ChatGPT Complete #05 — Context, Memory, Instructions and Projects: What Belongs Where?

AIYING | CHATGPT COMPLETE | #05

Context, Memory, Instructions and Projects: What Belongs Where?

Understand the platform. Discover possibilities. Accomplish useful work.

By Ravindra Dastikop

You told ChatGPT that the workshop lasts 30 minutes. You also asked it to use plain English. Later, you begin another task. Which details should carry forward?

That depends on their purpose and scope. A workshop duration is a project fact. A preference for plain English may apply much more widely.

This edition helps you give information a useful home—and check whether the next response uses the right version.

1. Four concepts, different jobs

Context is the broad category: information available for answering a request. Instructions, memories, conversation messages and retrieved sources can all contribute context. The other three concepts are not separate containers completely outside it.

Concept	Practical meaning	Example
Task context	Information needed for the current output.	The passage to turn into questions.
Memory	Personalization that can carry useful information from earlier work, when enabled.	A recurring preference for beginner-friendly explanations.
Instructions	Directions about how to perform the work; their intended scope matters.	For this worksheet, support every answer with a quotation.
Project	An organized home for related chats, sources and project guidance.	The workshop's brief, worksheet and facilitator-guide conversations.

OpenAI documents both cross-chat personalization and project-based organization. Available controls depend on the product, account and workspace. Check the options actually shown to you.

2. Choose the scope before choosing the setting

Use this AIYING placement guide as a starting point:

Information	Suggested home	Reason
“Make this reply under 80 words.”	Current message.	Applies to one output.
“Prefer plain English and explain unfamiliar terms.”	Personal custom instructions, if broadly desired.	Deliberate guidance across tasks.
A stable personal preference you choose to retain.	Memory, where available and enabled.	Supports personalization.
“For this workshop, use only supplied passages.”	Project instructions.	Applies to related workshop outputs.
The current approved duration and learning objectives.	A dated project brief.	Provides a reference you can update and inspect.
A passage used for one worksheet.	The worksheet chat or its attachment.	Keeps task-specific evidence close to its use.

Memory can include preferences and recurring context, but a critical project requirement deserves an explicit, inspectable source. Avoid relying on a remembered summary as the only record of an approved decision.

3. Meera organizes her workshop

Fictional continuation: Meera now has a workshop agenda, worksheet and answer key. She wants revisions to remain consistent.

She creates a project named AI Answer Verification Workshop, where available, and adds a short brief:

WORKSHOP BRIEF — VERSION 1
Status: Approved for preparation; not classroom-tested.

Audience: first-year college students.
Duration: 30 minutes.
Resources: smartphones; no coding or paid subscriptions assumed.
Learning objective: verify answers against a supplied passage.
Outputs: facilitator guide, worksheet and separate answer key.
Success check: each answer is supported, or marked as not stated.

She adds project guidance:

For this workshop:
- Use plain English suitable for beginners.
- Follow the current approved workshop brief.
- Use supplied passages as the evidence for worksheet answers.
- Identify conflicts between sources before finalizing.
- Label suggestions separately from approved decisions.
- Do not claim the workshop has been tested.

She then starts separate project chats for the agenda, worksheet and final consistency review. Each chat has a clear outcome; the project holds the common reference material.

4. What happens when a fact changes?

The college extends the workshop to 45 minutes. Meera updates the brief to Version 2, marks Version 1 as superseded and tells the active chat which brief governs the revision.

The attached Workshop Brief Version 2 is now authoritative.
It changes the duration from 30 to 45 minutes.
Version 1 is superseded.

Read Version 2 and revise the agenda.
Preserve its audience and learning objective.
Make the timings total exactly 45 minutes.
Flag any related materials that still mention 30 minutes.
Do not claim to have checked files you cannot access.

Simply uploading another file does not establish that every subsequent response will select the intended version. Name the current source and check the result.

5. Memory is useful; review still matters

OpenAI describes memory as a way to carry useful context into future work. Personalization controls let you review and manage available settings.

A practical question is: “Would I want this detail influencing an unrelated task?”

“I prefer concise explanations” may travel well. “This session lasts 45 minutes” belongs with that session. Keep changing project facts in maintained briefs rather than turning them into broad personal instructions.

A project organizes context; do not assume it automatically provides complete isolation from other personalization. Inspect available memory and project controls before relying on a particular scope.

6. Your fifteen-minute setup experiment

1. Choose one ongoing task with at least two related outputs.
2. Create a project if available. Otherwise, use one clearly named chat with a maintained brief.
3. Write a short brief containing purpose, audience, current facts and completion checks.
4. Add a few project-specific instructions where supported.
5. Request one output and inspect whether it follows the brief.
6. Change one fact, update the brief and explicitly request a revision.
7. Check both the changed requirement and the requirements that should remain stable.

Expected accomplishment: one organized task, one maintained brief and a revision you have checked against the updated requirement.

7. A reusable context-check prompt

Before drafting, summarize the task basis you can access:
1. Current goal and audience.
2. Source files and versions used.
3. Requirements that govern this output.
4. Conflicts or missing information.

Distinguish provided facts from your assumptions.
If you cannot inspect a source or determine its version, say so.
Then create the requested output using:
[Name the authoritative source and desired output.]

Check the summary against the actual sources. It is a working account of the task basis, not a complete audit of every internal influence.

8. Verification checklist

• Is the authoritative brief named and current?
• Are personal preferences separated from project facts?
• Do project instructions fit every output in that project?
• Are superseded documents clearly identified?
• Does the revision use the new fact correctly?
• Are unsupported claims about source access or remembered details flagged?

Give each detail the right scope. Keep important facts inspectable. Verify the version used.

Next in ChatGPT Complete

#06 — Working with Files: From Upload to Evidence.

Official references

Checked on 15 November 2026. Interfaces and available controls can change.

• OpenAI: Personalize ChatGPT
• OpenAI: Projects and chats

The placement guide, fictional case, versioning practice and exercise are AIYING teaching material, not guaranteed product behavior or claims of completed testing.

AIYING | Choose work that Matters.
ChatGPT Complete • Ravindra Dastikop
Independent educational series. Not affiliated with or endorsed by OpenAI.