With ChatGPT Pages, Do I Still Need Notion?

At DevDay on 29 September 2026, OpenAI introduced Space and Pages, giving ChatGPT a workspace for files and editable documents. It also named Notion as a partner for Sign in with ChatGPT. ⁸ For people who keep their knowledge in Notion and do their thinking in ChatGPT, these changes put the arrangement itself in question.

Notion is bringing an existing ChatGPT subscription into its workspace. ChatGPT is adding somewhere to keep the work a conversation produces. For a user, the question is what each workspace still contributes.

ChatGPT Pages is a credible candidate for narrative knowledge: notes, research and explanations. It is not a demonstrated replacement for operational records or Notion database relationships, and archive storage and recovery still need checking. My next step would be a small knowledge-work trial, rather than moving the collection.

The job I gave Notion has changed

I started using Notion as a notebook: somewhere to put notes, articles and reference material. Then I discovered databases. Pages could have properties, connect to other records and appear in different views. I could build a system around the way I worked without building the software underneath it.

The structure had a purpose. Relationships helped me navigate, and views made a growing collection manageable. I was building more than a place to save things.

More recently, that collection has become context for AI: documentation, research, business rules, project history and the reasoning behind decisions. I still want to read it, revisit it and share it. But increasingly, I keep it so I can ask questions, work through a problem and create something new.

That changed what I needed from the knowledge base. “Notion holds the knowledge; ChatGPT helps me think” used to be a sufficient explanation for the split. Now ChatGPT can also hold a document. Sending the result to another application needs to serve a purpose beyond keeping the old arrangement intact.

The same knowledge, approached from two directions

Notion lets members of Business and Enterprise workspaces connect their own ChatGPT Plus or Pro subscription to Notion Agent. Selecting a specific GPT model uses that ChatGPT allowance; Auto and other models use Notion credits. Matching verified email addresses are required, and the connection does not apply to Custom Agents. ¹

If your knowledge is already organised there, that is a reason to stay. An existing AI subscription can work inside the system rather than requiring you to move its contents.

ChatGPT's Space starts at the other end. It groups pages and files; Pages are editable documents with subpages and collaboration. A conversation's useful output can have a continuing home in the same application. ²

This is still a launch, with limits. As checked on 30 September 2026, availability is rolling out to Pro, Business and Enterprise. Editing works on web and desktop; mobile supports reading and sharing. Slides and Sheets are coming soon, automatic page updates are unavailable, and existing ChatGPT Projects remain separate. ²

The choice is becoming less about which application has AI. It is about where you want the work to live, and which structure it needs when it gets there.

Can ChatGPT Pages replace Notion?

Some of the relationships I created mainly helped me find things: an article belonged to a project, research supported an idea, a note about an integration belonged under AI. A language model can often recognise connections in the content itself. If it can retrieve the right sources, some manual linking may no longer earn the effort it takes.

Other relationships do a more exacting job. Notion's relation properties explicitly connect database records. A project can link to its client, and a rollup can summarise values from related records. ⁴ You can inspect which records are connected, rather than rely on an AI interpreting their text. If a project is assigned to the wrong client, its work can appear in the wrong account summary. That link has consequences beyond helping someone find a page.

Two illustrative questions from measurement work show why the distinction matters. “How does our restaurant booking tracking work?” needs an explanation: the journey, event meanings, limitations and decisions. A well-written page can hold that knowledge.

“Which production events have not been validated in the past 90 days?” needs a reliable selection of records. Event names, environments, validation dates and statuses must be explicit. An answer that selects the wrong events is wrong, however convincing the explanation sounds.

The first question asks a knowledge base to explain. The second tests structured fields and filtering, rather than a relation between databases, but it exposes the same need for an inspectable answer. A client-project relation adds another requirement: the connections between records must also be correct.

The Pages documentation I reviewed does not establish a native equivalent to Notion's relations and rollups. ³ Database-backed tools can exist in ChatGPT; that does not demonstrate that Pages can replace an existing relational system.

My rule would be to let AI help discover connections in narrative knowledge, while keeping relationships explicit when they determine an operational answer. That is a more useful boundary than whether an application can display a table.

A document editor still has to earn the archive

Even if Pages suits the knowledge work, moving years of material is a separate decision. I would need to know what I can store, who can see it and whether I can recover it later.

OpenAI's Library documentation lists 100 GB for Pro. ⁵ Space replaces Library for accounts with access, so that figure alone does not establish a Space-specific storage entitlement. ² Notion advertises an unlimited number of paid file uploads, with a roughly 5 GB per-file limit, subject to file-format and embedding limits. ⁶ The new terms need to be clear before I treat an accumulating archive of PDFs, images and reference material as a migration candidate.

Sharing has its own boundary. Linked files in Space retain their original permissions, but information copied onto a shared page becomes visible to its readers. Training and Memory settings also need checking for the intended workspace and collaborators. A collaborator's training setting can differ from your own; Business and Enterprise data is not used for training by default. ⁷

I would also test retrieval, export and recovery. Keeping a document is one thing; finding the evidence behind an answer, preserving its qualifications and recovering the full collection are different demands. I have not established those outcomes.

Where Notion still earns its place

For a team whose records, permissions and processes are built around Notion, the new ChatGPT connection could strengthen the case for keeping it. Moving the knowledge would mean rebuilding arrangements that already serve a purpose. Bringing AI to them may be simpler.

My priorities put more weight on where new work begins. I increasingly want to research, think and create in ChatGPT, then retain the result. That makes Pages interesting even while its wider replacement case remains unproven. I do not need it to beat every Notion feature. I need to establish whether Notion's structure is still doing a job I depend on.

Test the knowledge work before moving the knowledge

I would start with one non-sensitive topic, its supporting material and questions whose answers I already know. I would compare its answers with the original material: the right source, no missing qualification and no unsupported conclusion. Then I would check that sharing exposes only the intended content and that the export retains the page and its supporting material in readable form. A wrong source or an incomplete export would be a reason to keep the original workflow. That would test the job I want it to do without making the archive itself the experiment.

If the workflow relies on records, relations and rollups, I would keep the existing system until equivalent behaviour is demonstrated. If it mainly involves reading, explaining and developing narrative knowledge, Pages is a credible candidate.

The arrival of Pages makes me look differently at the structure I built. Some of it may remain essential. Some may have been a way to navigate knowledge that I can now explore through a conversation. I want to keep the parts that still do useful work.

I used to ask how to give ChatGPT access to my knowledge base. Now I also have to ask why the knowledge needs to live somewhere else.