Outcome: Build one focused project that carries the right instructions, files, and sources into each related chat, so you spend less time repeating context and less usage correcting avoidable misunderstandings.

A project is useful when work continues over time, produces several deliverables, or relies on the same sources. Keep one-time questions in a normal chat. Within a project, use a separate chat for each distinct outcome; the project supplies shared context while each chat stays focused.

Set up reusable context in five steps

1. Name the workstream. Use an outcome-oriented name such as “Fall advising communications” or “Annual equipment inventory,” rather than a broad name such as “Work.”
2. Write short project instructions. State the audience, preferred format, authoritative sources, review rules, and boundaries. Project instructions apply across the project’s chats, so include durable guidance rather than directions for one deliverable.
3. Add only reusable sources. Upload or connect the current handbook, approved template, glossary, or other material that several chats will need. Remove obsolete versions and avoid unrelated files.
4. Start one chat per result. Separate “draft the announcement,” “compare the two versions,” and “prepare meeting questions.” This keeps the transcript relevant while retaining the project’s shared files and instructions.
5. Refresh the project. When a policy, schedule, or template changes, replace the old source and note the effective date. Archive finished chats so active work remains easy to find.

Use less capacity: Luna is a good fit for naming, formatting, extraction, and first-pass organization. Use Terra for ordinary synthesis or judgment. Keep only relevant sources so each task has less unnecessary context to process.

Copy and adapt this project instruction

This project supports [workstream] for [audience].
Use the files and connected sources in this project as the primary context.
Treat [named policy/template] as authoritative.
For each request, give the requested deliverable first, then list missing or conflicting information.
Preserve names, dates, links, and approved terminology exactly.
Do not infer approvals, owners, grades, or deadlines.
Ask before sending, publishing, scheduling, or changing a record.

Illustrative SBS example

Example: A faculty member creates “Biology course refresh,” adds the current syllabus, course outcomes, and approved accessibility checklist, and sets instructions to cite page numbers and flag conflicts. One chat checks weekly topics against outcomes; another drafts practice questions; a third reviews a handout. The example is illustrative—the faculty member still verifies content and follows current SBS requirements.

Troubleshooting

• Remove or clearly label the older file, add the effective date to the current source, and ask ChatGPT to identify which version it used before revising the output.
• Start a new chat inside the same project and restate the single deliverable. Keep shared guidance in project instructions, not in a long catch-all conversation.
• Remove sources that are not reused, shorten instructions to durable rules, and attach task-specific material only in the chat that needs it.

Related SBS guides

• Write better prompts: a practical four-part method
• Choose a model, reasoning effort, and speed without wasting your limits
• Use AI responsibly: protect information and verify the result

Official guidance

• Projects and chats | ChatGPT Learn
• Set up a project teammate | ChatGPT use cases

Official guidance checked September 16, 2026. Features vary by product surface and workspace settings.