Three Questions That Fix Your AI Content Workflow (Ask Them Before You Prompt)

The expensive part of working with AI is not the first draft. It is the third one — the loop where you read something almost right, cannot say why it is wrong, and type “make it better.”

That loop is almost never the model’s fault. It is the brief’s. A vague request produces a competent, generic answer, because a competent, generic answer is the only safe response to a question that did not say who it was for.

Three questions fix most of it. Answer them before you open the prompt box, in one line each. They take about ninety seconds and they remove the majority of rewrites.

1. Who is reading this?

Not “a customer.” Not “my audience.” A specific person in a specific state of mind, including what they already know and what they are worried about.

The difference this makes is not tone — it is content selection. A reader who already knows the background does not need the background, and every sentence explaining it is a sentence you will delete later.

• Vague: “Write a follow-up email to a client.”
• Specific: “Reader: a client who approved the budget three weeks ago, has not replied since, and is probably worried this project has stalled.”

The second version already implies the first sentence of the email. The first version does not imply anything, so the model guesses — and its guess is the average of every follow-up email ever written.

2. What should they do after reading it?

One action. Written as a verb, not as a feeling.

“Understand our value” is not an action; nobody can do it, so nothing in the draft will be shaped toward it. “Reply with a date” is an action. Once you name it, the whole structure follows: the piece needs to make that one action easy, obvious, and low-effort — and everything that does not serve it is a candidate for deletion.

• Vague: “Make them interested in the service.”
• Specific: “Reply with one of three dates for a 20-minute call.”

This is also the question that tells you the length. A piece asking for a one-word reply should not be six paragraphs.

3. What must be in it, no matter what?

The two or three facts that make the piece correct, and that the model cannot invent: the number, the date, the constraint, the name of the thing, the one sentence legal or accuracy requires.

This is the question people skip, and it is the one that causes the worst failures — not clumsy phrasing, but confident, fluent text with the wrong figure in it. If you do not supply the fact, something will still appear in its place.

• Vague: “Mention the pricing and the timeline.”
• Specific: “Must include: the work starts the week of the 14th; the quoted figure is unchanged; nothing is billed before the kickoff call.”

The five-line brief

Put the three answers above whatever you were going to type. Copy this, fill it in, then add your request underneath:

Reader: [who they are + what they already know + what they are worried about]
Action: [the one thing they should do after reading]
Must include: [the facts that cannot be invented]
Must avoid: [the phrasings or claims that would be wrong here]
Shape: [format and rough length — email, 5 bullets, 200 words]

“Must avoid” is the line most people leave blank and then regret. It is where you put the promise you are not allowed to make, the jargon your reader hates, the apology you do not want to open with.

What this looks like in practice

Before: “Write a LinkedIn post about our webinar on inventory forecasting.”

After:

Reader: operations managers at mid-size retailers who have heard
  "AI forecasting" too many times and assume it means a big software project.
Action: click through to the recording.
Must include: the webinar was 45 minutes; the method uses data
  they already have in their existing system.
Must avoid: the words "revolutionary" and "game-changing"; any claim
  about a specific percentage improvement.
Shape: LinkedIn post, under 150 words, no hashtag pile at the end.

The second one does not need a rewrite loop, because the reasons you would have rejected the first three drafts are already written down.

What this does not fix

It will not make an AI draft sound like you — that takes examples of your own writing, not instructions about it. It will not rescue a piece you had no reason to write. And it will not help if you genuinely do not know who the reader is; in that case the brief has done its job by telling you that before you spent an hour on the draft.

But the ordinary daily failure — fluent, plausible, subtly useless output — is a briefing failure far more often than it is a model failure. Ninety seconds of it, up front, is the highest-return habit in this entire category of work.

Two books that take this further

We publish two practical guides that extend this into full procedures, both one-time purchases:

• Stop Typing, Start Asking — $29. 119 pages across 9 chapters: a one-page prompt system built to cut weekly life admin down to copy, paste, and send. The first chapter is free to read in full, no email required.
• The One-Person AI Content Agency — $39. Nine chapters on turning one long-form recording into twenty distribution-ready assets, and on selling that procedure as a monthly retainer. One full chapter — the one on stripping generic AI phrasing out of client-ready drafts — is free to read, no signup.

Both, along with the free chapters and everything else we publish, are here: avaloncompany.ai