On June 4, 2026, OpenAI began rolling out a new memory system for ChatGPT, built on a background process it calls “dreaming.” It reached Plus and Pro users in the U.S. first, and OpenAI said it would expand to more countries and to Free and Go users “over the coming weeks” (OpenAI).

For business owners who are tired of retyping project context every morning, this is a real change. It also means ChatGPT now decides on its own much of what it remembers about you and your clients. That makes understanding the controls part of the job.

Key Takeaway

What happened: ChatGPT’s upgraded memory now learns from your past conversations in the background and keeps that context current, without you asking it to remember anything.

For your business: Less re-explaining and better continuity on long projects. It also means more client and business detail stored automatically, so set boundaries before you rely on it.

For your family: If your teen uses ChatGPT, its memory is building a profile of them too. Linked parent accounts can manage teen memory settings.

What “dreaming” actually does

Dreaming isn’t brand new. OpenAI says it introduced the first version in April 2025 as “a method for ChatGPT to automatically curate memories in the background by referencing chat history.” The June 2026 release, which OpenAI calls Dreaming V3, is “a significantly more capable and compute-efficient memory architecture built on top of dreaming” (OpenAI).

In practice:

• It keeps context current. OpenAI says memory should “account for the passage of time.” Its example is a birthday party planned for “next Saturday,” which becomes outdated once Sunday arrives (OpenAI).
• It holds more. OpenAI is doubling memory capacity for Plus and Pro users. If you prefer the old system, “you can switch back in settings” (9to5Mac).
• You get a summary, not a full record. You can review and edit a memory summary page (OpenAI). But OpenAI’s help center says the summary “does not necessarily include every detail or source that Memory can reference” (OpenAI Help Center).

The workflow shift

The old routine was to paste your business background, project status, and style preferences into every new chat. With memory on, you can often skip that. OpenAI’s own framing is that memory lets “future conversations to start from shared context rather than from scratch” (OpenAI).

That matters most for strategic work that spans weeks, where re-explaining context used to be the bottleneck. Three uses worth trying (our suggestions, not OpenAI’s):

1. A standing project thread. Keep one project’s discussion in a ChatGPT Project. Update it weekly and ask it to challenge your assumptions.
2. A decision journal. Record major decisions (pricing, hires, partnerships) with your reasoning. Ask ChatGPT to flag when a new decision conflicts with an earlier one.
3. A goals check-in. Review progress against quarterly goals, with context carried forward.

The risks to manage

ChatGPT decides what to remember. Researchers analyzed 2,050 memory entries from 80 real ChatGPT users and found “a striking 96% of memories in our dataset are created unilaterally by the conversational system.” 28% contained personal data as defined by Europe’s GDPR, and 52% contained psychological insights about the user (arXiv). The study predates the June upgrade, but it shows how much memory captures that users never asked it to keep.

Your plan determines how your data is used. By default, OpenAI does not use content from ChatGPT Business, Enterprise, or Edu workspaces to train its models. Personal accounts, including Plus and Pro, may be used for training when “Improve the model for everyone” is on (OpenAI Help Center). Pro is a personal plan. It doesn’t get Business-level data protection.

Deleting takes more than one step. Deleting a chat “does not necessarily delete a separate saved memory created from that chat.” OpenAI also “may retain logs of deleted saved memories for up to 30 days for safety and debugging purposes” (OpenAI Help Center).

Vendor lock-in. The more context lives inside one assistant, the harder it is to switch. Keep your key project plans and decisions in tools you own, like Google Drive, Notion, or your CRM. Anthropic, for comparison, publishes instructions for importing and exporting memory in Claude (Claude Help Center).

How to set up memory deliberately

1. Find your controls. Go to Settings, then Personalization, then Memory. Depending on your plan, you’ll see options such as Reference saved memories, Reference chat history, and Memory summary (OpenAI Help Center).
2. Use temporary chat for sensitive work. Non-personalized temporary chats “do not create or update memories, even with personalization on” (OpenAI Help Center). This is more dependable than telling ChatGPT “don’t remember this.”
3. Wall off client projects. With project-only memory, ChatGPT uses conversations within that project but “won’t reference memories or conversations from outside the project,” and keeps project information out of your other chats (OpenAI Help Center).
4. Brief it once. Give ChatGPT a short overview of your business, clients, and style, then check that the memory summary reflects it accurately.
5. Audit monthly. Review the summary, correct anything wrong, and remember it may not show everything.

On a Business plan, workspace owners can turn memory off for the entire workspace. Doing so deletes members’ existing saved memories in that workspace (OpenAI Help Center).

For Your Business

• This week, open your Memory summary and read what ChatGPT thinks it knows about your business. Correct anything wrong.
• Move one client engagement into a project with project-only memory.
• Run your next conversation involving confidential client data in a non-personalized temporary chat.

For Your Kids

• ChatGPT requires users to be at least 13, with a parent’s permission under 18 (OpenAI). For a linked teen account, “a parent or guardian can manage supported memory controls through Parental controls” (OpenAI Help Center).
• Sit down with your teen and read their memory summary together. Talk about what they’d rather an AI not know.

What to watch

• Claude’s alternative. On August 25, 2026, Anthropic made Claude’s memory visible “topic by topic,” with every item editable. It also excludes sensitive subjects such as health or beliefs by default (Anthropic). Compare how much each assistant shows you.
• Regulators and researchers. The GDPR-focused memory study has drawn attention (Futurism). Watch for European regulators to weigh in.
• Rollout to Free users and other countries. Check whether dreaming is active on your team members’ accounts.

FAQ

Can I see everything ChatGPT remembers about me?

Not entirely. OpenAI says the memory summary is a high-level view that may omit some details or sources.

Does ChatGPT train on my business memories?

Not by default on Business, Enterprise, or Edu plans. On personal plans, including Plus and Pro, it can, unless you turn off “Improve the model for everyone.”

How do I keep a conversation out of memory?

Use a non-personalized temporary chat. It doesn’t create or update memories.

Sources

• OpenAI, “Dreaming: Better memory for a more helpful ChatGPT” (June 4, 2026), https://openai.com/index/chatgpt-memory-dreaming/

• 9to5Mac (Zac Hall), “OpenAI says ChatGPT’s memory feature is getting smarter and coming to free users” (June 4, 2026), https://9to5mac.com/2026/06/04/openai-says-chatgpts-memory-feature-is-getting-smarter-and-coming-to-free-users/

• OpenAI Help Center, “Memory in ChatGPT” (accessed October 28, 2026), https://help.openai.com/en/articles/8590148-memory-faq

• OpenAI Help Center, “ChatGPT release notes” (accessed October 28, 2026), https://help.openai.com/en/articles/6825453-chatgpt-release-notes

• OpenAI Help Center, “Memory FAQ (Business Version)” (accessed October 28, 2026), https://help.openai.com/en/articles/9295112-memory-faq-team-version

• arXiv (Abhisek Dash et al.), “The Algorithmic Self-Portrait: Deconstructing Memory in ChatGPT” (February 2026, ACM Web Conference 2026), https://arxiv.org/abs/2602.01450

• Futurism (Maggie Harrison Dupré), “ChatGPT Is Building Detailed Psychological Profiles of Its Users” (October 25, 2026), https://futurism.com/artificial-intelligence/chatgpt-building-profiles-users

• Anthropic, “Claude’s memory works everywhere, and you decide what’s in it” (August 25, 2026), https://claude.com/resources/articles/claudes-memory-works-everywhere-and-you-decide-whats-in-it

• OpenAI, “Terms of Use” (accessed October 28, 2026), https://openai.com/policies/terms-of-use/

• Claude Help Center, “Release notes” (accessed October 28, 2026), https://support.claude.com/en/articles/12138966-release-notes