The Art of the Follow-Up

Most founders treat the AI's first answer as a verdict. Read it, wince, close the tab, tell a friend the technology isn't there yet. Professionals treat the first answer as an opening bid — the start of a short negotiation, not the end of one.

The single biggest gap between people who get real work out of these tools and people who don't isn't the first prompt. It's what they type second.

Sarah's proposal

Sarah, the fractional CFO from chapter one, is pitching a prospective client: an eight-person supplements brand in Toronto doing about $2.4 million a year. She wrote a proper prompt using the structure from The Five-Part Prompt and got a solid draft in thirty seconds. Solid — not sendable. The opening read like a big accounting firm wrote it, and the pricing paragraph sounded apologetic about her $2,500-a-month fee.

Old Sarah would have copied the draft into a document and rewritten half of it by hand. Instead, she stayed in the chat. Everything already said in a conversation stays on the model's desk, so every note builds on the last — no re-explaining, no starting over.

Four follow-up moves

Point at the problem, specifically. "Make it better" is a coin flip; the model will change something, possibly the parts you liked. Compare: "Paragraph two sounds like a press release — make it sound like a text from a friend." The sharper you point, the better the fix. You're the editor here. The model is a very fast writer waiting for notes, and it takes notes without sulking.

Make it critique its own draft. Ask: "What are the three weakest parts of this proposal, from the client's point of view?" Models are strangely good editors of their own work — they'll flag padding, vague claims, and buried asks that they themselves just wrote. Then pick: "Fix one and three, leave the rest."

Request variations of just the weak part. Don't regenerate the whole draft to fix one paragraph; you'll lose what was already working. "Give me three versions of the pricing paragraph — one direct, one warm, one that leads with what the fee replaces" gets you options without collateral damage.

Ask the 20 percent question. When the draft feels done but flat, try: "What would make this 20 percent better?" The deliberately modest number matters — it invites sharp small improvements instead of a rewrite, and it routinely surfaces upgrades you didn't know to request.

The negotiation, in practice

Here's the middle of Sarah's actual session, condensed:

Four follow-ups, eleven minutes, and the proposal went out that afternoon instead of eating her Sunday night. No single message did the job. The sequence did.

When follow-ups stop working

Two or three rounds of notes should get you there. If they don't, stop pushing and diagnose, because the problem is usually upstream.

If the draft keeps missing facts, audience, or purpose, your original prompt skipped one of the five parts — read it back against the structure and you'll usually spot the hole in ten seconds. If the facts are right but the voice refuses to land no matter how you describe it, you've hit the limit of adjectives: paste an example instead, the way Dev did in Show, Don't Tell. Descriptions cap out. Demonstrations don't.

The skill worth keeping from this lesson is patience with a purpose. One more message is almost always cheaper than rewriting by hand — and the founders who internalize that stop asking "why is the AI mediocre?" and start asking "what's my next note?"

Try it now

Open a recent AI draft you gave up on — or generate a fresh one for a real task. Then run exactly three follow-ups: first, name the weakest paragraph and say specifically what's wrong with it; second, ask for the three weakest parts from your reader's point of view and fix the one that stings most; third, ask what would make it 20 percent better. Compare draft four to draft one. That distance is what you've been leaving on the table.