ChatGPT & Claude Projects: Setup, Settings and Memory Guide

Introduction

Why Projects matter

Most people start using AI the same way: one long chat, then another, then thirty more. Within a month the useful conversations are buried. The same background gets retyped, reference documents get re-uploaded, and nobody on the team can find last week's draft.

Projects address that organization problem. In both ChatGPT and Claude, a Project is a workspace that groups related chats with the instructions and reference material those chats should share. OpenAI describes projects as keeping “related chats, files, and instructions together.” Anthropic describes them as “self-contained workspaces with their own chat histories and knowledge bases.”

This guide shows how to set up Projects in both products, how each handles instructions, files, memory and sharing, and where they differ. They are not interchangeable. Several differences matter for business use: which plans can share, whether a Project can be archived, and how memory crosses, or does not cross, Project boundaries. For the wider product map, see ChatGPT and Claude ecosystems explained.

Comparison

ChatGPT Projects vs. Claude Projects: quick comparison

As of October 14, 2026. Plan limits change; confirm them in the vendor pages listed in the references. For a broader product comparison, see Claude vs ChatGPT.

Fundamentals

What is an AI Project?

The simplest model:

A Project called “Work | Daily Operations” holds many chats, such as “Draft Friday client update” or “Summarize Monday standup notes”, each handling one task. The Project can also carry instructions that apply to every chat inside it and reference material the chats can draw on.

Two cautions. First, organization is not the same as unlimited shared context. Anthropic states that “context is not shared across chats within a project unless the information is added into the project knowledge base” (memory is a separate mechanism, covered below). In ChatGPT, instructions and files are shared, and memory behavior depends on settings. Second, Projects do not change the underlying model. They change what context the model starts with.

Tutorial

Setting up Projects in ChatGPT

Steps reflect OpenAI's help documentation. Anything not confirmed there is marked for a live check.

1. Create the Project

1. In the sidebar, select New project.
2. Enter a project name.
3. Select an icon and a color.

OpenAI also documents a memory choice (Default or Project-only) at creation. This detail was not verified in a live account and may vary by plan or workspace: where this choice appears in the creation flow for each plan.

2. Rename or change appearance later

OpenAI documents that owners of shared projects can modify the project name. This detail was not verified in a live account and may vary by plan or workspace: how to rename a Project, and whether icon and color can be edited after creation; the help article does not say.

3. Open Project settings and add instructions

Select the more options menu (•••), then Project settings. OpenAI's example instruction: “Act like my marketing mentor. Be concise. Use bullet points. Ask clarifying questions.”

Project instructions “apply only within that project and override your global custom instructions.” OpenAI's Enterprise and Edu context table likewise lists custom instructions as “Not available (only project instructions)” for chats inside projects. No instruction-length limit is documented in the help article reviewed.

4. Add reference material

Upload PDFs, spreadsheets, documents or images, or paste text. After adding a file you can open it from the sources list to preview, download or delete it. Limits per Project:

• Free: 5 files
• Go and Plus: 25 files
• Edu, Pro, Business and Enterprise: 40 files
• Only 10 files can be uploaded at once.

You can also paste links to supported Google Drive files and folders or Slack channels as sources, connecting the app if prompted. OpenAI notes that Google Drive content added within a project is not synced in advance for retrieval.

If ChatGPT produces something you want to reuse, such as a decision note or summary, you can save that response as a project source. OpenAI lists the menu action as “Save to project / Add to project sources (label may vary).” This detail was not verified in a live account and may vary by plan or workspace: exact label.

5. Organize conversations

Start new chats inside the Project, or move an existing chat by dragging it onto the Project or choosing Move to project from its menu. Moved chats inherit the Project's instructions and file context. Use Remove from project, when shown, to move a chat back. Chats created with a GPT cannot be moved in. You also cannot add a temporary chat to a Project; saving it converts it to a regular chat first.

6. Pinning and Project home

This detail was not verified in a live account and may vary by plan or workspace: whether Projects can be pinned in the sidebar, and what the Project's main page is called; OpenAI's help article documents neither.

Tutorial

Setting up Projects in Claude

Steps reflect Anthropic's help documentation as of October 14, 2026. Anthropic is rolling out a new Projects beta, starting with Claude Code and select Pro and Max users. This guide covers the current chat version; your interface may differ if your account has been moved.

1. Create the Project

1. Open Projects from the left panel, or go to claude.ai/projects.
2. Click + New Project (upper right).
3. Enter a name and description. Anthropic states that Claude cannot see these details, so do not put instructions there.
4. On Team or Enterprise, choose visibility: Keep it private or Share with your broader organization, where enabled by administrators. This detail was not verified in a live account and may vary by plan or workspace: Anthropic's visibility article uses Public/Private wording; confirm current labels.

No icon or color options are documented for Claude Projects.

2. Add project instructions

Click Set project instructions, write your guidance, and click Save instructions. Claude uses them for all chats in the Project. No length limit is documented in the pages reviewed, and Anthropic does not state how project instructions interact with account-level preferences.

3. Add project knowledge

On the right side of the Project's main page, click + to add documents, text files or code snippets. Anything added is available in every chat in the Project.

Limits from Anthropic's upload article: 30MB per file; unlimited file count, but total content must fit Claude's context window; text extraction only (except multimodal PDFs). On paid plans (Pro, Max, Team, Enterprise), when knowledge approaches the context limit Claude automatically enables RAG mode, which Anthropic says can expand capacity “by up to 10x.” A visual indicator shows when a Project is RAG-enabled, and Claude then uses a project knowledge search tool to retrieve relevant material instead of loading everything at once (source).

4. Organize conversations

Move a standalone chat in with the dropdown arrow next to its name → Add to project, then choose the Project in the modal. The same menu removes a chat from a Project or moves it between Projects.

5. Manage the workspace

• Star a Project from its three-dot menu on the Projects page or with the star icon inside the Project; starred Projects appear in the left panel.
• Archive from the three-dot menu in the Project and confirm. Archived Projects sit at the bottom of the list and in an archived tab; their conversations stay accessible.
• Rename: This detail was not verified in a live account and may vary by plan or workspace: Anthropic's article says it does not cover renaming.

Organization

Naming conventions and Project organization

This section is an AIBX recommendation, not vendor documentation.

Use a consistent pattern: Category | Project Name

• Personal | Morning Routine
• Personal | Getting Ready for Work
• Work | Start of Day
• Work | Daily Operations
• Work | End of Day
• Business | Client Management
• Research | AI Tools

Both products show Projects as plain lists, so a shared prefix keeps related Projects together and makes them easier to scan.

Keep scope focused. A Project whose chats share a purpose can use one set of instructions and one set of files well. A “General | Random Questions” Project is optional and may be counterproductive: unrelated conversations then share instructions and, depending on memory settings, context that does not fit them. If you want a catch-all, consider leaving unrelated chats outside Projects.

Instructions

Project instructions and maintaining a master template

Project instructions tell the model how to behave for every chat in that Project. Good instructions cover role, purpose, relevant context, response style, accuracy requirements, output format and clarification rules.

Illustrative example for Work | Daily Operations:

Role: You are an operations assistant for a small professional-services firm.
Purpose: Help me plan the day, draft internal communications, and document recurring procedures.
Context: Our team is ten people. We use a weekly status email and a shared task list.
Style: Plain language, concise, bullet points where useful. No marketing tone.
Accuracy: Do not invent facts about our clients or policies. If information is missing, say so and list what you need.
Output: Lead with a one-sentence summary, then actions with owners and dates.
Clarification: If the request is ambiguous, ask up to three questions before drafting.

Keep a master instructions file

Store your instructions in a file you control rather than only inside the product. Preferred formats are Markdown (.md) or plain text (.txt). A Word document (.docx) works if you prefer editing there, but it is not required. Use version names such as work-operations-instructions-v1.md, then follow a loop:

Paste the draft into the Project's instructions, run a few realistic chats, check whether the output follows the rules, then revise and bump the version.

Memory

Understanding Project memory and context

Several mechanisms can bring earlier information into a chat. They are different things, and the two products handle them differently.

ChatGPT

OpenAI documents two memory settings per Project (Projects, Memory):

• Default memory: behavior depends on plan. On Enterprise and Edu, chats stay contained within the Project. On other plans, including Business, chats can reference both Project and non-Project conversations unless the other Project is set to Project-only.
• Project-only memory: your previously saved memories are not referenced in Project chats. Chats can reference other conversations in the same Project but not outside it, and chats outside cannot reference the Project.

You choose when creating the Project and can change it in ••• → Project settings → Memory → Save. OpenAI says changes “may take a few hours to take effect.” Switching to Project-only removes that Project's information from memory used outside it. On most plans, Reference saved memories and Reference chat history must be on; on Enterprise, Reference saved memories plus workspace memory. There is no global setting to make all Projects Project-only, and no list of Project memories. This detail was not verified in a live account and may vary by plan or workspace: current setting labels; OpenAI notes they vary by account.

Two points matter for teams. Shared Projects are automatically Project-only and cannot be switched back to default, even after all collaborators are removed. And memory is not exhaustive: OpenAI says it “does not retain every detail from every conversation.”

Claude

Anthropic states that each Project has its own memory (create and manage, chat search and memory). While you work in a Project, Claude uses that Project's memory alongside your memory from chats outside Projects, but not memory from other Projects. New memory is saved to the Project and, if memory is on, Claude can reference it in chats outside Projects too. To keep a Project fully separate, turn off Use account memory on the Project's page.

Memory is on by default for Free, Pro and Max. On Team and Enterprise an owner must enable it, and members then turn it on. Manage it in Settings → Memory, where Topics can be viewed, edited or deleted, and memory can be paused or reset. Resetting permanently deletes all memories including Project memories. Chat search is a paid-plan feature and is limited to non-Project chats or to within one Project.

To keep one chat out of memory, turn off Memory in the + menu before sending the first message. That chat does not use or add to Project memory but stays in your history. The setting cannot be changed once the chat starts. Anthropic also says some information is never saved to memory, including government ID numbers and financial account numbers, and that sensitive topics are excluded by default unless you opt in. Treat this as product behavior, not a confidentiality guarantee.

Temporary and incognito chats

• ChatGPT temporary chats are not saved to history, do not create or update memories, and are not used for training while they remain temporary. OpenAI may keep a copy for up to 30 days for safety. You cannot add a temporary chat to a Project (source).
• Claude incognito chats are not saved to history or memory, are not used for training, and are retained for 30 days by default. They are available only outside Projects, so the ghost icon does not appear inside one. Incognito chats can still see your profile preferences (source).

Files

Files, libraries and knowledge bases

ChatGPT Project sources are files attached to one Project, capped by plan (5, 25 or 40 files). ChatGPT Library is a separate store for files you upload to or create in ChatGPT, with its own storage limits: Free 500MB, Go 4GB, Plus and Business 20GB, Pro 100GB (source). You can add a Library file to a chat through the composer menu. Library files are not deleted when a Project is deleted; files stored only in the Project are. OpenAI's help article does not describe how ChatGPT retrieves from Project sources. This detail was not verified in a live account and may vary by plan or workspace: per-file size limits for Project sources specifically; OpenAI lists 512MB per file for conversation and GPT uploads, with 2M-token and spreadsheet caps.

Claude project knowledge has a 30MB per-file limit and is bound by the context window until RAG engages on paid plans. Supported documents include PDF, DOCX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON and XLSX (XLSX requires code execution and file creation to be enabled); images include JPEG, PNG, GIF and WebP. For project files, Anthropic says text extraction applies, except for multimodal PDFs.

Business use

Sharing, privacy and business use

What the vendors document

ChatGPT. Project sharing is available on Free, Go, Plus, Pro, Business, Enterprise and Edu. Invitees get chat access (see and interact with chats, files and instructions) or edit access (update instructions, upload or remove files, invite others but not remove members). Owners can remove members and change permissions. Members can view the Project's chats, files and member list (names and email addresses), and all members can download files. Owners can choose “Only those invited” or “Anyone with a link,” in which case any logged-in ChatGPT user with the link can join. Collaborator limits are plan-based (Free 5, Go and Plus 10, Pro 100; workspace Projects up to 100). Sharing a single chat from a private Project shares only that chat. Workspace owners can control who can create or share Projects (source).

Claude. Sharing is limited to Team and Enterprise plans, with Can view and Can edit roles. Members can see contents, knowledge and instructions and chat within the Project. Chats inside a shared Project remain private unless their author shares them. Admins can disable sharing, and archiving does not remove members or reset permissions (source). Anthropic's pages describe who can remove members differently. This detail was not verified in a live account and may vary by plan or workspace: which roles can remove access.

Training, retention and deletion

• ChatGPT personal plans: content may be used to improve models if “Improve the model for everyone” is on. For a shared Project, that applies only if the owner and every contributor have it on. By default OpenAI does not use content from Business, Enterprise, Edu or Healthcare workspaces for training (data controls).
• ChatGPT retention: deleted chats leave your view immediately and are scheduled for permanent deletion within 30 days, with security and legal exceptions, and cannot be restored. Files added to a Project, including a shared Project, are kept until the Project or file is deleted, then scheduled for deletion within 30 days (source).
• Claude consumer plans (Free, Pro, Max): chats may be used if you choose to allow it, if a conversation is flagged for safety review, or if you opt in explicitly, for example through a trusted tester program. Incognito chats are excluded. Thumbs up/down feedback can be stored for up to 5 years (source). Deleting a chat removes it from history immediately and from backend storage within 30 days; data you allow for training can be kept de-identified for up to 5 years (source).
• Claude commercial plans (Team, Enterprise, API): by default Anthropic does not use inputs or outputs to train its models, though feedback you submit or data you choose to allow can be used (source). Deleted conversations are removed from history immediately and from backend storage within 30 days (source).
• Memory and deletion: deleting a chat does not necessarily delete memory derived from it, in either product. OpenAI says deleting a chat alone “does not necessarily delete a separate saved memory,” and Anthropic says memory entries from a deleted conversation are not removed, though individual memories can be deleted at any time.

Maintenance

Managing Projects over time

• Rename and reorganize. Keep the naming pattern. This detail was not verified in a live account and may vary by plan or workspace: rename path in ChatGPT and in Claude; Anthropic's article does not cover renaming.
• Move chats. ChatGPT uses Move to project when available; Claude uses Add to project or the move option. ChatGPT notes that memory settings and workspace restrictions can limit which chats can move.
• Pin or star. Claude documents Star. This detail was not verified in a live account and may vary by plan or workspace: whether ChatGPT Projects can be pinned.
• Update instructions from your versioned master file, then retest.
• Remove outdated knowledge. In ChatGPT, open the file from the sources list and delete it; in a shared Project this removes it for everyone. In Claude, content can be added and removed at any time.
• Archive (Claude only). Archiving declutters finished work while keeping members and knowledge. Archived Projects must be unarchived before they can be deleted. ChatGPT has no documented Project archive, so rely on clear naming and delete only after copying out what you need.
• Delete. In ChatGPT, deleting a Project “permanently removes its chats and instructions, along with files stored only in the project.” Library files are kept, members lose access, and it cannot be undone. In Claude the steps are documented (••• → Delete → Yes, delete), but Anthropic's pages reviewed do not say what is removed or whether it is reversible. Treat it as permanent and export anything you need first.

Examples

Four practical Project examples

All examples are illustrative. They are not results from real deployments.

Efficiency

Do Projects actually save time or tokens?

Projects can reduce repetitive prompting and improve consistency, because instructions and reference files travel with every chat. Those are organizational benefits. Token and cost savings are a different matter, and neither vendor documents them as a feature:

• OpenAI states that project access carries no extra cost and that “any credits or usage charges… apply to project chats the same as they do to regular chats.”
• Anthropic states that projects draw from your plan's usage like any other use of Claude. That sentence appears in the section on its new Projects beta, so confirm it for your interface (source). Large project knowledge can also increase what Claude processes, and RAG changes how it is retrieved.

Do not assume Projects lower your usage; long instructions and many files could add to it. AIBX recommends measuring your own workflow rather than relying on a general claim, and this guide makes no numerical savings claims. For usage habits specific to Claude, see Stop Burning Through Claude Tokens.

Checklist

Final setup checklist

• Confirm you are in the web version and note your plan.
• Choose a naming pattern (Category | Project Name).
• Create a Project with a focused scope.
• Write instructions in a versioned .md or .txt master file.
• Paste instructions into the Project and test in a new chat.
• Add only current, relevant reference files with clear filenames.
• Decide memory behavior (ChatGPT Default vs Project-only; Claude Use account memory).
• Review sharing permissions and who can see files.
• Check your plan's data controls before adding sensitive material.
• Schedule a periodic review: update instructions, remove stale files, archive (Claude) or clean up finished Projects.

FAQ

Frequently asked questions

Conclusion

Better workspaces, not smarter models

Projects improve the process of using AI, not the underlying intelligence of the model. They give related work a home, make instructions repeatable and keep reference material close at hand. The two products organize this differently: ChatGPT leans on per-Project memory modes and broad sharing availability, while Claude offers archiving, per-Project memory with an account-memory toggle, and sharing on business plans.

Name Projects deliberately, keep scopes narrow, maintain your instructions in a versioned file, load only relevant knowledge, and review everything periodically. Before putting business data in a shared Project, read the current vendor documentation for your plan. Wider context on the platforms is in Top AI Chat Platforms 2026.