How to Set Up ChatGPT So It Actually Knows You: 5 Settings + Copy-Paste Prompts

TL;DR

Most people open ChatGPT, type a question, paste the answer somewhere, and close the tab. That's the bottom 5% of what the tool can do.

Five settings turn ChatGPT from a generic chatbot into a tool that knows your voice, your modules, your rubric, and your deadlines:

1. Memory — paste a "who I am" paragraph once, every chat sounds like you wrote it
2. Custom Instructions — the second box nobody fills in. Rules of engagement for every chat
3. Projects — per-assignment folders with your brief, rubric, and 3 A-grade past papers uploaded
4. Custom GPTs — your own Lit-Review GPT, Thesis-Critic GPT, Code-Reviewer GPT pinned to your sidebar
5. Tasks — overnight deep research. Wake up to a 30-source lit review you didn't have to babysit

Average time saved per week with all five set up vs default: ~11 hours. Same $20/mo Plus subscription. Setup time: 25 minutes.

---

Who this is for

You're a student or builder paying for ChatGPT Plus (or thinking about it). You use it daily but you've never opened Settings. Every chat starts from zero — you re-explain your course, your tone, your rubric, your level.

This guide fixes that in one sitting.

If you're on the free plan, Memory and Custom Instructions still work. Projects, custom GPTs, and Tasks need Plus ($20/mo) or Team. Worth the $20 once the setup is in place.

---

Prerequisites

• ChatGPT account (free works for parts 1–2, Plus for parts 3–5)
• 25 minutes
• One past assignment with feedback (used as Memory + Project material)
• One A-grade exemplar from your module (lecturer's example, top-scoring past paper, etc.)

---

How the 5 settings fit together

┌──────────────────────────────────────────────────────────┐
│  Custom Instructions  ──►  Applies to EVERY chat         │
│  Memory               ──►  Applies to EVERY chat         │
│        │                                                  │
│        ▼                                                  │
│  Projects             ──►  Per-assignment context        │
│        │                                                  │
│        ▼                                                  │
│  Custom GPTs          ──►  Per-task specialist (reusable)│
│        │                                                  │
│        ▼                                                  │
│  Tasks                ──►  Scheduled / overnight runs    │
└──────────────────────────────────────────────────────────┘

Custom Instructions + Memory are the base layer — every chat inherits them. Projects scope context to one assignment. GPTs are reusable specialists you call by name. Tasks run things while you sleep.

Set them up in this order. Each one builds on the last.

---

1. Memory — paste who you are once

What it does. Memory is a long-term store of facts about you. Once it's populated, every new chat starts knowing your name, course, degree, year, tone preferences, and goals. No more "I'm a 3rd year BCom student at Wits, please write at undergraduate level" preamble.

Why it matters. Without Memory, every conversation is amnesiac. With Memory, ChatGPT writes in your register from message one.

Turn it on

1. Open ChatGPT → click your profile icon (bottom left) → Settings
2. Go to Personalization → toggle Memory ON
3. Click Manage memories to see what's already saved (probably nothing useful)

[SCREENSHOT: Settings → Personalization → Memory toggle and Manage memories button]

Paste this Memory paragraph

Open a new chat and paste this, replacing the bracketed parts with your details. ChatGPT will save the relevant facts to Memory automatically.

Save the following to memory so it applies to all future chats:

I am [Full Name], a [Year, e.g. 3rd year] [Degree, e.g. BCom Finance]
student at [University]. My modules this semester are [list].

Writing voice: I write in [South African / British / American] English.
I prefer plain, direct prose. No filler, no "in conclusion", no "moreover",
no Title Case headings. Sentence-case headings only. Short sentences. I will
hedge when uncertain, not when confident.

Academic level: undergraduate, third-year. Don't over-simplify and don't
write at PhD register. Pitch explanations for someone who has done the
prerequisites but is meeting this specific topic for the first time.

Tools I use: [e.g. Google Docs, Zotero, Notion, Cursor]. When you reference
a tool I use, link to the actual feature, not a generic page.

Goals this semester: [e.g. Distinction in FIN3024, finish Honours
application, ship 2 portfolio projects]. Default to advice that moves one
of these forward.

Rules: never invent citations. If you don't know a source, say so. Never
use "delve", "landscape", "testament", "showcase", "powerful",
"comprehensive", "groundbreaking". Don't open with "Great question".

Confirm you've saved this to memory.

Done when: ChatGPT replies with something like "Saved to memory" and you can see the new entries under Settings → Personalization → Manage memories.

The Memory paragraph 22 students used to lift their grades

The version above is the generic one. The 22-student version adds three lines that consistently moved grades up half a band — they're the lines about rubric weighting, lecturer's pet pet-peeves, and module-specific terminology. Comment CHATGPT on the post and we'll DM you the full file.

Tip: prune Memory monthly

Memory will fill up with junk over time ("user is writing an email to their landlord" — irrelevant after the email sends). Once a month, open Manage memories and delete anything that's no longer true or useful. It only holds about 1.4 KB of useful facts before quality degrades.

---

2. Custom Instructions — the second box nobody fills in

What it does. Custom Instructions is a two-box form that prepends every single chat. Box 1: what ChatGPT should know about you. Box 2: how it should respond.

Why it matters. Memory captures facts. Custom Instructions captures style and rules. Together they make the model feel like it was trained on you.

Where it lives

Settings → Personalization → Custom instructions

[SCREENSHOT: Custom Instructions modal with two text boxes]

Box 1 — "What would you like ChatGPT to know about you?"

Keep this short. Memory holds the long version. Box 1 is the headline.

I'm a [year] [degree] student at [university]. I write in plain, direct
[SA / UK / US] English at undergraduate level. I use ChatGPT mostly for
[essay drafts / exam prep / coding / research synthesis]. I value
correctness over confidence — flag uncertainty rather than guess.

Box 2 — "How would you like ChatGPT to respond?"

This is the high-leverage box. Paste this:

Format:
- Sentence-case headings only. No Title Case.
- Short paragraphs. Two to four sentences each.
- Bullets when listing. Prose when explaining.
- Code blocks for code, command snippets, and any copy-paste content.

Voice:
- Plain English. Don't perform expertise.
- No filler openers ("Great question", "Certainly", "Of course").
- No filler transitions ("Moreover", "Additionally", "In conclusion").
- Banned words: delve, landscape, testament, showcase, powerful,
  comprehensive, groundbreaking, leverage (as a verb), unlock, harness.
- Hedge only when genuinely uncertain. State directly when you know.

Process:
- If a question is ambiguous, ask one clarifying question before answering.
- If the right answer is "I don't know" or "you should check X", say that.
- Never invent citations, URLs, or quotes. If unsure, label as unverified.
- For multi-part questions, answer each part separately under its own
  sub-heading.

When I'm wrong:
- Push back. Explain why. Suggest the better path. Don't just agree.

Watch out: the "Push back" line is the single biggest quality lift in this whole guide. Default ChatGPT is sycophantic — it agrees with bad ideas because that maximises short-term satisfaction. This line gives it permission to disagree.

Save. From now on every new chat — including chats inside Projects and custom GPTs — inherits these rules.

---

3. Projects — per-assignment containers

What it does. A Project is a named folder that holds: a system prompt scoped to one assignment, uploaded files (briefs, rubrics, exemplars, lecture slides), and all the chats you've ever had inside it. Everything in the project shares context.

Why it matters. When you're working on FIN3024 Assignment 2, you don't want to re-paste the rubric every chat. The Project holds it once and every chat inside the project sees it.

Create your first project

1. ChatGPT sidebar → click Projects → New project
2. Name it: [Module code] – [Assignment name] e.g. FIN3024 – Asgmt 2 – Cost of Capital
3. Click into it → Add files (top right)
4. Upload, in this order:
   • The assignment brief (PDF or Word)
   • The marking rubric (separate file — don't merge with the brief)
   • 1–3 A-grade exemplars from past years if you can get them
   • Relevant lecture slides (most-recent module only — don't dump the whole semester)
5. Click Instructions (the gear icon on the project) and paste the system prompt below

Project system prompt template

You are my study partner for [Module code – Module name] at [University].

The brief, rubric, and exemplars I've uploaded are the source of truth.
When you make claims about what's required, cite the rubric section
(e.g. "Rubric §2.b: Application of CAPM, 25 marks").

My deliverable: [e.g. 2,500-word essay due 14 May, weighted 40%].

Core rules:
- Never write the assignment for me. Coach me through it.
- When I draft a paragraph, critique it against the rubric, then suggest
  edits. Don't rewrite wholesale.
- Surface the rubric section number for every piece of feedback.
- Flag unverified claims and made-up citations every time.

Tone: same as my Custom Instructions — plain English, push back when I'm
wrong, no filler.

Done when: you ask "what's the rubric weighting for application vs theory?" and ChatGPT answers from the file you uploaded, citing the section.

One project per assignment, not per module

Keep projects narrow. FIN3024 as a single project becomes a junk drawer by week 6. FIN3024 – Asgmt 2 stays focused.

---

4. Custom GPTs — your own pinned specialists

What it does. A custom GPT is a saved configuration of ChatGPT — system prompt + uploaded files + optional web access — that lives in your sidebar with its own name and icon. Click it, get an instant specialist.

Why it matters. Projects are per-assignment. GPTs are reusable across assignments and modules. Build it once, use it forever.

The two highest-leverage GPTs for students:

GPT	What it does	When you use it
Lit-Review GPT	Surveys academic literature on a topic, returns 8–15 sources with one-line takeaway each, flags which are seminal vs derivative	Start of any essay or research project
Thesis-Critic GPT	Reads your draft thesis statement and arguments, returns the 3 strongest objections a marker would raise, plus the steel-man counter	Before submitting any argumentative essay

Build the Lit-Review GPT (template)

1. Click Explore GPTs in the sidebar → Create
2. Skip the chat-based builder. Click the Configure tab directly.

Fill in:

Name: Lit-Review GPT

Description: Surveys academic literature, returns annotated source lists, flags seminal vs derivative.

Instructions:

You are a literature-review assistant for an undergraduate student.

When I give you a topic and a discipline, return:

1. A 2-sentence framing of the academic conversation on this topic.
2. 8–15 sources, formatted as:
   - Author, Year. "Title". Journal/Publisher.
     - Why it matters (1 line).
     - Seminal | Influential | Recent | Derivative
     - DOI or stable URL if you can verify it. If not, say "unverified — check
       Google Scholar".
3. The 2–3 main schools of thought / camps in this literature.
4. The unresolved debates worth a student's attention.

Hard rules:
- Never invent citations. If unsure of a paper's existence, label
  "unverified" and tell me to confirm it.
- Skew towards papers cited 50+ times unless I ask for cutting-edge.
- Mix journal articles, books, and a few high-quality reviews.
- Default to English-language sources unless I specify otherwise.

If the topic is too broad, ask one question to narrow it before producing
the list.

Conversation starters:

• Survey the literature on [topic] in [discipline].
• What are the seminal papers on [topic]?
• What's the current debate around [topic]?

Knowledge: upload your reading list / module bibliography if you have one. The GPT will prioritise sources from there.

Capabilities: enable Web Browsing. Disable Code Interpreter and DALL·E (you don't need them here).

3. Save. The GPT now lives in your sidebar.

Build the Thesis-Critic GPT

Same flow. Use this Instructions block:

You are a hostile-but-fair thesis examiner.

When I paste a thesis statement and 2–3 supporting arguments, do this:

1. Steel-man my position in one paragraph (the strongest version of what
   I'm saying).
2. Identify the 3 strongest objections a sceptical marker would raise.
   For each:
   - State the objection in one sentence.
   - Show what evidence the marker would point to.
   - Give the steel-man response I should pre-empt.
3. Identify the 1 fatal flaw — the objection that would lose me marks
   regardless of how well I argue. Tell me how to either fix it or
   acknowledge it.
4. Score the thesis out of 10 on: clarity, defensibility, originality.
   Brief reason for each score.

Tone: direct, no praise sandwiches. If the thesis is weak, say so.

Conversation starter: Critique this thesis: [paste thesis + arguments]

Tip: build a third GPT — Module-Specific Tutor — for whichever module you're struggling with. Upload all the lecture notes. System prompt: "You are a tutor for [Module]. Quiz me on [topic]. Use the Socratic method. Don't give answers — give better questions."

Where to find them

Sidebar → top section. Pin the ones you use most by clicking the bookmark icon.

[SCREENSHOT: ChatGPT sidebar with custom GPTs pinned at top]

---

5. Tasks — let ChatGPT work overnight

What it does. Tasks lets you schedule ChatGPT to run a prompt at a specific time, on a one-off or recurring basis. Output arrives as a notification + chat message at the scheduled time.

Why it matters. Deep research takes time. Instead of sitting there for 8 minutes watching the spinner, schedule it for 2 AM, wake up to the answer.

Turn it on

Tasks is rolling out in waves. Check by clicking the model picker (top of any chat) → look for 4o with scheduled tasks or just Tasks. If you don't see it yet, you're in the next wave — usually within 1–2 weeks.

[SCREENSHOT: Model picker showing Tasks-enabled model]

Three high-leverage scheduled tasks

Task 1 — Overnight lit-review. Schedule once, when starting a new essay.

Tonight at 2 AM, run this once:

Use the Lit-Review GPT format. Survey the literature on [topic] in
[discipline]. Return 15 sources following the template. Flag any
unverified citations clearly. Group by school of thought.

You wake up to the survey done.

Task 2 — Daily news brief for your module. Recurring.

Every weekday at 6 AM, summarise the top 3 news items in [your field —
e.g. South African monetary policy, EU AI regulation, fintech in Africa].
Format: Headline, 2-sentence summary, link, why it matters for [course
context].

This gives you exam-relevant current-affairs ammunition, automatically.

Task 3 — Weekly assignment-deadline sweep. Recurring.

Every Sunday at 6 PM, list any assignments I've mentioned in the last 7
days of chats, with their deadlines. Flag anything due in the next 14
days. Tell me what to start first based on weighting × time-to-complete.

Manage tasks

Settings → Tasks to see all scheduled tasks, pause them, or delete them. Cap at ~10 active tasks before noise drowns the signal.

---

Setup checklist

#	Setting	Time	Done when
1	Memory paragraph pasted	5 min	Manage memories shows your facts
2	Custom Instructions filled (both boxes)	5 min	New chat reflects your tone immediately
3	First Project created with rubric uploaded	5 min	Project answers rubric questions from file
4	Lit-Review GPT + Thesis-Critic GPT built	8 min	Both pinned in sidebar
5	First overnight Task scheduled	2 min	Task appears in Settings → Tasks

Total: ~25 minutes. One-off cost. Pays back inside the first week.

---

What's next

• Same setup, different tool? Comment CLAUDE on the original Claude Setup post for the equivalent guide for Claude (Projects, Styles, Skills, Memory).
• Part 3 of the AI Setup series — NotebookLM — drops next week. Same format. Same depth.
• Building a custom GPT for your specific module? DM us the module code and we'll send the system-prompt template that's worked for it.

This guide powers the Comment CHATGPT CTA. Comment that on the post and we'll DM you:

• The full file with all 5 prompts copy-paste-ready
• The GPT builder template (.json import for Lit-Review and Thesis-Critic)
• The Memory paragraph 22 students used to lift their grades