Editing the Junior Analyst: Critiquing an AI Positioning Draft
Objective: Given a raw AI-generated positioning doc excerpt for Mailchimp's AI Growth Assistant, identify every place the draft is generic, contradicts known institutional memory, or needs a specific human-sourced detail before it's fit to ship.
You're the PMM at Mailchimp reviewing a first-pass positioning doc that a teammate generated with an LLM for the AI Growth Assistant feature. It reads fine on a skim, but you know things the model doesn't.
Treat the draft like a junior analyst's first pass: mark what's generic filler, what contradicts what sales has told you, and what's missing a real proof point.
Before you start
What you'll need
Free path (everything below is enough to finish)
Free, comment threads make the critique easy to hand back to the original drafter
Free tier is enough to redraft a paragraph once you've told it what to fix
The process
1 step
Step 01 of 01
The lesson treats an LLM's first draft as useful structure and phrasing options with zero shipping authority, because it has no memory of which campaigns already flopped or which words your own sales team refuses to say out loud.
This draft calls the AI Growth Assistant 'a seamless, all-in-one solution.' You know two things the model doesn't. What are they, and what should replace that line?
Procedure
- Read the full draft excerpt once without editing.
- Flag any phrase that could describe literally any competitor's AI feature (generic filler).
- Cross-check each specific claim against what you know: sales' actual objection language, past campaign angles that already failed.
- Mark any claim with no source or number attached as 'needs a real proof point.'
- Rewrite the flagged lines with the missing institutional detail.
DRAFT: 'Mailchimp's AI Growth Assistant is a seamless, all-in-one solution that transforms your marketing.' FLAGS: - 'seamless' -> sales has told you prospects roll their eyes at this exact word, drop it - 'all-in-one solution' -> the 'all-in-one platform' angle already flopped in Q1 campaigns per the last debrief, don't reuse it - 'transforms your marketing' -> no proof point attached, needs a real stat REWRITE: 'Mailchimp's AI Growth Assistant flags your highest-conversion customers automatically, so you stop guessing who to email first.'
Healthy
Every generic phrase gets replaced with something only someone who talks to customers weekly would know.
Unhealthy
Shipping the AI's draft with a light copyedit pass and no institutional-memory check at all.
What this means
The gap between a usable AI draft and a shippable one is exactly the information the model was never given, what already failed and what your own team refuses to say.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| A positioning doc reads fine but sales won't use the language in it | Run it past sales and support before it ships, not after | 30 min |
Final deliverable
An annotated version of the draft with every generic or unverified line flagged, plus a rewritten version for at least three flagged lines.
See a reference example
Squarespace AI positioning critique (excerpt) DRAFT: 'Our AI website builder is the smartest way to launch your site.' FLAG: 'smartest way' has no proof point, and 'launch your site' ignores that most Squarespace users are redesigning an existing site, not launching a new one. REWRITE: 'Squarespace's AI design tools rebuild your existing site's layout in minutes, no template shopping required.'
Success criteria
You're done when you can:
- Every generic, competitor-interchangeable phrase in the draft is flagged with a reason
- At least three flagged lines are rewritten using a specific, sourced detail