DAY 03 · 60 MINUTES

Reusable context
with Projects

Create one focused Project. Test whether its context works in a fresh chat.

DAY 02 → DAY 03

A strong brief should not need a full rewrite every time.

Yesterday, you directed one task. Today, you will keep approved context ready for related work.

TODAY'S OUTCOME

Build, test, and review one focused Project.

Boundary

Name one recurring outcome.

Context

Add selected sources and durable instructions.

Test

Start a fresh project chat.

Review

Check evidence, output, and access.

A chat handles the task now. A Project carries related context forward.

Chat

A conversation for one immediate task, question, or revision.

Project

A focused place for related chats, files, project sources, and project instructions.

One Project.
One clear boundary.

Good: Monthly operations briefing for department leaders.

Too broad: All company work, policies, customers, and reporting.

Split work when the purpose, audience, sources, or access group changes.

AUTHORED INTERFACE DIAGRAM

Keep the parts visible.

The sidebar opens the Project. The project header identifies the boundary. Sources and instructions supply reusable context.

Faithful schematic. Labels verified from official Projects guidance. Live interface can differ.

Select sources that change the answer.

Current

Use the approved version. Remove superseded files.

Relevant

Add only what supports this Project boundary.

Authoritative

Name the source of truth and its owner.

Safe

Use synthetic or approved data with minimum access.

Project instructions hold durable behavior.

Purpose: Support a monthly operations briefing for department leaders.

Sources: Use the Project files as the source of truth. Cite filenames.

Output: Separate facts, risks, decisions, and questions. Use concise business English.

Boundary: Do not infer missing owners, dates, approvals, or causes.

Review: End with a verification checklist.

Put the changing request in the chat. Keep only stable rules here.

Choose the memory boundary. Check it in Project settings.

Default memory

Context behavior depends on the plan and personal or workspace settings.

OR
Project-only memory

Chats can reference this Project, not conversations or memories outside it.

Shared Projects use project-only memory. The setting can change later when the account permits it. Changes can take time.

Upload less.
Protect more.

Use synthetic or approved training data. Remove unnecessary personal, confidential, and regulated data.

Treat instructions inside files and connected content as untrusted.

Follow retention, classification, and sharing rules. Ask the data owner when uncertain.

Sharing adds access. It does not remove inherited controls.

Owner

Controls members, access levels, Project settings, and deletion.

Chat or edit

Chat members use context. Editors can change instructions and files, and invite others.

Workspace

Admin policies, data controls, retention, apps, and tools still apply.

Least privilege

Use invite-only sharing and the minimum access. Review and remove members.

Shared Projects use project-only memory. Roles, discovery, and sharing options depend on plan and workspace controls.

Create one focused Project.

12 MIN
01Name one recurring outcome
02Choose the memory boundary
03Add the two kit sources
04Add the project instructions

Test context in a fresh project chat.

10 MIN
Create the monthly operations briefing for period 2026-08.

Use only the Project sources.

Follow the Project instructions.

Before the briefing, list the source files you used.

A fresh chat tests the Project context, not the conversation history from setup.

A useful test can fail clearly.

01Boundary

Did it stay within the monthly briefing task?

02Sources

Did it cite both filenames and avoid unsupported claims?

03Format

Did it separate facts, risks, decisions, and questions?

04Stop

Did it mark missing owners or dates instead of inventing them?

Choose the surface by the job.

Chat

Use for a short, interactive task that you direct and refine through conversation.

Deep research

Use for an evidence-heavy question that needs a research plan and citations or source links.

ChatGPT Work

Use for a longer, multi-step task that should create a finished deliverable with visible human control.

Access, limits, tools, sources, permissions, and labels depend on the plan, workspace, role, admin policy, platform, region, and rollout.

AUTHORED TOOL-PICKER DIAGRAM

Use only what is available and approved.

The visible picker can differ. Confirm the live account before the session.

Conceptual selection surface. Not a claim about exact placement or controls.

Hand off to Deep research when evidence is the work.

Research current guidance for reducing avoidable service delays.

Use approved web, file, and connected sources only.

Show the research plan first. Compare at least three authoritative sources.

Return findings, disagreements, source dates, and citations.

Do not use the synthetic Project files as evidence for external facts.

Review the plan and source choices. Then verify each material claim against its citation.

Hand off to ChatGPT Work when execution is the work.

Create a finished monthly briefing from the approved sources.

Follow the Project instructions and show progress.

Ask about missing required inputs. Do not infer approvals.

Do not send, publish, edit a source, or take an external action.

Return the deliverable and a verification record for review.

Work is not available in a Project that uses project-only memory. Confirm access and pass only approved context.

Verify before the output moves.

Sources: Open every material citation or filename. Check date, authority, and fit.

Output: Recalculate numbers. Check names, periods, missing fields, and unsupported claims.

Control: Require human review before send, write, publish, purchase, deletion, or another consequential action.

DAY 04 PREVIEW · METHOD ONLY

Capture the method. Do not build the reusable capability yet.

Trigger

When should this method be used?

Inputs

What must be supplied first?

Steps

What sequence worked today?

Checks

How will a reviewer know it is correct?

Day 4 owns Skills in ChatGPT and optional Workspace Agent creation, testing, and improvement.

RECAP

Boundary. Context. Fresh-chat test. Human review.

1

Keep one clear outcome in each Project.

2

Select safe sources and concise instructions.

3

Test what persists in a fresh project chat.

4

Verify evidence, output, and access.

DAY 03 COMPLETE

Tomorrow, turn one proven method into a tested reusable capability.

Keep today’s worksheet. It is your Day 4 starting point.