// The Method

The Prompt Stack.

A four-stage method I built for getting useful answers from AI on investment decisions. Asking "what do you think of this stock?" turned out to be the worst possible way to use it — so I built something that forces evidence before opinion, and makes the final answer earn its place.

// Four stages. Run in order.

Each stage is an instruction to the AI — and a checkpoint for you — before moving to the next one.

01ROLE: Be a cautious analyst, not a cheerleader.

02FILTER: Separate observable facts from assumptions.

03RISK: Find what could go wrong, when, and what would prove the view wrong.

04VERDICT: Give one practical action and a confidence level.

// 01 — ROLE

Set the stance before the question.

AI defaults to agreeable. Ask "what do you think of XYZ?" and you get a balanced-sounding summary that reads, on closer inspection, as mildly positive about almost anything. The first stage hands the AI a stance it has to argue from before it sees your question.

My default is "cautious analyst, not a cheerleader." For specific jobs I narrow it — "value-oriented short-seller", "compliance officer reading this filing for the first time", "risk committee member who has to defend approving this trade." Most of the work the rest of the prompt is meant to do gets done here. With the right role set, the AI stops trying to be helpful to a fault.

// 02 — FILTER

Separate facts from filler.

Most AI output on a company looks like analysis and is actually narrative. The filter step asks for two lists: what's in the source material, and what the AI has inferred or assumed on top of it.

The instruction I use: "List the observable facts you're drawing on. Then separately, list the inferences you're making from those facts. Mark anything that's neither — anything you're asserting without a basis — and remove it." It does more work than any other step in the method, and what comes out in the cut is usually the bit that sounded most confident going in.

// Meta Q4 2024 — filter step applied

// Observable — kept

Revenue $48.4bn — up 21% year-on-year
Operating income $23.4bn — margin 48%
Family of Apps MAU: 3.35bn
Reality Labs operating loss: $5.0bn
Q1 2025 guidance: $39.5–41.5bn

// AI inferences — flagged

"continued dominance in digital advertising"
"exemplary management execution"
"unassailable position in social media"
"Reality Labs showing encouraging progress"
"signals strong management confidence"

The strikethrough items are what a typical AI response adds on top of the source. The filter step removes them before you act on anything.

// 03 — RISK

Make the downside explicit.

"What are the risks?" gives you a checklist, and a checklist isn't what this stage is for. I ask for three specific things: what could go wrong on this timeframe, what would have to be true for the whole view to be wrong, and what I'd actually see in the next 90 days if it was going wrong.

The third question matters most. A risk you can't spot in advance is just background worry. A risk with a specific, observable warning sign is something you can act on. If the AI can't give you one of those, it doesn't really have a view — it's pattern-matching off the consensus.

// 04 — VERDICT

One action. With a confidence level.

By this point there's something worth reading. The verdict step is short on purpose: one practical action, a stated confidence level — low, medium, or high — and a sentence on why. "Hold and reassess after Q3 results, medium confidence — the thesis depends on margin recovery that won't be visible before then." That shape, no more.

The confidence label is what keeps the rest honest. An AI that has to commit to "low confidence" is more useful than one allowed to hide behind a balanced summary, and the label gives you something to check against later: was it right when it sounded sure, and what happened when it wasn't?

// Why it works

Discipline, not magic.

The Prompt Stack isn't a clever trick to get better answers out of AI. It's a way of refusing to take the first one. Each stage gives you a place to stop and check the working before the next step gets built on top of it.

Most of what the method does is slow you down at the points where you'd otherwise jump to a conclusion — usually the points where the AI sounded most confident and the evidence was thinnest. That's where the useful work happens.

// Built by Ben Dixon

I run this method against my own investment portfolio — real money, real decisions, no demos. The Prompt Stack is the system I landed on after trying most of the obvious alternatives and finding them wanting. More about me →

// Use it this week

Get the prompts. Free.

I've put the Prompt Stack into five copy-paste prompts for the moments that actually matter — when a stock has moved and you need to decide what to do next. Drop your email and they're yours immediately.