How to use ChatGPT for projects: a PM's guide to the Projects feature

If you're managing multiple client engagements and using ChatGPT as part of your workflow, you've probably hit the same wall I did: dozens of one-off chats, zero continuity, and the constant feeling that you're training the AI from scratch every Monday morning. ChatGPT's Projects feature is OpenAI's answer to that problem. It gives you persistent workspaces where chats, files, and instructions live together. This guide walks through what Projects actually does, how to set it up for professional services work, and where it stops being useful. Knowing that boundary matters just as much.

What is ChatGPT Projects?

ChatGPT Projects is a workspace feature that lets you group related conversations, upload reference files, and set custom instructions, all tied to a single initiative. Think of it as a persistent container for everything ChatGPT needs to know about a specific piece of work.

Instead of starting every chat cold, a Project carries context forward. You set the ground rules once (your role, the deliverable format, the client's preferences) and every new conversation inside that Project inherits them. It's the difference between briefing a new contractor every morning and working with someone who already knows the account.

OpenAI rolled out Projects as part of its push to make ChatGPT more useful for ongoing work, not just one-off questions. For PMs who already use ChatGPT to draft documents, analyze data, or brainstorm solutions, Projects adds the organizational layer that was missing. You're no longer searching through dozens of unnamed chats to find the one where you outlined the scope for Client X.

Here's what each component does:

Why PMs need a better system for AI conversations

A few years ago, I was using ChatGPT for everything from drafting proposals to summarizing meeting notes. But after a few months, my sidebar looked like a graveyard of unnamed conversations. I'd spend 10 minutes searching for a prompt I wrote last Tuesday about a client's scope change, only to give up and write it again from scratch.

That's the pattern we keep seeing across Teamwork.com customers too. PMs are using AI more than ever, but most are doing it in a way that creates more chaos, not less. Separate chats for each task, no shared context between conversations, and no way to pick up where you left off on a client engagement. According to Teamwork.com's Sprint to AI research, 58% of professionals confirmed they're now using 3 to 5 separate tools to get the job done. ChatGPT is increasingly one of those tools, but without structure, it just becomes another source of fragmentation.

The real cost isn't the time spent searching. It's the quality gap. When you start every ChatGPT conversation from zero, you get generic outputs that need heavy editing. When the AI already knows your client, your deliverable format, and your role, the first draft is dramatically closer to useful. That's the promise of Projects, and for PMs managing multiple client accounts, it's a meaningful improvement over the default experience.

ChatGPT Projects doesn't fix the multi-tool problem entirely. But it does fix one specific friction point: the lack of continuity in your AI conversations. For PMs juggling multiple clients, that continuity is worth more than any individual prompt.

How to set up ChatGPT Projects for client work

Most of the setup guides out there walk you through the buttons. Open the menu, name your project, done. What they don't cover is how to structure Projects for the way professional services teams actually work: multiple clients, overlapping engagements, deliverables that need to be client-ready. Here's how I'd set it up.

Create a project and name it for the client or engagement

Open ChatGPT and click the "+" next to Projects in the left sidebar. The naming convention matters more than you'd think. I recommend using a format like "Client Name, Engagement Type" (for example, "Acme Corp, Q3 Website Redesign" or "FinServ Co, Retainer Ops"). This makes it easy to scan your sidebar when you're switching between accounts.

If you're running multiple workstreams for the same client, create separate Projects for each. One Project per engagement keeps the context clean. Mixing a brand strategy conversation with a technical implementation chat in the same Project muddies the instructions and confuses the memory.

A good rule of thumb: if the engagement has its own SOW, it gets its own ChatGPT Project. If two workstreams share a brief and the same stakeholders, they can coexist. When in doubt, split them. Merging later is easy; untangling muddled project memory is not.

Write custom instructions that match your delivery context

This is where most PMs leave value on the table. Custom instructions aren't just a "nice to have." They're the mechanism that makes every chat inside the Project immediately useful.

Here's a template I recommend for professional services PMs:

• Role: I'm a delivery lead managing client projects for a [type of firm]. My team includes [roles].

• Deliverable format: All outputs should be formatted as [status updates / scope documents / meeting agendas / etc.]. Use bullet points for action items. Keep language professional but conversational.

• Client context: This project is a [engagement type] for [client description]. The key stakeholders are [roles]. The timeline is [duration].

• Constraints: Do not fabricate specific metrics or timelines. Flag assumptions clearly. Keep responses under [word count] unless I ask for more detail.

The more specific your instructions, the less editing you'll do on every output. In my experience, teams that skip this step end up spending as much time reformatting ChatGPT's responses as they would have spent writing the thing themselves.

One thing to note: custom instructions for a Project are separate from your global ChatGPT instructions. The Project-level instructions take precedence inside that workspace, so you can have different tones and formats for different clients without changing your default settings.

I've found that the most effective custom instructions include a "don't do this" section. Telling ChatGPT what to avoid is just as valuable as telling it what to do. List the jargon the client hates, formatting patterns that don't match your templates, and assumptions it shouldn't make. This is particularly useful for delivery teams managing standardized processes where consistency across client accounts matters.

Upload reference files your AI needs

Attach the documents ChatGPT will need to reference. For most PM engagements, that means the SOW or brief, any brand or style guidelines the client has shared, and templates you use for recurring deliverables. ChatGPT can pull from these files when generating responses inside the Project.

Keep your uploaded files current. If the scope changes (and in client work, it always does), update the brief in the Project. Stale files lead to stale outputs. I've seen teams leave an outdated SOW in a Project for weeks, then wonder why ChatGPT keeps referencing deliverables that were descoped a month ago.

Organize chats by workstream or phase

Within a Project, each new conversation shows up in the sidebar. Name your chats clearly: "Kickoff agenda, May 2026," "Scope change, Phase 2 additions," "Weekly status draft, Week 12." When you come back to the Project three weeks later, you'll know exactly where to find what you need.

A pattern that works well for client delivery is organizing chats by project phase (discovery, planning, execution, closeout) or by deliverable type (proposals, status updates, meeting notes). Pick one convention per Project and stick with it. Mixing both in the same workspace creates the same organizational chaos you were trying to escape.

Five ways PMs use ChatGPT Projects in practice

The setup is only half the story. What makes Projects useful for professional services PMs is what you do inside them. Here are five use cases I see working well, all specific to client delivery work rather than the generic "plan your vacation" examples you'll find elsewhere.

Drafting client-ready status updates

Set your custom instructions to include the status update format your client expects (bullet points, traffic-light RAG status, or narrative summary). Then start a chat with "Draft this week's status update" and paste in your task list or notes. Because the Project already knows the engagement context, the output is 80% there on the first pass.

Over time, project memory kicks in. ChatGPT starts to recognize recurring items (the integration that's always behind schedule, the stakeholder who needs extra detail) and adjusts accordingly. This is especially useful for agencies managing multiple retainer clients where the update format varies by account.

Building and refining project scopes

This is where Projects really earns its keep. Upload the initial brief or client request to the Project. Then use a series of chats to iterate on the scope: one chat to break the brief into workstreams, another to draft the deliverables list, a third to pressure-test the timeline.

Because every chat shares the same context, you don't lose the thread between iterations. I've found this particularly useful for agencies responding to vague briefs. Instead of a single marathon prompt, you can have a structured conversation that builds on itself over days or weeks.

The important distinction here is that ChatGPT is helping you think through the scope. It's not managing the scope for you. Once the scope is agreed, it needs to live in your project management platform where tasks, deadlines, and budgets are tracked, not in a ChatGPT chat. That's where tools like Teamwork.com's AI Project Wizard come in, turning that agreed scope into an actual project plan with dependencies, owners, and timelines.

Running post-project retrospectives

Upload your retro notes, client feedback, and project metrics into a chat. Ask ChatGPT to identify the top three recurring themes, the biggest delivery risk that materialized, and one process change to test next quarter. A single paragraph of context in the custom instructions ("This is a retro for a 6-month website build for a financial services client") sharpens the analysis significantly.

In my experience, retros produce better insights when the AI already has context on the project's history. A Project that's been active throughout the engagement has accumulated enough memory to spot patterns a fresh chat would miss entirely. The themes ChatGPT surfaces can feed directly into process improvements for your next engagement. That's exactly the kind of continuous learning loop that separates firms that scale delivery from firms that just add headcount.

Creating RFP response frameworks

Upload two or three past proposals that won work. Set the custom instructions to: "Use these past proposals as reference for structure and tone. All new responses should follow the same format." When a new RFP comes in, start a chat with the requirements and let ChatGPT draft a first-pass response anchored in your proven structure.

This alone can cut RFP response time in half for the initial draft. The editing and tailoring still needs a human, but the structural heavy lifting is done. For consulting firms and IT services teams that respond to dozens of RFPs each quarter, this compounds into significant time savings. The trick is keeping your "winning proposals" folder updated inside the Project. Outdated reference material produces outdated responses. Review your uploaded files quarterly, just as you'd review your project templates to make sure they still reflect how you actually deliver.

Generating meeting agendas and action items

Before a client meeting, start a chat in the project with "Draft an agenda for our weekly sync." Because the Project has context on the engagement, the agenda includes relevant items without you listing them. After the meeting, paste your notes and ask for a formatted list of action items with owners and deadlines.

The key is using the same Project for pre-meeting and post-meeting work. That way, ChatGPT can reference what was discussed last week when drafting this week's agenda, creating a continuity loop that mirrors how a well-run delivery team actually operates.

One thing to watch for: ChatGPT's action items are suggestions, not tracked tasks. If someone says "I'll have the wireframes done by Thursday" in a meeting, that commitment needs to land in your task management system with an owner and a deadline, not just in a ChatGPT chat that nobody checks again.

ChatGPT Projects plans and limits

Not every ChatGPT plan includes Projects, and the limits vary. Here's what you need to know before committing.

For most professional services PMs, the Plus plan covers individual use. If your team needs to share Projects and collaborate in the same workspace, you'll need the Team plan. The sharing capability is the key upgrade for agencies and consulting firms where multiple people work on the same client account.

One detail worth noting: project memory on the Free plan is limited. If you're working on complex engagements where context retention matters (and for most client work, it does), the Plus plan at $20/month is worth the upgrade. You can verify the latest plan details on OpenAI's pricing page.

For professional services firms evaluating the Team plan, the per-user cost adds up quickly across a delivery team. That's worth factoring into your AI tool stack decisions. The value proposition is strongest for teams where multiple people need to collaborate inside the same ChatGPT Project. That means sharing context across account managers, strategists, and delivery leads working on the same client.

Where ChatGPT Projects falls short for professional services

In my years working in agencies before joining Teamwork.com, I saw teams adopt every new productivity tool that promised to "change how they work." What I've learned is that the tool is never the problem. The problem is when teams start using a tool for something it wasn't built to do.

ChatGPT Projects organizes your AI conversations. It does that well. But here's what it doesn't do:

• No task management: You can't create tasks, set deadlines, assign owners, or track completion. A chat about deliverables is not a task list.

• No resource planning: There's no visibility into who's available, who's overbooked, or how work is distributed across your team.

• No budget or profitability tracking: ChatGPT has no concept of billable hours, project budgets, or cost rates. If you're managing client profitability, this is a non-starter.

• No time tracking: You can't log time against tasks or projects. For professional services firms where billable time is revenue, that's a critical gap.

• No client permissions: You can't give a client view-only access to project status. Everything lives inside your ChatGPT account.

• No reporting: There's no project health dashboard, no utilization report, no budget burn-down chart.

A pattern I kept seeing in my prior career, and still see at Teamwork.com, is teams cobbling together AI tools, spreadsheets, and messaging apps into a fragile system that works until it doesn't. Usually right around the time a project goes over budget or a deadline slips. For a broader look at how AI is shaping the project management tool category, we've covered that separately.

How Teamwork.com fills the gaps ChatGPT can't

ChatGPT Projects handles the thinking side of project work: drafting, analyzing, iterating on ideas. Teamwork.com handles the doing side: tasks, timelines, resources, budgets, and reporting. The two aren't competitors. They're complements.

At Teamwork.com, we build project and resource management software specifically for client work. That means the features are designed around the realities professional services teams actually face: scope changes, budget pressure, and utilization targets. They also handle keeping clients in the loop without adding more admin.

Here's where Teamwork.com picks up where ChatGPT leaves off:

AI Project Wizard turns a client brief into a fully structured project in clicks. Instead of spending an hour building out task lists and dependencies manually, you paste in the brief and get a ready-to-refine project plan. One of the reasons we built this at Teamwork.com is because I spent years watching teams rebuild the same project structure from scratch every time a new engagement kicked off.

Workload Planner shows exactly who's available, who's at capacity, and where work needs to be redistributed. For delivery teams managing multiple client accounts, this is the difference between proactive planning and reactive firefighting. When Invanity, a UK-based digital agency, moved to Teamwork.com, they cut time spent on weekly workload management by 80% and saw a 20% increase in on-time delivery.

Time tracking runs in the background while your team works. Built-in timers, retroactive logging, and automated reminders mean you actually capture the hours that turn into invoices. In my experience, teams that don't track time in real time leave significant revenue on the table every month.

Project health reports give you task progress, budget usage, and delivery status at a glance. No more pulling data from three different tools to answer "are we on track?" during a client call. Combined with profitability tracking, you can see not just whether a project is on time but whether it's still profitable. That matters enormously for client services firms where margins determine whether an engagement was worth taking on.

And if you want ChatGPT and Teamwork.com working together, we've built a ChatGPT connector via MCP that lets you query your Teamwork.com data directly from ChatGPT. Ask questions like "who's overbooked this week?" or "what's the budget status on the Acme project?" and get answers pulled from your live project data. We've written about the connector setup in detail if you want to dig deeper.

FAQ

Is ChatGPT Projects available on the free plan?

Yes, ChatGPT's free plan includes basic access to Projects with limited functionality. For full file upload support, project memory, and the ability to share Projects with teammates, you'll need the Plus ($20/month) or Team ($25/user/month) plan. Check OpenAI's pricing page for the latest details.

How does ChatGPT project memory work?

Project memory lets ChatGPT retain key details across multiple conversations within the same Project. It builds on what it's learned from previous chats, so you don't need to re-explain context each time you start a new conversation. This is separate from ChatGPT's global memory, which applies across all your chats outside of Projects.

Can I share a ChatGPT Project with my team?

Sharing is available on the Team and Enterprise plans. When you share a Project, teammates can view and contribute to the same workspace, including its chats, files, and instructions. On the Plus and Free plans, Projects are limited to your individual account.

What file types can I upload to a ChatGPT Project?

ChatGPT Projects supports common document types including PDFs, Word documents, text files, spreadsheets, and code files. You can upload reference materials like SOWs, brand guidelines, and templates that ChatGPT will draw on when generating responses within the Project. For details, see OpenAI's Projects documentation.

Is ChatGPT Projects a project management tool?

No. ChatGPT Projects is an AI workspace for organizing conversations, not a project management platform. It doesn't include task tracking, resource planning, budgets, time tracking, or reporting. For professional services teams, it works best as a complement to a dedicated PM tool like Teamwork.com that handles project delivery end to end.

Can ChatGPT connect to project management software like Teamwork.com?

Yes. Teamwork.com offers a ChatGPT connector via MCP that lets you query live project data directly from ChatGPT. You can ask about task status, team workload, and budget health without leaving your ChatGPT conversation.