ChatGPT

If you only ever open one AI tool, it's probably this one — and that's a reasonable place to start. ChatGPT is a big general assistant: it drafts, summarizes, explains, plans, and talks through almost anything. It's the generalist, not the specialist. That's its strength and its trap — it'll happily attempt a job another type would do better, and sound just as confident doing it.

What it's genuinely good at

Think of ChatGPT as the smart, fast, widely-read assistant from Chapter 1 — the one who needs a real brief. Give it the whole assignment (deliverable, audience, constraints) and it's excellent at first drafts you then make yours: the email you're avoiding, the messy notes that need a shape, the thing you understand but can't yet phrase, the "explain this to me like I'm smart but new." It's also a strong thinking partner — talk a decision through and have it play back the trade-offs.

Where it's the wrong tool

It's a generalist, so it'll attempt jobs that belong to a specialist and you won't always notice until it matters. For research you can cite, a research specialist with sources beats it. For a polished, print-ready graphic, a visual tool beats it. For anything where the numbers have to be right, you are still the one who checks. Remember the rule from Chapter 1: a clean sentence is not a receipt.

What it costs

ChatGPT has a real free tier — genuinely usable for everyday drafting, summarizing, brainstorming, and getting unstuck, with daily limits on the smarter models. Paid tiers add capacity, the better "thinking" models, deep research, agent mode, and (on Business and Enterprise) keep your work data out of training by default. Prices and tiers change — OpenAI's pricing page is the one place to check before quoting a number.

Setting it up the LAiDIES way

Five ChatGPT-specific moves worth making early. Do them in any order; once they're in place, you don't redo them.

1. Custom Instructions. Settings → Personalization → Customize ChatGPT. Fill the fields with the basics you'd tell any new assistant — your role, your context, how you want things written, what "good" looks like, and what's off-limits ("don't invent facts; ask me before drafting if something's missing"). It applies to every chat without retyping. The fields are up to 1,500 characters each — use them.

2. Memory. Settings → Personalization → Memory. Turn it on if you want ChatGPT to carry useful preferences across chats — your tone, your project codenames, who's on your team — without you reminding it every time. The summary page shows exactly what's been saved; you can edit or delete any individual memory, or clear it all. Turn Memory off entirely on accounts you use for sensitive work.

3. Projects. Projects in the left menu. Make one for any ongoing job — a launch, a manuscript, a client engagement. Drop the reference files in, add project-specific instructions, and every chat inside stays on-topic with that context. Projects keep their own memory of where you left off; standalone chats don't.

4. Custom GPTs (Plus, Business, Enterprise). When you have a job you do repeatedly the same way — same audience, same format, same tone — build a Custom GPT for it. It's a Custom Instructions block + saved reference files + a name, bundled into one assistant you can pin and reuse. Keep them private or share inside your org.

5. Temporary Chat. The icon at the top-right of the chat screen. Won't appear in your history, won't write to memory, won't be used to train the model — and auto-deletes after 30 days. Reach for it when you're pasting something one-off you don't want associated with your account.

How to prompt it (the ChatGPT-specific moves)

The universal brief — deliverable, audience, constraints, check — is the foundation. These are the moves layered on top, pulled from OpenAI's current prompting guidance (the part that actually changed between models). Each one with a vague version, a better version, and the why — so it sticks.

1. Skip "you are an expert" — give context instead.

Vague: "You are a world-class marketing expert. Write me a launch email."
Better: "I'm drafting a launch email to B2B SaaS buyers who've heard of us but haven't bought. Tone: confident, not breathless. Length under 150 words. Highlight one specific feature benefit. End on a soft CTA. Show me three versions."

Why: Old advice said start every prompt with "You are a world-class…" — and for older models, that helped. OpenAI's own GPT-5.2 guide now calls that noise: the model already knows how to write a marketing email. What it doesn't know is your situation, your audience, your taste. Treat ChatGPT like a sharp new hire — they don't need you to tell them they're smart. They need to know the job, the audience, and when they're done.

2. Tell it what "done" actually looks like.

Vague: "Help me with my landing page."
Better: "Rewrite this landing page hero so a first-time visitor knows what we do, who it's for, and what to do next inside 5 seconds. Headline under 30 words; sub under 80. No buzzwords."

Why: Without explicit success criteria, ChatGPT picks one of a thousand ways to "help" — at random. It's like asking Cher for "a cute outfit." She'll absolutely pull something, but the version of cute in her head might be plaid pants with a yellow blazer. Spell out what done looks like and the right thing falls out.

3. Specific beats vague — every time.

Vague: "Write a poem about OpenAI."
Better: "Write a short inspiring poem about OpenAI, focused on the recent DALL-E launch."

Why: OpenAI's own example. The first prompt produces mush; the second produces something you might actually use. The pattern is always the same: add audience, length, format, tone, and the actual job the output has to do. More context in, better output out.

4. Iterate in the same chat. Don't restart.

Vague move: first reply is too corporate → open a new chat and try a different prompt.
Better move: stay in the same chat → "Make this less corporate." "Half the length." "Drop the bullets, rewrite as one paragraph." "Closer, but the opener still sounds like a vendor pitch."

Why: Every message you send becomes context the model uses for the next reply. Starting fresh throws away everything ChatGPT learned about your taste from this draft. Steering mid-chat compounds — each correction makes the next reply better. Restarting resets you to zero.

5. Break complex jobs into steps.

Vague: "Write me a launch campaign — email, landing page, three social posts, internal announcement, and press release."
Better: The brief first. Then headlines. Then body copy. Then social cuts. Each accepted step becomes the spine of the next.

Why: One mega-prompt forces ChatGPT to commit to a strategy without you weighing in. By the time you see the press release, it's built on five other decisions you never approved — and pulling on one thread unravels the whole thing. Step-by-step lets you catch the wrong shape before everything else is built on top.

6. Match the model to the task.

The default model is fast and good enough for almost all everyday chat — drafting, summarizing, brainstorming. The reasoning / "thinking" models (the slower ones in the model picker) are worth switching to when the job needs careful logic — multi-step analysis, math, code, untangling a complex decision with trade-offs. The picker is in the chat header; switch per task. Don't waste the slow one on chitchat; don't use the fast one for a multi-step audit.

The one safety note for ChatGPT

On a personal free or Plus account, your conversations may be used to improve the model by default — you can turn that off in Settings → Data Controls → Improve the model for everyone. For one-off sensitive things, use a Temporary Chat instead. Anything confidential — work data, client info — belongs in a company-approved Business or Enterprise account, used within policy (those don't train on your data by default). The full rule is in Chapter 1, Part 5.