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.

ChatGPT Complete #04 — Give ChatGPT a Clear Brief: From Vague Requests to Useful Results

AIYING | CHATGPT COMPLETE | #04

Give ChatGPT a Clear Brief: From Vague Requests to Useful Results

Understand the platform. Discover possibilities. Accomplish useful work.

By Ravindra Dastikop

“Make it better.”

Better for whom? Shorter, more accurate, easier to teach, or ready to send?

ChatGPT can produce a polished response while missing the work you actually need. A clear brief makes your intention easier to act on and the result easier to judge.

1. What makes a brief useful?

OpenAI's prompting guidance recommends describing the goal, relevant context, desired output and boundaries. It also makes clear that no rigid formula is required: use the details that matter and refine through conversation.

For this edition, we add one practical teaching question: How will I check whether the result works?

Part of the brief	Workshop example
Purpose	Help students practise checking AI-generated answers.
Situation	First-year students; smartphones; 30 minutes.
Source material	The passage supplied below.
Deliverable	A worksheet and separate answer key.
Requirements	Five questions; plain English; passage-supported answers.
Evidence of completion	Each answer has a supporting quotation that I can verify.

This is an AIYING checklist for the exercise, not an official product feature or a mandatory prompting formula.

2. Meera's before-and-after brief

Fictional continuation: Meera is preparing the workshop introduced in edition #01. This time, she wants a worksheet that makes verification visible.

Broad request:

Create a worksheet about checking AI answers.

The topic is clear. The learners, source, question format and standard of correctness remain unspecified.

Clearer brief:

Create a worksheet for first-year college students.
Purpose: practise checking whether answers are supported by a passage.
Use plain English. No coding is required.

Use only this source:
<passage>
The college library opens at 8 a.m. on weekdays and 9 a.m.
on Saturdays. It is closed on Sundays. Students may borrow
two books for seven days. Reference books must remain
inside the library. A student ID is required to borrow books.
The reading room closes at 6 p.m. on weekdays and 1 p.m.
on Saturdays.
</passage>

Produce:
1. Five short-answer questions.
2. A separate answer key.
3. For each answer, the exact supporting sentence.
4. One additional question the passage cannot answer,
   clearly labeled “Not stated in the passage.”
5. A three-item student verification checklist.

Keep all answers within the information provided.
Do not invent library rules.
Make the distinction between stated information and
missing information visible.

The passage and scenario are fictional teaching material. The delimiters simply separate the passage from the instructions; they are not a special feature you must use.

3. Make the difference observable

Consider two possible questions about the passage:

• “How many books may a student borrow?” The passage supports “two books.”
• “What is the fine for returning a book late?” The passage supplies no fine.

A usable answer key preserves that distinction. An invented late fee fails the task even if it sounds realistic.

You can verify the supporting sentence for the first answer directly: “Students may borrow two books for seven days.”

Success here means traceability: the learner can connect an answer to evidence, or recognize that the evidence is absent.

4. A brief can also be short

Detail should resolve uncertainty. A small task may need only one sentence:

Rewrite the instructions below for first-year students in
under 80 words, preserving every library rule exactly.

For a larger deliverable, more context may help. Repeated demands such as “be excellent” add less direction than a specific requirement such as “include five questions and a separate answer key.”

5. A reusable brief template

Help me accomplish: [specific outcome]

Audience and situation:
[Who will use the result, and where?]

Relevant source material:
[Paste or attach the information needed.]

Create:
[Deliverable, format and useful length.]

Requirements:
[What must be included or preserved?]

Check:
[Observable conditions the result must satisfy.]

If essential information is missing, ask a focused question.
For minor gaps, state any assumptions clearly.
Flag requirements you cannot meet or verify.

Fill only the parts that affect your task. Remove irrelevant headings rather than inventing context to complete the template.

6. Revise with a specific correction

Suppose the worksheet is accurate but too difficult. Name the problem and the required change:

Revise only the student-facing question wording.
Use shorter sentences and familiar vocabulary.
Keep the facts, five questions and separate answer key.
Check that every revised question still matches its answer.
List any substantive changes.

Then inspect the revision. A request to preserve facts is an instruction, not proof that preservation occurred.

7. Your ten-minute experiment

1. Choose a real request whose output you can check.
2. Write a brief version in your usual words.
3. Run it and keep the response.
4. Add the missing audience, source, requirements and completion check.
5. Run the revised brief in a fresh chat, using the same model where possible.
6. Compare the outputs using the checklist below.

• Does the output address the actual purpose?
• Does it preserve the supplied facts?
• Does it include every required item?
• Does it identify missing information?
• How much correction is still required before use?

Expected accomplishment: one reusable brief, two outputs and a short note explaining which output is more useful.

This is a practical comparison, not a controlled benchmark. Outputs vary, and the revised brief may still require correction.

8. What a clear brief cannot guarantee

A well-specified request can still receive an incorrect answer. Source quotations can be wrong, calculations can fail and instructions can be missed. Check the parts that determine whether you can use the result.

For this exercise, that means reading every question, answer and quotation against the passage.

Describe the result you need—and the evidence that will make it usable.

Next in ChatGPT Complete

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

Official reference

Checked on 15 November 2026: OpenAI: Prompting.

The library passage, fictional case, checklist, prompt examples and exercise are original AIYING teaching material. No completed learner or classroom outcomes are claimed.

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

ChatGPT Complete #02 — Chat, Work and Codex: Choosing How to Work

AIYING | CHATGPT COMPLETE | #02

Chat, Work and Codex: Choosing How to Work

Understand the platform. Discover possibilities. Accomplish useful work.

By Ravindra Dastikop

You need help with a workshop. Do you want to discuss an idea, prepare the teaching materials, or build a small quiz application?

The subject is the same. The work you need is different.

In edition #01, we mapped conversation, model, context, tools and review. Now we make a practical choice: which way of working fits the result you need?

1. Three starting points

OpenAI's current documentation describes Chat for conversational work, ChatGPT Work for carrying larger tasks through to reviewable outputs, and Codex for developer-oriented work. These are useful starting points, with overlapping capabilities.

Starting point	Typical intention	Workshop example
Chat	Discuss, understand or refine.	Help me decide what students should practise.
Work	Produce a deliverable I can inspect.	Prepare my facilitator guide and student worksheet.
Codex	Build, inspect or change software using technical tools.	Implement and test a browser-based practice quiz.

Check your account: access, interface labels, tools and allowances vary. If Work or Codex is unavailable, you can still discuss the task and draft materials in Chat. Running code or producing a downloadable file depends on available tools.

2. Keep the distinctions clear

• A working experience shapes how you approach the task.
• A model supplies AI capability.
• A tool supplies a specific ability, such as searching or running code.
• A plan affects access and allowances.

These are different choices. You do not need a technical workflow simply because your topic is technical. Equally, a short request can require substantial technical work.

3. One case, three useful requests

Fictional case: Meera is preparing the 30-minute workshop from edition #01. Students will create five study questions from a short passage and verify the answers.

A. Chat: clarify the learning activity

Meera first needs to make a teaching decision. A focused conversation can help her examine options.

I am planning a 30-minute workshop for first-year college students.
They will create five study questions from a short passage and verify the answers.

Compare these approaches:
A. Demonstrate first, then let students try.
B. Let students try first, then discuss mistakes.

Explain one benefit and one limitation of each.
Recommend an approach for beginners and state your assumptions.
Keep the response under 250 words.

What to check: Does the recommendation account for beginners and the time available? Treat the recommendation as a proposal to judge.

B. Work: prepare the materials

After choosing the activity, Meera needs usable teaching materials. Her brief should specify the outputs and their acceptance criteria.

Prepare materials for a 30-minute introductory AI workshop.
Audience: first-year students, smartphones, no coding experience.
Do not assume paid subscriptions.

Create:
1. A facilitator guide with an agenda totaling 30 minutes.
2. A student worksheet with one original 120–150-word passage.
3. Five questions and a separate answer key.
4. A checklist for verifying answers against the passage.
5. A fallback activity without AI access.

Use only the passage for the answer key.
Keep questions and answers consistent across the materials.
State assumptions and flag anything requiring classroom testing.
Provide downloadable files if supported; otherwise provide clearly separated text.

What to check: Open the actual files when provided. Check timings, word count, answer support and consistency. An announcement that a file is ready is not a substitute for inspecting it.

C. Codex: implement the quiz

Later, Meera wants an interactive version. When she needs implementation, debugging and technical evidence, a developer-oriented workflow becomes useful.

Using the passage, five questions and answer key below,
build a small browser-based practice quiz.

Requirements:
- One self-contained HTML file.
- Works offline without external libraries.
- Shows the passage while students answer.
- Uses five multiple-choice questions.
- Awards one point per correct answer, maximum five.
- Shows feedback after submission.
- Reset clears answers, score and feedback.
- Includes visible labels and keyboard-operable controls.
- No accounts, tracking or network requests.

Inspect the implementation and run the checks available
in your environment. Report which checks actually ran,
their results and anything that remains untested.
Do not publish or deploy.

[Paste the verified passage, questions and answer key here.]

What to check: Try all-correct, all-wrong and unanswered submissions. Test reset and keyboard use. If the environment cannot run a browser, record that gap and test the interface yourself.

This is an optional extension. Completing this edition does not require programming.

4. Choose by the evidence you need

Before starting, finish this sentence: “I will know this task is complete when…”

• Discussion: I can explain the choice and its trade-offs.
• Teaching materials: I have inspected a consistent guide, worksheet and answer key.
• Quiz implementation: I have a working file and evidence of the checks performed.

These are acceptance criteria for this example, not guarantees attached to any mode. Your brief and review still matter.

5. Your ten-minute experiment

1. Choose one real task you need to complete.
2. Write its completion criterion in one sentence.
3. Choose a suitable available starting point: Chat, Work or Codex.
4. Provide the relevant information and request one concrete output.
5. Inspect the output against your criterion.
6. Ask for one specific correction and record what improved.

Expected accomplishment: one reviewed output and a clear reason for choosing that workflow. If the task needs longer, completing a defined first part is enough for this experiment.

6. A quick judgment check

“Explain what an API is to a beginner.” Chat is a reasonable start.

“Prepare a worksheet from my attached lesson notes.” Work is a reasonable start where available; check the source coverage and the output.

“Fix this quiz so resetting clears the previous score.” Codex is a reasonable start when implementation files and an execution environment are available.

The useful question is whether the experience supports the work and evidence you need.

Define the finish. Choose how to work. Inspect what was accomplished.

Next in ChatGPT Complete

#03 — Models, Plans and Limits: Understanding Your Choices. We will separate model capability from subscription access and usage allowances.

Official reference

Product distinctions checked on 15 November 2026 against OpenAI's Use ChatGPT documentation. The case, prompts, experiment and review criteria are original AIYING teaching material. No classroom or software outcomes are claimed here.

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

ChatGPT Complete #01 — Start Here: A Practical Map of ChatGPT

AIYING | CHATGPT COMPLETE | #01

Start Here: A Map of ChatGPT

Understand the platform. Discover possibilities. Accomplish useful work.

By Ravindra Dastikop

You open ChatGPT. A message box waits. You type a question, receive an answer, and wonder what to ask next.

A more useful starting question is: What do I want to accomplish, and what does ChatGPT need to help me?

ChatGPT Complete is an independent AIYING learning series for students, educators and professionals. We will explore the platform through practical tasks, reusable prompts and results we can check. This first edition gives you a map before we explore individual features.

1. Understand the names

OpenAI is the organization behind ChatGPT. ChatGPT is the product you interact with. A model is the underlying AI system used to interpret your request and generate a response. A tool adds a capability, such as searching the web or working with a file.

A plan determines access and usage allowances; it is different from a model. Selecting a model and selecting a tool also solve different problems. A capable model still needs relevant information to work with.

2. A practical map: five parts of the experience

The following is a learning framework, rather than an official classification of ChatGPT.

Part	What it contributes	Your useful question
Conversation	A place to explain the task, ask follow-ups and refine results.	Have I stated the outcome clearly?
Model	The AI capability used for the task.	Which available option suits the work?
Context	Your audience, notes, examples, constraints and relevant files.	What information does the task depend on?
Tools	Capabilities such as web search, file work, image input, voice or image creation.	Does this require information or actions beyond a text response?
Review	Your checks against sources, requirements and intended use.	What evidence would make this result usable?

For example, a request about today's announcements needs current sources. A request to summarize a particular report needs that report. A request for a workshop agenda needs the audience, duration and learning objective.

3. Choose a way of working

OpenAI's current documentation distinguishes Chat for conversation and back-and-forth, ChatGPT Work for carrying a larger task through to a reviewable result, and Codex for developer tools and technical work. Use the options actually available in your account; this exercise can begin in an ordinary chat.

For work that continues across several conversations, Projects can organize related chats and provide shared files and context. For a single self-contained task, a separate chat may be enough.

Availability note: features, labels, limits and access can vary by plan, workspace, device and rollout. This map is a guide to choosing a workflow, not a promise that every account has every feature.

4. One running example: plan a practical AI workshop

Fictional scenario: Meera, a college lecturer, needs a 30-minute introductory AI workshop for first-year students. They have smartphones, no coding experience and no paid subscriptions. She wants each student to produce and check a short study aid.

A broad request such as “Tell me about AI workshops” may produce background information. A clear brief gives ChatGPT something concrete to prepare.

Help me prepare a 30-minute introductory AI workshop.

Audience: first-year college students with no coding experience.
Resources: smartphones; do not assume paid subscriptions.
Desired accomplishment: each student creates a five-question study aid from a short passage and checks every answer against that passage.

Produce:
1. A timed agenda totaling exactly 30 minutes.
2. One original sample passage of 120–150 words.
3. A student prompt that uses only the passage.
4. A checklist for checking the five questions and answers.
5. A fallback activity if students cannot access ChatGPT.

Keep the activity manageable for beginners.
State assumptions. Do not invent institutional requirements.
Do not claim the workshop has been tested.

Illustrative agenda, not a tested result: introduction—4 minutes; demonstration—5; student activity—10; answer verification—7; reflection—4. Total: 30 minutes.

This brief specifies the audience, constraints, output and success criterion. No web search is necessary to create an original practice passage. If Meera later asks about current account limits, that becomes a separate task requiring current official information.

5. Your first experiment: from a request to a checked result

1. Open ChatGPT and start a new chat.
2. Paste the workshop prompt. You may substitute your own audience and topic.
3. Read the complete output before asking for changes.
4. Add the agenda timings yourself. Count the passage words.
5. Answer the generated questions directly from the passage. Check whether any answer introduces unsupported information.
6. Request a focused revision using the prompt below.
7. Save the revised workshop plan and record one improvement you made.

Review the workshop plan against my original brief.
Show a checklist with: requirement, evidence in the draft, and any gap.
Then revise the plan to address the gaps.
Keep the agenda at exactly 30 minutes.
Label anything that still requires my judgment or testing.

ChatGPT's self-review is another draft to inspect. Compare it with the actual output and the original brief.

6. The verification check

• Does the agenda total exactly 30 minutes?
• Is the sample passage 120–150 words?
• Are there five questions, with answers supported by the passage?
• Does the activity work without coding or paid subscriptions?
• Is the fallback usable without ChatGPT access?
• Are assumptions visible, with no invented claims of classroom success?

If a requirement fails, name the failure and request a correction. A polished response can still contain errors. Open source links when research is involved, check calculations independently and test instructions before using them with learners.

7. What have you accomplished?

Your expected output is one workshop plan checked against six explicit requirements. Whether it works in a classroom remains a question for a real trial.

The transferable discovery is simple: a useful ChatGPT workflow connects a clear outcome, relevant context, appropriate tools and human review.

Begin with the accomplishment. Give the context. Check the result.

Next in ChatGPT Complete

#02 — Chat, Work and Codex: Choosing How to Work. We will use practical examples to distinguish a quick conversation, a larger deliverable and a technical task.

Official references

Product documentation checked on 15 November 2026. The exercise, five-part map and fictional example are AIYING's editorial material.

• OpenAI: Use ChatGPT — ways of working and reviewing results.
• OpenAI: Features — capabilities and workflows.
• OpenAI: Projects and chats — organizing related work and context.

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