Step 1 — I write the prompt
The position goes in first
AI doesn't have a point of view. I give it one, then ask it to map the angles that follow.
Step 2 — I pick the angle
Five angles, all safe
- Features vs outcomes
- The language gap: "collaboration" hides the chaos
- What buyers search vs what vendors publish
- When a feature solves an unnamed problem
- Reverse-engineer feature copy to issue-led
The bold one, if the brand will back it
Sub-point 2, the language gap. It names something vendors do and don't admit. The other four are true, but nobody argues with them. I check appetite with the client before writing, because a position only works if they will stand behind it in public.
Step 3 — I pull the thread
Accurate, and completely flat
The same point, put in a real moment
Step 4 — I go find the proof
Research comes after the argument, so I'm only looking for what supports it or breaks it.
Grok + Perplexity
Live data, recent stats, industry signals
GPT + Gemini
Named examples, analogies, case references
Claude
Stitches data and examples into clean sentences
Step 5 — I check what survived
It reads as one piece now
What I changed before it shipped
AI handled the shape. I cut "nobody in business", which claims more than I can prove, and I swapped the derived 83% for the 17% Gartner actually reports. Hit publish at the end to read the version that went out.
Step 6 — I align to brand tone
Flags off-brand language
My argument, their voice
I accept or override. The position is mine, the way it sounds is theirs, and both have to be true before it goes out.
Step 7 — I ask three questions
Three questions, every time
- Is it specific? → "Three days late, account at risk" ✓
- Does it sound like the brand? → Conversational, no jargon ✓
- Is there a line they'll repeat? → "Collaboration is what vendors say when they won't name the real thing" ✓