://help.openai.com/en/articles/8983148-does-memory-function-with-gpts) says GPTs do not use saved memory, custom instructions or previous conversations, and each conversation starts fresh. That article is not recent, so treat it as likely rather than certain. Recognize it by the GPT's name at the top of the chat. Fix: ask your question in a normal chat, or put what the GPT needs into its own instructions.

## Check your plan, region and rollout

If the switches are missing or greyed out, the cause may be availability rather than a setting you broke.

OpenAI rolled chat-history referencing out in stages during 2025. Release notes from that year say it reached Plus and Pro first, was off by default and enabled manually in the EEA, UK, Switzerland, Norway, Iceland and Liechtenstein, and later reached Free users in a lighter form. Whether those regional rules still apply today we could not confirm, and the plan table on the [pricing page](https://openai.com/chatgpt/pricing/) is the place to check. Recognize it by a missing toggle in Settings. Fix: confirm your plan and region on that page, then update the app and look again.

> [!NOTE]
> OpenAI changes memory often. Its [release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) mention an upgraded memory in June 2026 and, in August 2026, project-only memory and a setting for Temporary Chats. If a label in this post does not match your screen, the notes are the quickest way to see what moved.

## Look at your Project's memory scope

Projects can narrow memory. Per OpenAI's [Projects article](https://help.openai.com/en/articles/10169521-using-projects-in-chatgpt), each project has a memory setting, Default or Project-only, and it respects your personal and workspace memory settings. Shared projects are always project-only.

Recognize it by ChatGPT knowing your facts in ordinary chats but not inside one project, or the reverse: facts from the project never showing up elsewhere. That is project-only memory doing its job. Fix: open the project's settings and choose the scope you want, or move the conversation out of the project. [How memory works in ChatGPT Projects and Claude Projects](https://esment.notas.ai/blog/chatgpt-claude-projects-memory) explains the scopes.

## If you are on Business, Enterprise or Edu, ask your admin

On a workspace plan, memory has an extra layer. OpenAI's Projects article says project-only memory is available when memory is enabled for both the workspace and your account, and that a workspace-level off switch disables it in every project. Enterprise members also need Reference saved memories on in their personal settings.

Where admins find the control has changed over time, and OpenAI's own Enterprise pages do not agree on one path, so we do not give a menu route here. Recognize it by a toggle that is greyed out or an account that never saves anything while a personal account works. Fix: ask your admin to check the workspace memory settings, and read [OpenAI's Business memory FAQ](https://help.openai.com/en/articles/9295112-memory-faq-business-version) together.

> [!WARNING]
> Ask before anyone switches memory off for a whole workspace. OpenAI's wording on whether members' existing saved memories are deleted at that point is something we could not verify.

## Separate "memory full" from "Chat memory full"

These sound alike and are not. A full saved-memory list means ChatGPT has too many stored facts. A "Chat memory full" or "Context window full" notice, per OpenAI's [help article](https://help.openai.com/en/articles/20001639-why-does-chatgpt-say-context-window-full-or-chat-memory-full), is about one conversation getting too long, and deleting saved memories will not clear it.

Recognize the second one by it appearing mid-conversation, after many messages. Fix: start a new chat, with a short summary of where you left off pasted at the top. For the saved-list case and how the two limits differ, see [ChatGPT memory full: the two limits](https://esment.notas.ai/learn/chatgpt-memory-full). OpenAI added automatic management in October 2025 that moves less relevant memories to the background, which reduces how often the saved list fills. If the list itself is the problem, [a repeatable memory cleanup routine](https://esment.notas.ai/blog/chatgpt-memory-full-cleanup) helps.

## Update, sign out and test properly

If every setting looks right, do the general fixes: update the app, sign out and back in, and compare the web and mobile apps. These are ordinary troubleshooting, not documented OpenAI steps, and we found no official source on app-versus-web differences.

Then test the same way each time:

> [!STEPS]
> 1. In a normal chat, state one harmless fact: "I work as a product designer in Lisbon."
> 2. Check for **Memory updated**, or look for the fact in your saved memories list.
> 3. Open a second normal chat and ask "Where do I work?"
> 4. Repeat on the web and in the app, and note which one fails.

If the fact is saved but ChatGPT does not use it, say so directly: "Use my saved memories and tell me what you know about my work." Several forum threads report saved memories that are recalled but not applied. Those reports are anecdotal, and OpenAI has not confirmed them as a bug.

## The whole list in one table

| Symptom | Likely cause | Fix |
|---|---|---|
| No "Memory updated" ever | Temporary Chat, custom GPT, or Reference saved memories off | Use a normal chat; turn the switch on |
| Chat history not used, saved facts fine | Reference chat history off | Turn it on in Settings › Personalization |
| Both switches off after you changed one | Saved memories off also turns off chat history | Turn saved memories back on first |
| Knows you in some chats, not in a project | Project-only memory scope | Change the project's memory setting |
| Toggle missing or greyed out | Plan, region, rollout or admin setting | Check your plan; ask the workspace admin |
| "Chat memory full" mid-chat | Conversation too long | Start a new chat with a short summary |
| Saved but not applied | Retrieval is not guaranteed (anecdotal) | Ask it by name to use your memories |

## When to contact support

Write to OpenAI when your settings are right, you are in a normal chat on a supported plan, a new fact never reaches the saved list, and signing out and updating the app change nothing. Send your plan, platform (web, iOS, Android, desktop) and a screenshot of your memory settings.

Forum threads suggest support does not always recognize regressions after a rollout, so a clear reproduction from the test above helps.

> [!ESMENT]
> If memory is switched off by your plan, region or workspace admin, or you keep landing in custom GPTs, a memory kept outside ChatGPT does not depend on those switches. Esment is one such memory: it does not repair OpenAI's own, but it can [import your ChatGPT memory](https://esment.notas.ai/docs/memories/capture) and hold it for any assistant you connect. See [how Esment works](https://esment.notas.ai/learn/how-esment-works).

For the bigger picture of why a single assistant's memory keeps being this fragile, read [why AI assistants need a memory they can't lose](https://esment.notas.ai/blog/why-ai-assistants-need-memory) and [what local-first memory means](https://esment.notas.ai/blog/local-first-ai-memory).

## Questions, answered

### Why is ChatGPT memory not saving anything?

Usually one of four things. Reference saved memories is off, you are in a Temporary Chat, you are inside a custom GPT, or a workspace admin turned memory off. Check your settings, then test in a normal chat with one harmless fact.

### What is the difference between saved memories and chat history?

Saved memories are specific facts ChatGPT keeps on their own. Reference chat history lets it draw on past conversations. Turning off saved memories also turns off chat history, but turning off chat history leaves saved memories alone.

### Does a Temporary Chat use memory?

It does not create or update memories. Recent release notes suggest there is a setting for whether it can read existing ones, but check that in your own app before relying on it.

### Does memory work in custom GPTs?

OpenAI's help article says custom GPTs do not use saved memory, custom instructions or previous conversations, and each conversation starts fresh. That article is not recent, so test it. If you need continuity, use a normal chat or a Project.

### Is "Chat memory full" the same as a full memory list?

No. The "Chat memory full" or "Context window full" notice is about one conversation getting too long. Deleting saved memories will not clear it. Start a new chat instead.

### When should I contact OpenAI support?

When your settings are right, you are in a normal chat on a supported plan, and a new fact still never shows up in your memories after you sign out and update the app. Send your plan, platform and a screenshot of your settings.
