DEPTH · CHATGPT · 03 OF 04

Prompting ChatGPT well

Three skills that work everywhere, with ChatGPT's specific levers

13 min read·3 of 4 in the Claude track

The three core skills — prompt anatomy, the iteration loop, magic phrases — are identical to the Claude track. If you read that one, you already know most of this.

Where ChatGPT is different is that it gives you more levers to set your defaults. Custom Instructions. Memory. Voice mode. Custom GPTs. None of these are required, but each one shifts what 'prompting ChatGPT' even means once you set them up.

This page covers the three core skills briskly, then spends real time on the ChatGPT-specific levers.

If you haven't read the Claude Skills piece, the three core skills are explained in more detail there. This page assumes the basics and focuses on ChatGPT's specific moves.

PART 01

The three core skills (the short version)

1. Prompt anatomy: context + goal + constraints + example

Context: what you're working with. Who are you, who's the audience, what's already been tried.

Goal: what you want at the end. The shape — email, list, decision, draft. Be explicit.

Constraints: what would make a bad answer. Length, tone, format. Constraints are the most underused part.

Example: optional but powerful. Show what good looks like instead of describing it.

2. The iteration loop

First response is a starting point, not a final draft. Read it. Notice what's wrong, missing, or off — specifically. Tell ChatGPT exactly what to change. Repeat until ready.

The skill is in being surgical: 'cut the second paragraph to two sentences,' not 'make it shorter.' Vague feedback gets vague improvements.

3. Magic phrases that consistently work

"Explain your reasoning."
"What could go wrong with this?"
"What are the tradeoffs?"
"Is there a simpler way?"
"Push back on this."
"Give me three versions."
"Explain it like I'm new to this."

These work identically in ChatGPT and in Claude. Pull them when you want a different shape of answer.

PART 02

ChatGPT-specific: Custom Instructions

Custom Instructions are the single highest-leverage feature in ChatGPT and most people never set them up. They're two text boxes in Settings → Personalization → Custom Instructions, and what you write there gets applied to every new conversation automatically.

The two boxes are usually labelled something like 'What would you like ChatGPT to know about you?' and 'How would you like ChatGPT to respond?' The first is about you. The second is about voice.

What to put in box 1 (about you)

Whatever ChatGPT would benefit from knowing every time you talk to it. Some examples:

What you do for work (so it doesn't explain your domain to you).
Where you live (so it doesn't assume US defaults when relevant).
What you're working on currently (a book, a business, a project) — only if it matters across many chats.
Any standing preferences about how you process information (e.g. 'I learn faster from examples than from definitions').

What to put in box 2 (about voice)

This is where you fix ChatGPT's defaults. Some good lines to consider:

"Skip the introduction. Get to the point in the first sentence."
"Don't open with restating my question."
"No emoji unless I explicitly ask for them."
"No bullet points unless they meaningfully help."
"Push back when you disagree. Don't be polite about it."
"If I ask a yes-or-no question, lead with yes or no."
"Flag low confidence explicitly: 'guess:' before things you're inferring."

Setting up Custom Instructions once saves you typing the same anti-yes-man preamble at the start of every conversation. It's a five-minute setup that compounds for every chat afterward. Do this before anything else.

PART 03

ChatGPT-specific: working with Memory

ChatGPT remembers things about you across conversations if you have memory turned on. This is helpful when it stores genuinely useful context (your job, the project you're working on) and unhelpful when it stores stale facts (an opinion you held two months ago, a name of a person you stopped working with).

Two habits make memory work for you instead of against you.

Habit 1 — Be deliberate about what you teach it

ChatGPT often forms 'memories' from passing remarks. If you mention that you're considering moving to Lisbon as a hypothetical, it might remember 'user is moving to Lisbon' as a fact. You can prevent this by saying explicitly: 'don't remember this.' You can also turn off memory in Settings.

Habit 2 — Audit memory occasionally

Settings → Personalization → Memory. Scroll through what ChatGPT remembers about you. Delete the stale stuff. It takes a minute and prevents weird drift.

PART 04

ChatGPT-specific: Projects

ChatGPT now has a Projects feature similar to Claude's. You create a project, attach reference documents and instructions, and every chat inside that project picks up the context automatically.

Use it for the same things you'd use Claude Projects for: ongoing work, persistent reference material, separating different threads of your life. The setup is similar; the value is similar.

Available on Plus and Pro plans.

PART 05

ChatGPT-specific: Custom GPTs

Custom GPTs are mini-ChatGPTs you (or someone else) configure for a specific purpose. There's a 'recipe generator' GPT, a 'meeting note summarizer' GPT, a 'travel planner' GPT, and tens of thousands of others.

When to use one: when you do the same kind of task often (weekly meeting notes, daily journal prompts, recurring writing tasks), a Custom GPT pre-loaded with your preferences saves the setup time every conversation.

When to skip them: for one-off tasks, the GPT layer is overhead. Just chat normally.

If you want to make your own Custom GPT, the setup is mostly conversational — ChatGPT walks you through it. The whole thing is more accessible than it sounds. Worth trying once even if you don't end up using the result, just to see what's possible.

TAKEAWAY

The order to do this in

If you're going to invest one hour in getting better at ChatGPT, this is the order:

1. Set up Custom Instructions. Five minutes. Single biggest payoff.

2. Turn on memory, then audit it once a month. Two minutes per audit.

3. Get comfortable with the iteration loop and magic phrases. Practice on real work.

4. Make a Project the first time you find yourself pasting the same context twice.

5. Try a Custom GPT only if you have a specific recurring task in mind. Otherwise skip.

Try it now

Three exercises that build on each other. Do them in order if you can.

Practice 1 — Write your Custom Instructions

(This is a setup exercise — do it once, then forget it.)

Go to Settings → Personalization → Custom Instructions.

In box 1 ("about you"): write three sentences about who you are, what you do, and what you tend to use ChatGPT for.

In box 2 ("about responses"): try these as a starting point and adjust to taste:
"Skip introductions. Get to the point. Push back when you disagree — don't soften it. No emoji unless I ask. No bullet points unless they meaningfully help. If I ask a yes-or-no question, lead with yes or no. Flag low confidence with 'guess:' before things you're inferring."

Save. Then test it in your next conversation and see how the voice changes.

Practice 2 — Build a full-anatomy prompt

Use all four parts.

Context: I [WHO YOU ARE] and I need [WHAT YOU'RE TRYING TO DO]. The situation: [2-3 SENTENCES]. The audience: [WHO IT'S FOR].

Goal: [SPECIFIC OUTCOME — what shape, how long, what should it contain].

Constraints: Don't [WHAT TO AVOID]. Keep it [TONE / LENGTH / STYLE].

Example: Here's something in the tone I want — [PASTE 2-3 SENTENCES].

Give me a first draft.

Practice 3 — Use three magic phrases in one conversation

(After ChatGPT has given you any answer.)

Round 1: Explain your reasoning for that recommendation.

Round 2: What could go wrong with this approach? Be specific about the failure modes.

Round 3: Is there a simpler way to achieve the same outcome?

Notice how each one changes the shape of the answer.