ChatGPT Projects: How to Build Persistent AI Workspaces That Actually Remember Your Business

What you'll learn in this article:

• Why the re-briefing tax is the biggest hidden cost in most executive AI use
• How Projects works technically: what persists and what does not
• A complete setup walkthrough for 3 high-value executive use cases
• The custom instructions that make Projects genuinely useful
• Common failure modes and the fixes that work

A CMO at a 60-person SaaS company uses ChatGPT every day. He re-explains his company, product, team structure, preferred tone, and current priorities at the start of almost every session. He estimates 5-10 minutes per conversation on setup context before he gets to the actual work. Five conversations a day, five minutes each: 25 minutes of briefing, not AI work. Over a year, that is more than 100 hours spent briefing a tool that forgets everything the moment the conversation closes.

ChatGPT Projects, launched December 13, 2024 for all plans, solves this problem. Each Project is a persistent environment with its own files, custom instructions, and conversation history. When you open a Project, ChatGPT already knows everything you have told it in that context. The re-briefing tax disappears.

How Projects Works

Each Project holds: uploaded files (persist across all conversations), custom instructions (apply to every conversation automatically), and conversation history (grouped and searchable). Free users get 5 files per project; Plus users get 25; Pro/Business get 40.

ⓘ Important: What Projects Does Not Do

Projects does not automatically recall every specific exchange from every prior conversation. The custom instructions and uploaded files persist, but prior conversations do not auto-populate into new conversations. Think of it as: the Project remembers the briefing documents and standing instructions, not every conversational detail. Set expectations accordingly.

Setup: 3 High-Value Executive Projects

Project 1: Client or Account Management. Upload the account brief, recent communications, contract terms, and relationship notes. Write custom instructions that specify the relationship context and preferred communication tone.

CLIENT PROJECT CUSTOM INSTRUCTIONS

"This project is for managing the [Client Name] account. [Client] is a [size, industry, relationship stage]. Key contacts: [name, role, tone notes]. Current priorities: [list]. When drafting communications for this client, match the professional but direct tone I use with their team. When I share meeting notes, extract action items in a table. When I ask for analysis, reference the uploaded contract and account brief first."

Project 2: Board and Leadership Communications. Upload your current board deck, the last three board updates, your leadership team OKRs. Write instructions that specify the board's communication preferences and required output format.

BOARD PROJECT CUSTOM INSTRUCTIONS

"This project is for board and leadership communications. Board members have limited reading time. When I ask you to draft board updates, match the format of the uploaded examples: 3-5 bullet executive summary, metrics table, decisions needed section, appendix. Use a direct, no-hedging tone. Every output should be executive-ready on the first pass."

Project 3: Weekly Operations Review. Upload your Q1 goals, current metrics, and standing agenda items. Write instructions for a Monday briefing format and a mid-week update format.

OPS REVIEW PROJECT CUSTOM INSTRUCTIONS

"This project is for my weekly operations review. Each Monday, give me: (1) Top 3 priorities based on my Q1 goals in the uploaded document; (2) Risks or blockers based on what I have shared; (3) One question I should be asking that I have not asked yet. Keep each section to 3 bullet points. If I share meeting notes, organize them as: decision made / action item / open question."

The Custom Instruction Craft

Good custom instructions do three things: (1) set context once in 3-4 sentences; (2) define the exact output format: table, bullet, reading level; (3) state the defaults: what should ChatGPT do when it is uncertain?

The most underused technique: upload a "bad output" example: a real AI response or human-written document that missed the mark: and add a line to your instructions: "The uploaded bad-example document shows the tone and format I do NOT want." This calibrates the model faster than any positive description.

Action Steps Summary

• Identify your 3 top re-briefing contexts : the situations where you most often re-explain ChatGPT from scratch.
• Create your first Project with a clear name and upload 3-4 relevant files: context brief, output example, data reference.
• Write specific custom instructions : not filler sentences, but behavioral constraints with format definitions and defaults.
• Test with a real task on day one : measure how much setup context is now automatic.
• Iterate instructions after 3-5 uses : every Project should go through 1-2 rounds of refinement before it is production-ready.