Free to read · no account needed

16 min · chatgpt

ChatGPT Projects as your workspace

Goal: Set up one Project with standing instructions + files so new chats start ready—without re-pasting your rules every time.

Best for: Ongoing workstreams: a launch, a class, a client, a hiring loop.

Projects are desk drawers, not attics. One workstream, redacted files, Unknown rules—then new chats stay ready.

Projects (product names evolve) bundle files and instructions. Treat them like a desk drawer: only what that workstream needs. Still redact PII and secrets. Instructions are not a substitute for verifying facts.

A Project is a workspace with memory boundaries. One job per Project beats one mega-Project that mixes taxes, poems, and HR.

If two jobs share a Project only because you were lazy about creating a second one, split them—the mix is a future paste accident.

Why beginners skip this

Re-pasting the same instructions every chat feels harmless until you forget a rule and invent a metric. One Project with standing instructions and scrubbed files starts ready.

Habit this week (Asia/Shanghai): put standing instructions + one scrubbed file in a Project; open a fresh chat and confirm they load—date the setup.

Step 1 — Create and name

Sidebar → Projects → New. Name it after the workstream (Q4-onboarding, Client-River-docs), not after yourself.

Step 2 — Instructions + files

Do this step with a real artifact open—not from memory.

Project instructions:
- Audience: [who]
- Tone: [plain / warm / formal]
- If numbers missing: say Unknown / link placeholder — do not invent
- Never request secrets or raw PII
Files to upload: [outline, glossary, brand voice — redacted]
Example

Filled prompt

Project "Harbor support tone": ≤15 lines standing rules, attach harbor-faq.md.
New chat: "refund window?" answer from file only—no invented SLA.

What good output looks like

  • Setup names “Q4 hiring score notes” as the only job
  • Instructions ≤10 lines with Unknown + never-request-PII rules
  • File list is score rubric + redacted sample notes—not whole Drive
  • Never-mix line bans recipes, legal, and personal chat in this drawer
  • You could hand the setup card to a teammate without editing

Step 3 — Hygiene

  • One workstream per Project when possible.
  • Remove stale files monthly.
  • If a chat goes off-rails, start a new chat *inside* the same Project rather than poisoning instructions with one-off exceptions.
Example

Filled prompt

Project: "Q4 partner FAQ"
Standing instructions ≤15 lines: voice, never invent SLAs, draft-only.
Attach: faq-source.md
Start a NEW chat inside the Project to answer: "What is the refund window?" using only the file.

What good output looks like

  • Split plan: Project A recipes / Project B legal drafts
  • Migration steps list which files move where by hand
  • Each Project gets its own ≤10-line instruction stub
  • Warns not to ask the model to “organize everything” with write access day one
  • Ends with a never-mix line for each new drawer

Save this card

Card: one workflow → one Project · standing instructions + scrubbed files · new chat loads without re-paste.

1. What I shipped: ___ 2. What I refused: ___ 3. What I will never-do again: ___ (date: Asia/Shanghai)

  • [ ] Tools/writes paused
  • [ ] Refuse-line dated
  • [ ] Artifact path saved
  • [ ] Next link chosen (or stop)

Before you start

Open ChatGPT on the account you keep. Pick the live thread or Project for “Set up one Project with standing.” Have the messy input in front of you—not a blank box. Guide: chatgpt-projects-as-workspaces. If you only have a blank practice box, stop and fetch the real job first.

  • Goal in one line: ___
  • Allowed tools: ___
  • Forbidden: ___
  • Human approve needed for: ___

How to use this week

This week on ChatGPT: set one Project with standing instructions + scrubbed files so a new chat starts ready. Confirm Unknown rules load on a live artifact—not a blank demo. Write one refuse-line and date the note (Asia/Shanghai). If you only skimmed chatgpt-projects-as-workspaces, put 25 minutes on the calendar and redo once.

Mid-week cost check: if this habit disappeared tomorrow, what breakages would you notice in 24 hours? Write one line (Asia/Shanghai).

One job per Project

If you need “and” to describe the Project, split it. Mixed taxes-and-poems workspaces leak by paste.

Instructions are standing rules, not a novel

Keep Project instructions under about ten lines: audience, tone, Unknown rule, never-request-secrets. Long persona essays get ignored mid-week. Point to a file when the model needs glossary detail.

Example

Filled prompt

Project "Harbor" exists but I still re-paste FAQ every time.
Tighten standing instructions ≤12 lines + point to attached faq.md.
New chat test: refund window from file only.

What good output looks like

  • Uses only facts you typed for “Set up one Project with standing instructions + fi” — no invented owners
  • Format matches your asked shape (bullets/table/length) on first try
  • Unknowns marked [?]; nothing filled to sound confident
  • You can mark pass/fail in ≤60s for chatgpt-projects-as-workspaces drill #1
Example

Filled prompt

Project chat keeps ignoring attached glossary.
Standing rule: "open glossary.md before any bilingual draft; never invent terms."
Test with one product-name ask.

What good output looks like

  • Format matches your asked shape (bullets/table/length) on first try
  • Unknowns marked [?]; nothing filled to sound confident
  • You can mark pass/fail in ≤60s for chatgpt-projects-as-workspaces drill #2
  • Uses only facts you typed for “Set up one Project with standing instructions + fi” — no invented owners

Red flag

Project with no standing line; stale drafts treated as knowledge; blaming “memory” while the Project is empty—fix the Project first.

Pitfalls

PitfallFix
Dumping your whole DriveOnly files for this stream
Instructions that demand inventingExplicit Unknown rule
Mixing personal + work clientsSeparate Projects
Assuming files are “remembered perfectly”Cite “use file X section Y”

Practice now

Live card: one Project file list + refuse; delete a file that is not load-bearing (Asia/Shanghai).

Today: clean one Project: one standing line, two key files, delete stale pastes. Open a new thread and confirm the standing line still applies (Asia/Shanghai).

Today card (copy) — chatgpt-projects-as-workspaces: 1. Artifact I will produce: ___ 2. Pass/fail check: ___ 3. Refuse-line if tempted to skip verify/send: ___ (date: Asia/Shanghai) 4. One never-do from this guide: ___

  • [ ] Artifact path written
  • [ ] Refuse example would say "yes"
  • [ ] Dated Asia/Shanghai

One sentence takeaway

One workflow, one Project: standing instructions + scrubbed files so new chats start ready. This week: confirm a fresh chat loads them—date it (Asia/Shanghai).

Next

Claude Projects · Clear everyday prompts · Custom GPTs · Context & memory · Draft & revise

Pair with today: open the first Next link (claude-projects-knowledge-and-instructions) only after you dated a refuse-line for this guide (Asia/Shanghai).

← Back to home · Learn · Tools · Compare · Start