How to Use an Analyst for Positioning Brainstorms

A positioning brainstorm with three PMMs in a Zoom room produces, on average, the same five ideas you've written down before. Not because your team is unimaginative — because the room is small, the priors are shared, and the first plausible answer absorbs the next thirty minutes. By "analyst," we mean a working AI sparring partner — a model with context on your category, your competitors, and your past messaging, used as a structured interlocutor rather than a search box.

The point of bringing one in is not to outsource the thinking. It's to break the priors.

The brainstorm fails when the room agrees too early. The analyst's job is to keep disagreement alive long enough to find the second idea.

Why most positioning brainstorms converge on the obvious

The pattern is consistent across the audits we've run. A team blocks 90 minutes, opens a Miro, and within the first 20 minutes someone says the thing — "we're the [adjective] [category] for [segment]" — and the rest of the session is everyone refining that sentence. The fluency feels like progress. It usually isn't.

Three things cause the convergence:

• Shared inputs. The team reads the same competitor pages, the same analyst notes, the same win/loss decks. They walk in with overlapping mental models.
• Social cost of dissent. Pushing back on the VP of Product's framing in week two of the launch cycle is expensive. The best ideas die in the chat, unsent.
• No external counterweight. There's nobody in the room whose job is to argue the opposite, generate ten worse ideas, or surface the framing you're systematically not seeing.

An analyst — used well — fixes the third one. It doesn't fix the first two, but it gives the room something to react to that didn't come from inside the room.

The setup that actually works

The mistake most teams make is treating the analyst like a search engine: "give me five positioning angles for our product." The output is generic because the prompt is generic. The analyst has no context, so it generates the median answer for your category — which is, by definition, the same answer your competitors already use.

The setup that works has three parts:

What "load the context" looks like in practice

Vague context produces vague output. The analyst needs roughly the same brief you'd give a new PMM on day one.

The last item is the one most teams skip and the one that matters most. Without it, the analyst will cheerfully regenerate the angle you killed eight months ago.

The prompts that move past the obvious

A handful of prompt structures, used in sequence, reliably produce angles a six-person Zoom room wouldn't reach:

• The inverse prompt. "If we wanted to position ourselves as the worst fit for our ICP, what would we say?" Reveals which claims your current positioning is implicitly making and which it could disown.
• The competitor-mirror prompt. "Write our competitor's homepage as if they were positioning against us. What's the gap they'd attack?" Surfaces the soft underbelly of your own claim.
• The five-years-ago prompt. "What category did our buyer think they were buying five years ago? What do they think they're buying now? What's the next shift?" Forces a temporal frame that static brainstorms miss.
• The customer-language prompt. "Here are six verbatim quotes from win/loss calls. What category noun does the buyer keep reaching for? What words do they avoid?" The buyer's vocabulary is almost never the vocabulary on your homepage.

What the analyst is bad at

Worth naming, because the failure modes are predictable:

• It doesn't know your customers. It knows the median customer in your category from public data. The specifics — the CTO at a 400-person fintech who keeps asking about SOC 2 timelines — are yours to bring.
• It will hedge when you want a take. Models are trained to be balanced. Force a position by asking "if you had to bet, which angle wins?" Then argue with the bet.
• It can't tell you what's true. It can tell you what's plausible, what's differentiated, what's defensible given the inputs you provided. Whether the claim survives a customer call is empirical work you still have to do.

What to do Monday

Block 90 minutes. Don't invite anyone. Spend the first 30 loading context — homepage copy, competitor copy, ICP, win/loss, past rejected angles. Spend the next 30 generating ten angles ranked obvious-to-improbable, then arguing each one against itself. Spend the last 30 writing up the three angles you'd actually be willing to defend in a launch meeting, with the strongest counter-argument under each.

Then — and only then — bring the room in. The brainstorm isn't the meeting; the meeting is where the team kills the bad angles and sharpens the survivors. The analyst earned its keep if the room is now arguing about angles it wouldn't have generated on its own.