ChatGPT Projects

ChatGPT Projects let you keep a client’s brief, reporting rules, files and working conversations together, so the next report starts with useful context already in place. For an operator repeating the same job, the payoff is less time rebuilding the brief and more time checking the work.

Verified against OpenAI’s live help pages on November 1, 2026.

Start with one recurring deliverable. A Project called “Northstar Weekly Client Report” is easier to maintain than a bucket called “Work.” Northstar is the example client throughout this guide.

When ChatGPT Projects Beat a Plain Chat or Custom GPT

Choose a Project when the job stays the same while the evidence and conversations change.

Think of it as a client workroom: the files are the reference shelf, the instructions are the working rules, and the chats are the conversations around the table. Project memory lets relevant context from that room inform later work. OpenAI describes Projects as a place for related chats, files and instructions. OpenAI’s Projects guide.

A custom GPT is a version of ChatGPT configured for a purpose. Its conversations do not use saved memory or previous conversations. Personal Free, Go, Plus and Pro accounts cannot create or publish new GPTs; Business and Edu availability remains subject to product and workspace permissions. For a new recurring client workflow, my recommendation is to start with a Project. OpenAI’s GPT guide.

Since October 26, 2026, new GPT creation has ended for affected Enterprise workspaces, except FedRAMP. OpenAI’s current FAQ schedules standard retirement for December 11, 2026, with February 11, 2027 reserved for FedRAMP and affected Enterprise workspaces with a qualifying, approved deferral. Follow any different notice for the account or workspace where the GPT was created. The creation cutoff and retirement are separate; existing GPTs remain usable until their applicable retirement date. See OpenAI’s FAQ and our full custom GPT retirement timeline.

Moving a simple custom GPT into a Project

A Project is a practical replacement when the GPT’s useful work comes from instructions and reference files for an ongoing job. For Northstar, that might be a report-writing assistant whose job is to follow the account brief and approved report format. Rebuild that setup manually using the documented Project controls; OpenAI’s built-in migration route goes to a plugin.

Carry Over the Instructions and Reference Material

1. Preserve the working instructions. If you maintain the GPT, copy its instruction text into your own working document. Keep its rules about audience, evidence, format and review; remove directions that depend on tools you have not replaced.
2. Gather the maintained source files. Use the original reference documents and approved examples. Upload them to the Project within its file allowance. This recreates the reference collection; it is not an automatic transfer of the GPT’s knowledge.
3. Paste and adapt the instructions. In the Project’s more options menu, open Project settings. Make the rules specific to this client or deliverable, choose the memory boundary, and start a new chat inside the Project.
4. Compare the work before switching. Try a familiar report request and a case with missing evidence. Check source use, calculations, format and whether it asks for clarification. Then invite a colleague and check that they can complete the same job.

These are suggested rebuild steps using OpenAI’s Project instructions and file controls. A Project’s ability to draw on earlier Project chats changes the context available to it, so copied instructions alone do not establish identical behavior.

What Does Not Come Across Automatically

GPT conversations cannot be moved into a Project. Start fresh and add any approved decisions or summaries as source material. Existing GPT conversations remain accessible after retirement, but they do not become Project history. Project chat restrictions, conversation retention.

Conversation starters need a deliberate replacement. OpenAI describes them as example prompts in the GPT configuration; its current Projects guide does not document an import of those buttons. Keep useful starter text in a short source document or the working instructions and reuse it as a prompt. That is a suggested workaround, not a confirmed starter-button migration. GPT configuration, Project setup.

Custom actions need separate integration work. Copying instructions and files does not recreate an API connection. Even OpenAI’s plugin migration does not transfer custom actions. Check whether an available app performs the required operation; otherwise a custom MCP server, a connection exposing tools to the assistant, may be needed. Test the operation separately. OpenAI’s action-migration guidance.

Review sharing afresh. Use the Project’s Share controls to invite the intended people. A shared Project exposes its chats and files to members, so moving an assistant’s instructions into a shared workroom also requires a decision about which working material belongs there. Project visibility.

When OpenAI’s Plugin Migration Is the Better Home

Choose a plugin when the value is a reusable method across different clients or jobs, especially when it includes connected apps. A plugin can package a skill, meaning reusable workflow instructions, with reference material and apps. For example, a proposal-review method belongs in a reusable plugin; Northstar’s changing account brief and weekly discussions belong in its Project. This is a workflow recommendation based on OpenAI’s plugin capabilities.

When available in the owning account, use My GPTs → Migrate to plugin. Migration takes the latest published version: instructions become a skill, knowledge files become reference files, and connected apps are added. Unpublished edits and the selected model do not transfer. The plugin starts private, so install, test and share it deliberately. Migration makes the original GPT read-only. Access varies by account and workspace; follow the official migration steps.

Put Sources, Rules and Working History in Their Proper Places

Give the Project enough context to do its job, with a clear distinction between approved facts and unfinished discussion.

Project instructions override your global custom instructions inside that Project. Project memory can use its chats and files as context; it is not a promise that every detail will appear in every answer. OpenAI’s broader memory guidance explicitly says memory does not retain every detail. Put an important reporting rule in instructions or an approved source instead of relying on a passing remark. Projects, Memory in ChatGPT.

Set Up a Project for a Weekly Client Report

Build the smallest useful reporting room, then run an actual report through it. These controls follow OpenAI’s setup instructions; the Northstar content is a suggested workflow. OpenAI’s setup guide.

1. Create the Project. Select New project in the sidebar. Enter “Northstar Weekly Client Report,” then choose an icon and color.
2. Add the source material. Upload the account brief, metric definitions, an approved report example and the current reporting export. Projects accept PDFs, spreadsheets, documents, images and pasted text. Keep filenames clear enough that a colleague can identify the current source.
3. Add project instructions. Open the more options menu, then Project settings. Paste the instruction below and adapt the audience and report sections.
4. Choose the memory boundary. For a separate client account, I recommend Project-only memory. In Project settings, select it under Memory, then Save. You can change an existing private Project’s setting; changes may take a few hours. Shared Projects already use this setting and cannot switch to default memory.
5. Bring over useful history. Drag an eligible chat onto the Project or select Move to project from its menu. It inherits the Project’s instructions and file context. Chats created with a GPT cannot be moved this way; if the option is missing, start a new Project chat.
6. Draft, check and preserve the result. Start the report conversation inside the Project. After reviewing the output, open the response’s message menu and choose Add to project sources. Keep only a reviewed version as the example for future work.

Use this as the project instruction:

Help prepare Northstar’s weekly client report for the account lead. Use the account brief for goals and the metric definitions for calculations. Follow the approved report’s tone and structure. Separate results, interpretation, blockers and next actions. Identify the source and reporting period behind each numerical claim. If information is missing or sources disagree, flag the gap and ask instead of inventing an answer. Treat exploratory chat as discussion, not approval. Mark drafts for review before they are sent to the client.

Then start each reporting cycle with a request like this:

Draft this week’s Northstar report using the current reporting export and the approved metric definitions. Compare it with the previous approved report where the periods and definitions are comparable. Explain the changes, list the evidence behind them, and flag missing inputs. Finish with the decisions the account lead needs to make.

The business change is specific: you stop rebuilding the account brief and report format at the start of every conversation. You still supply fresh evidence, review calculations and approve the client-facing output. Measure the time spent on preparation and review before claiming an hours-saved result.

Use Connected Apps Without Assuming Everything Is Synced

Connected apps, often called connectors, can bring outside information into a Project chat. OpenAI lists app support in Projects on Free, Go, Plus, Pro, Business, Enterprise and Edu. That does not promise every app or operation on every plan: availability also depends on the app, region, role, model, interface and workspace settings. Provider permissions still apply. Projects app support, connected-app availability.

In a Project chat, select the tools menu + to choose an app, or refer to it by name. OpenAI also documents adding Google Drive file or folder links and Slack channel links through Add sources in a private Project, connecting the app if prompted. Google Drive added this way does not sync content in advance for retrieval. For Northstar, confirm the current reporting export is accessible before asking for a report; a saved link alone does not prove the evidence is current. App links and sync behavior.

Share the Workroom With the People Doing the Job

Project sharing is available on Free, Go, Plus, Pro, Business, Enterprise and Edu, on the web and mobile apps. Workspace settings can restrict it. OpenAI’s sharing rules.

Open the Project and select Share. For the Northstar team, use Only those invited, enter the intended collaborators and assign access. Free, Go, Plus and Pro support individual invitations. Business, Enterprise and Edu also support workspace groups; managed workspaces may limit invitations to their own members.

Owners can also enable Anyone with a link, which allows logged-in ChatGPT users with the link to join where that sharing mode is available. Workspace links have their own membership restrictions and initially grant chat access. For client work, my recommendation is to start with named invitations. OpenAI’s access controls.

Everyone in the Project can view and download its files. Chat access is therefore useful working access, not a way to hide internal documents from someone. Keep the agency’s internal working Project separate from any material you intend to hand to the client.

Members can branch a conversation to explore another draft while preserving the original. They do not edit the same chat together synchronously. Agree which reviewed report is the current deliverable so alternative branches do not become competing approvals. OpenAI’s collaboration guide.

Sharing Automatically Turns On Project-Only Memory

A shared Project can draw on conversations within that Project. It cannot use a member’s personal memories, custom instructions or chat context from outside it. Conversations outside a Project using project-only memory cannot draw on its conversations either. Removing collaborators does not restore default memory. OpenAI’s memory rules.

That boundary means the account brief must actually be in the Project. Your teammate’s separate conversation about the client does not silently become shared knowledge.

Default memory behaves differently. Enterprise and Edu keep Project conversations contained, while other plans, including Business, can reference conversations across Project boundaries unless project-only memory applies. For separate client accounts, choose the boundary deliberately. OpenAI’s default-memory explanation.

The Limits That Affect This Setup

Instructions, reference files and sharing are available across the listed plans; memory remains subject to the settings described above. The Project owner’s plan determines the collaborator limit. These are the file and collaboration limits listed on OpenAI’s Projects page. Plans and limits.

You can create unlimited Projects, but upload only 10 files at a time, within the Project’s total allowance. Chat rate limits follow your subscription; the Projects page does not give a separate numerical message allowance.

File count is only one limit. OpenAI also lists 512 MB per document or presentation, 2 million tokens per text or document file, approximately 50 MB per spreadsheet or CSV, and 20 MB per image. Its general upload allowance is up to 80 files every 3 hours, with Free limited to 3 uploads per day; limits may be lowered during peak hours. Shared upload storage is capped at 25 GB per user and 100 GB per organization. These account-level limits also matter when filling a Project. OpenAI’s file-upload limits.

For an ongoing report, the other practical constraints are:

• Keep the evidence current. Uploading a file with an existing filename offers Upload anyway or Skip. Do not assume it replaces the old file. Deleting a file from a shared Project removes it for everyone.
• Check memory settings if continuity is missing. Turn on Enable memory if that is the control shown. With separate controls, Enterprise requires Reference saved memories plus workspace Memory. Other subscriptions require Reference saved memories and Reference chat history. Managed workspaces must allow memory.
• Keep an approved source instead of an invisible memory checklist. Project memory has no itemized list to inspect. To exclude a conversation, delete it or move it when the Project’s settings permit.
• Do not mistake a Project for scheduled reporting. OpenAI says ChatGPT Work is unavailable in Projects using project-only memory. Google Drive added within a Project does not sync content in advance for retrieval. The workflow above depends on someone updating and reviewing the inputs.

These behaviors are documented in OpenAI’s Projects guide. If your main problem becomes refreshing business data, the separate ChatGPT Data reporting workflow is the more relevant next comparison.

Project-only memory also does not mean a model-training opt-out. For personal accounts, shared-Project data may be used to improve models only when the owner and every contributor have Improve the model for everyone on. Business, Enterprise and Edu workspace content is not used for training by default. OpenAI’s data-use explanation.

Where Projects Pay Off Most

The strongest use cases have a recurring deliverable, a stable audience and source material someone owns. These are proposed workflows, ranked by practical fit for a small team.

The weak fit is a job with no maintained source material and no owner. A Project will give that work a home, but the team still has to decide what is correct and current.

What Is Worth Building Around This

The strongest opportunity is a weekly reporting setup service for agencies or small operations teams. The buyer gets a usable workflow: organized source files, metric definitions, project instructions, an approved report pattern and a review checklist.

The smallest sellable version could be a prepared source bundle plus an assisted setup for the buyer’s recurring report. The buyer creates the Project and owns its access. The difficult work is agreeing on definitions and approval rules. Generic prompt packs are easy to copy; knowledge of the buyer’s reporting process is the stronger reason to hire someone.

A client handoff kit is the next opportunity. An agency could buy a structured intake document, account brief, decision log and handover prompts that make a client Project useful to the next account manager.

Its first version could package existing client material for manual upload and include a walkthrough with the outgoing and incoming owner. The catch is stale information: a polished handoff becomes unreliable unless someone maintains the source documents. These are proposed services around the documented workflow, not claims of an integration that automatically creates or administers Projects.

Common Questions

How can I create a project in ChatGPT?

Select New project in the sidebar, enter a name and choose an icon and color. Add reference material, then use Project settings for instructions and the memory setting. OpenAI’s setup steps.

How many projects does ChatGPT allow?

OpenAI lists unlimited Projects. File limits apply separately to each Project, and collaborator limits depend on the owner’s plan. OpenAI’s limits.

Can ChatGPT make a report?

Yes. A Project can hold the reporting evidence and instructions for drafting recurring updates. For the suggested workflow here, supply current inputs, ask for source-backed claims and review the output before using it. OpenAI’s reporting use case.

Why are ChatGPT Projects not working as expected?

Check that the chat is inside the intended Project, its sources are current and the required memory settings are enabled. Workspace restrictions can also affect features. A chat cannot be added while it is temporary, and a chat created with a GPT cannot be moved into a Project. OpenAI’s troubleshooting details.

Your Monday Move

Take the next client report you already owe. Create its Project, add the approved brief and definitions, paste the reporting instruction, and run the draft through your normal review. Then ask a colleague to continue the work from the shared context. The useful test is whether they can understand the job and produce a reviewable next draft without another briefing from you.

If you want help turning that recurring report into a maintained business workflow, we can build the automation with you.