TL;DR
A ChatGPT Project is a container that gives every chat inside it the same instructions, the same uploaded files, and shared memory. Instead of re-pasting your resume, brand voice, or spec into every new conversation, you set them once at the project level. Projects are free for everyone (since September 2025); the only thing your plan changes is how many files each project holds: 5 on Free, 25 on Plus, 40 on Pro (as of June 2026). The default model behind project chats is GPT-5.5. The setup below takes about 15 minutes and pays off whenever a task runs longer than a single sitting.
What a Project actually is
Every chat inside a Project shares three things:
- One set of instructions — a project-level system prompt that applies to all chats inside it.
- One pool of files — uploaded references that every chat in the project can read.
- Memory — the project remembers context across its chats, so you stop re-explaining who you are and what you want.
The problem it solves: long-running task context stops getting lost between conversations. A standalone chat forgets everything the moment you start a new one. A Project does not.
Plan limits and what carries over (June 2026)
| Free | Plus ($20/mo) | Pro ($100 / $200/mo) | |
|---|---|---|---|
| Projects | Yes | Yes | Yes |
| Files per project | 5 | 25 | 40 |
| Per-file size cap | 512 MB | 512 MB | 512 MB |
| Per-file token cap (text) | 2M tokens | 2M tokens | 2M tokens |
| Default model | GPT-5.5 | GPT-5.5 | GPT-5.5 |
| Project-only memory | Yes | Yes | Yes |
| Custom color / icon | Yes | Yes | Yes |
Two caps to remember: each file can be up to 512 MB, and each text document is capped at 2 million tokens of extractable content (a long PDF can hit the token cap well before the size cap). Numbers above reflect OpenAI’s published limits as of June 2026 and shift over time — check the official Projects help page if a hard limit matters for your work.
When a Project beats a normal chat
Use a Project when the work continues across days or weeks and leans on the same reference material every time:
- Job search — one resume plus several job descriptions plus an answer bank, iterated over weeks.
- A book or long manuscript — chapters that must stay consistent in voice and canon.
- Running a creator channel — brand voice, audience personas, and a content calendar in one place.
- Long-running product work — PRD, interview notes, and competitor teardowns feeding every draft.
Skip it for one-off questions. A single throwaway chat is faster than building a container you will use once.
Step by step
-
Create the Project. In ChatGPT’s sidebar, open the Projects section (or the top ”+”) and choose New project. Name it after the outcome, not the topic:
- Good:
2026 job search - backend,SaaS content calendar Q2,Product X PRD v2 - Bad:
Test,Job stuff,AI project(future-you won’t know what it holds) - Optional: set a color and icon so the project is easy to spot in the sidebar.
- Good:
-
Decide on project-only memory. When you create a project you can enable project-only memory. With it on, chats reference other conversations inside the project but ignore your global ChatGPT memory — useful when a client project must not bleed into your personal context. Leave it off if you want the project to build on what ChatGPT already knows about you.
-
Write the project instructions. Open the project header, then Instructions. Paste this template and replace each
[...]placeholder with your real info:Your role: [e.g. "Senior PM focused on B2B SaaS tools"] Hard rules you must follow: - Output language: [English / Chinese / match the input] - Output format: [Markdown / plain text / tables when comparing] - Length: [specific, e.g. "<=200 words" or "3 paragraphs"] - Style: [e.g. "direct, no adjective stacks, <=3 sentences per paragraph"] Before answering, always: - Restate my question in one sentence so I can confirm you understood - List which project files / sources you drew from Always ask me first (do not generate) when: - [e.g. "you would have to make competitive claims about another product"] - [e.g. "the output will be published publicly"] - [e.g. "you would have to invent numbers or quotes"] Forbidden phrases: [e.g. "Introducing / Excited to / revolutionary / game-changer"] -
Upload reference material. In the project, open Files then Add files. Supported types include PDF, DOCX, TXT, MD, CSV, XLSX, PNG/JPG, and code files such as
.py. Stay under your plan’s file count (5 / 25 / 40) and upload only what should actually steer answers. Common bundles:- Job search:
resume_2026.pdf+target_jd_company_X.md+behavioral_qa_bank.md - Content calendar:
brand_voice.md+audience_personas.md+keyword_map.csv - PRD:
prd_v2.md+user_interview_notes.txt+competitor_teardown.pdf - Put a date or version in the filename (
v2,2026q1) so you can retire old copies.
- Job search:
-
First chat: make the model play the project back. Open a new chat inside the project and send:
This is the first chat in this project. Before generating anything: 1. Restate the core rules from project instructions (5 bullets max). 2. List the files you can see (filename + what you think each is for). 3. State what you think this project's goal is (2 sentences max). 4. List any missing information you still need to do high-quality work. -
Correct its understanding. Patch every wrong assumption now, before it writes resumes, articles, or code:
You missed that audience_personas.md is used to verify voice match, not just as background. The goal is not "produce a blog outline" but "produce a publish-ready blog post". Add this line to instructions: "After every draft, list a 3-bullet check against audience_personas.md confirming voice match."Have it restate the updated understanding before you let it generate anything.
-
Move related old chats in. In each existing chat, open the top ”…” menu, choose Move to project, and pick your project. Moved chats inherit its instructions and files, so search and context stay centralized.
-
Connect external sources (optional). Beyond uploads, ChatGPT can read live sources through connectors — Google Drive, Gmail, Google Calendar, Slack, GitHub, Notion, and Outlook/Teams are supported as of June 2026, and Deep Research can also point at any MCP server. Enable a connector once, then reference it inside the project so drafts pull from current data instead of a stale snapshot.
-
Share, with the right permission. Broad project sharing rolled out first on Team, Enterprise, and Education and is expanding to Plus and Pro. Where available, open the project, click Share, and pick Only those invited or Anyone with the link. Before sharing, scrub the Files panel for passwords, customer lists, and unpublished financials — anything in project context can surface in a collaborator’s answers.
-
15-minute weekly cleanup.
- Delete files not touched or referenced in 30+ days (remember the count cap is tight: 5 / 25 / 40).
- Promote any prompt pattern that worked this week back into the instructions.
- Archive finished sub-task chats with a
[done]prefix so they stay findable. - Keep one pinned “Project brief” chat with the current version number, recent decisions, and open questions.
Quality check
- Ask a fresh chat inside the project to restate the instructions and key files. If it misses something important, the instructions or sources need tightening.
- Compare an answer inside the project against the same question asked outside it. The project answer should reuse your saved context with far less repeated explanation. If it doesn’t, your instructions are probably too vague.
- For shared projects, double-check privacy: anything added as project context can influence answers visible to every collaborator.
Projects vs Custom GPTs
| Project | Custom GPT | |
|---|---|---|
| Built for | Yourself, ongoing work | Others, reusable behavior |
| Holds chats | Yes (shared history + memory) | No (each chat is fresh) |
| Files + instructions | Yes | Yes |
| Shareable as a product | Limited (collaborators) | Yes (publish to anyone) |
Rule of thumb: a Project is a working context for you; a Custom GPT is a packaged tool for others. If you want a behavior other people can run on their own, that’s a Custom GPT, covered in ChatGPT Custom GPTs.
Common mistakes
- Treating a Project as a folder. The value is the combination of instructions, sources, and reusable chats, not storage.
- Uploading everything. With only 5 / 25 / 40 file slots, fill them with material that should genuinely guide answers.
- Hoarding stale files. Old drafts and outdated sources produce stale answers; cut them weekly.
- Sharing before reviewing. Scrub instructions and files for private data first.
FAQ
Q: How many files can a Project hold? A: As of June 2026, 5 files on Free, 25 on Plus, and 40 on Pro. Each file can be up to 512 MB, and each text document is capped at 2 million tokens of extractable content.
Q: How are Projects different from Custom GPTs? A: Projects are private to you and bundle ongoing chats, files, and instructions under shared memory. Custom GPTs are reusable templates you can publish for anyone to run. Use Projects for your own long-running work; use Custom GPTs to package a behavior for others.
Q: Do uploaded files persist across chats? A: Yes. Every chat inside the Project sees the same files and instructions, so you don’t re-upload your resume, brand brief, or spec each session.
Q: What does project-only memory do? A: With it enabled, chats can reference other conversations in the same project but won’t pull in your global ChatGPT memory, keeping the project self-contained. Useful for client work you don’t want mixing with personal context.
Q: When should I start a new chat inside a Project versus continue an existing one? A: Start a new chat when the sub-task shifts (resume tailoring vs. interview prep). Continue when you’re iterating on the same artifact. Shared project context follows you either way.
Q: Can I share a Project with my team? A: Broad sharing launched on Team, Enterprise, and Education accounts and is expanding to Plus and Pro. Where available, use Share and choose Only those invited or Anyone with the link — scrub sensitive files first.