2.3 — The Follow-Up Habit | AI for Real Life

The first answer is a draft. Always.

The first answer is a starting point: a draft Claude is offering you so you can react.

This lesson is the steering wheel. Ten short follow-ups, copy-paste-able, that turn an okay answer into a useful one without rewriting your prompt from scratch.

Why "just rewrite the prompt" is the wrong move

When the first answer is off, the instinct is to scrap it and rewrite a longer prompt. Sometimes that's right. Often, it's not, and a 1-line follow-up is faster than a full rewrite. Following up also keeps the context: Claude already knows your subject, your tone, your constraints. You don't have to brief it again. You just have to steer it.

The boss reframe: say it like you mean it

Claude is calibrated to be agreeable. If you politely ask "could it maybe be a little shorter?" you'll get something a little shorter. If you say "cut it by 40%" you'll get something 40% shorter. Be specific. Be a little blunt. You're not being rude: you're directing. The model is not your friend; it's your assistant. Treat the chat like you're the boss who knows what they want, not the intern who hopes the deliverable is okay.

The 10 follow-ups.

1 · The cut-30 (when the answer is too long)

Cut this by 30%. Same content, fewer words. Kill any sentence that's just throat-clearing. Don't rewrite the meaning, just tighten.

2 · The "explain it to a 12-year-old"

Re-explain the [concept / paragraph] to me like I'm a smart 12-year-old. No jargon. Use a concrete example. If I still don't get it after this, the example was wrong.

3 · Three versions

Give me 3 different versions of this: at the angle level, not the word level. Make them genuinely different (different argument, different tone, or different structure). Label each one and tell me which you think is strongest and why.

4 · "What am I missing?"

What am I missing? List the 3 strongest objections to my [thesis / plan / approach], and the 1 question a smart skeptic would ask first. Don't soften them.

5 · Push back on it

Push back on what you just said. What's the strongest counterargument to your own answer? Where might you be wrong?

Five more: the ones for shape, voice, and depth.

The rest of the ten. These come up later in the chat, after the structural problems are fixed and you're trying to land the version you'll actually use.

6 · Make it sound like me

Rewrite this in my voice. [Paste 2–3 sentences of your real writing as a sample.] Match my sentence length, my word choice, and how I actually transition. If a sentence doesn't fit my voice, leave a [need rewrite] tag and I'll handle it.

7 · The "go deeper on point 2" move

Stay on point 2 from your last answer. Go 3 levels deeper on that one specifically. Skip the others. Don't summarize what you already said.

8 · Show me your sources / receipts

Where is this coming from? For each claim you just made, tell me whether you're (a) confident and could cite it, (b) repeating something common but I should verify, or (c) inferring. Be honest about which is which. Don't fabricate citations.

9 · The "be more specific" pry

You're being vague. Quote the 3 most generic sentences in your last answer and replace each one with a more specific version that names a real example, a real number, or a real moment.

10 · The "what would change your answer?"

What information, if I gave it to you, would change your answer? Name 2 things you don't know about my situation that would matter. Then ask me about them.

The follow-up workflow: what good steering looks like.

The pattern: short, fast follow-ups in series. Don't dump 4 follow-ups at once. One at a time. React to what you get back. Steer.

A real follow-up sequence: outlining a paper

Send the briefed prompt (Module 2.1 formula): outline a 5-page essay on [topic], thesis is [X], etc.
First reply lands. Read it. Probably good but generic. → Use #4 ("what am I missing?") to surface objections you should address.
Second reply: 3 objections. One of them is a real hole. → Use #7 ("go deeper on point 2") to think it through.
Third reply: deeper analysis. Now you can actually see your argument. → Use #3 ("three versions") to see different ways to frame the thesis.
Fourth reply: 3 thesis options. Pick the one that sounds like you. Don't argue with Claude about which is best: pick the one you'd actually defend.
Send revised brief based on the chosen thesis. Maybe ask for a fresh outline now that the thesis is locked.
Then walk away. Write the draft yourself. Module 4.5 will tell you what to do with the draft.

Try this: run a 4-follow-up chat tonight

Pick anything from your real life: an email you're avoiding, a study question you don't get, a decision you're stuck on. Open Claude. Send a properly briefed prompt (Module 2.1 formula).

Then run at least four follow-ups, in series, before you stop. Use #1, #4, and one of #3 / #5 / #9 minimum.

Notice the answer at the start of the chat vs. the answer at the end.

The "okay, that's it" closer: and when to start a new chat

End the chat on purpose, with a one-line closer like, "Okay, that's the version I'm using. Summarize what we landed on in 3 bullets and we're done." Then: when you start a different task, start a new chat. A 60-message chat about your bio quiz is a bad place to draft an email to your boss: Claude will keep dragging quiz framing into it. Rule of thumb: same task, same subject, same week → keep going. Different task, different subject, or you're stuck in a rut → new chat. (Pro Cowork Projects from Module 7 fix the "but I want it to remember my context" half of this.)

Up next: 2.4: Hallucinations, Outdated Info & Generic Slop

You can now write a clean prompt and steer the answer. Time for how Claude actually fails: confidently making things up, missing recent events, defaulting to generic: and the verify habits that mean none of those failures land in your submitted work.

Continue to 2.4:Hallucinations & Slop →