Focus-group a draft against AI versions of your customers

Before a real customer ever sees your email, landing page, or offer, run it past a panel of synthetic buyers — detailed AI personas built from real customer research. Each one reacts in character, from its own fears and priorities. You revise, re-run, and loop until the panel stops finding real problems.

How it works

The skill has two halves: a library of personas and a review loop.

A persona isn't a one-line "busy mom" sketch. It's a page or two of dense, specific detail drawn from real voice-of-customer research — what this person fears, what makes them skim, the exact words they'd use, the objection that ends the conversation. The more specific the persona, the more useful the reaction. A vague persona gives you vague feedback; a sharp one tells you exactly where you lost them.

The loop is simple and it's the part most people skip:

1. Each persona reacts to the draft, in character — not as a helpful editor, but as themselves, with their real reflexes.
2. You extract the concrete feedback — the specific line that lost them, the missing piece, the thing that read as a pitch.
3. You revise and run it again.
4. You stop when the panel goes quiet — when the only notes left are nitpicks, not real problems.

The discipline is in step four. The temptation is to stop after one round, or to keep "improving" past the point of real feedback. The loop keeps you honest in both directions: don't ship on the first draft, don't polish forever.

The skill itself

A "skill" here isn't a program — it's a plain-English Markdown file you write once and then run by name. No coding required. This is roughly what the review-loop skill says:

# Synthetic Review Loop

Run a draft past a panel of synthetic customers and
revise until they stop finding real problems.

## The panel

Use the personas in the library — each one a dense,
real dossier (not a "busy mom" sketch). For this draft:
the skeptical owner, the overwhelmed operator, the
curious builder.

## The loop

1. Show each persona the draft. Have them react IN
   CHARACTER — as themselves, not as a helpful editor:
   "React only from your own life. You are NOT here to
   be helpful. If it doesn't land, say so. Then name
   the ONE change that would make it work for you."
2. Pull out the real problems — the line that lost
   them, the missing piece — and ignore the nitpicks.
3. Revise the draft to fix them.
4. Run it again. STOP when the panel goes quiet —
   when only nitpicks are left, not real problems.

That's the whole skill. The leverage is in two things the instructions can't give you: personas sharp enough to react like real people, and the judgment to tell a real problem from a nitpick.

Why this beats asking AI "is this good?"

Ask a model to review your copy directly and it defaults to being agreeable — it'll polish your grammar and tell you it's great. A persona built to react as a specific skeptical customer does the opposite: it tells you the email reads like a pitch, that you lost them in the second paragraph, that the offer is aimed one notch above where they are. That's the feedback that actually changes the outcome — and it's the feedback a real customer would never bother to give you. They'd just not reply.