Start Here · Before The Floors

The GCES Method

Goal · Context · Expectations · Source

Every great prompt on this site is built the same way. Master one simple formula and you stop getting generic AI answers — and start getting output that reads like it came from your sharpest analyst. Four parts. That's the whole method.

The difference between a beginner and a power user isn't the AI — it's the prompt. A vague ask gets a vague answer. GCES turns a one-line question into a precise brief the AI can actually deliver on.

The Four Parts

G Goal

What do you want?

State the actual outcome — the thing you want produced. Not the topic, the deliverable. "Summarize this" is a topic. "Give me a 4-bullet summary I can paste into a client email" is a goal.

Ask yourself: What would a finished, useful answer look like?

Weak: "Tell me about this stock." Strong: "Give me a balanced bull/bear case I can present to my investment committee."

C Context

What should it know?

Set the scene. Who are you, who's the audience, what's the situation, what are the constraints? The AI has no idea you're a commodities PM presenting to a risk committee — until you tell it. Context is what makes the answer yours, not generic.

Ask yourself: What would a new colleague need to know to do this for me?

"I'm a sell-side analyst. The audience is a sophisticated PM who already knows the sector. We're 30 minutes from market open."

E Expectations

What should it look like?

Define the shape of the answer: format, length, tone, structure. Table or bullets? 100 words or 500? Punchy or formal? Ranked by impact? This is where you stop the AI from rambling and force it into the exact format you'll actually use.

Ask yourself: How will I use this, and what format fits that?

"Output as a table. Rank by likely market impact. Keep it under 200 words. No hedging — take a position."

S Source

What should it work from?

Give it the raw material — the report, the transcript, the numbers, the email thread. And tell it how to treat that material: stick to what's provided, cite the data point behind each claim, and flag anything it's inferring vs. reading directly. This is what keeps AI honest in a field where a made-up number is a real problem.

Ask yourself: What information should it use — and how do I keep it from guessing?

"Use only the earnings release I'm pasting below. Cite the line behind each claim. Mark anything you're inferring."

See It Work

Same question, two ways. On the left, how most people ask. On the right, the same ask built with GCES. The AI is identical — only the prompt changed.

Without GCES

What do you think about Apple's earnings?

The problem: No goal, no context, no format, no source. You'll get a generic, hedge-everything paragraph that could've been written a year ago — and you can't tell what's fact vs. guess.

With GCES

I'm a portfolio analyst briefing a PM before the open. [C] I'll paste Apple's latest earnings release below. [S] Give me a teardown: (1) beat/miss vs consensus on revenue and EPS, (2) the 3 numbers that actually matter for this name, (3) the one line management hopes I skim past. [G] Be fast and blunt. Under 200 words. Cite the figure behind each point and flag anything you're inferring. [E]

The result: A specific, sourced, format-ready brief you can act on in seconds. Every part of GCES is doing a job — and it's labeled so you can see the formula.

Your Reusable Template

Copy this, fill in the brackets, and you've built a strong prompt for almost anything. Keep it handy until the four parts become second nature.

GCES Prompt Template

[Context] I'm a [your role] working on [situation]. The audience is [who]. The constraints are [time / scope / rules]. [Source] Use the following material: [paste your report / data / email / transcript]. Stick to what's here, cite the figure behind each claim, and flag anything you're inferring vs. reading directly. [Goal] I want you to [the specific deliverable — what a finished, useful answer looks like]. [Expectations] Format it as [table / bullets / memo]. Keep it [length]. Tone: [punchy / formal]. [Rank by X / take a position / show your math].

Power Moves

Once you've got GCES down, these small phrases punch far above their weight. Drop them into the Expectations or Source part of any prompt.

"Don't be polite."

Forces the AI to red-team your idea instead of flattering it. The single fastest way to get real pushback on a thesis.

"Show your math."

Non-negotiable for anything quantitative. Makes the AI expose its reasoning so you can audit the logic instead of trusting a number.

"Flag what you're inferring."

Separates fact from guess. Critical in finance, where presenting a hallucinated figure as fact is a genuine problem.

"Take a position."

Stops the endless on-the-other-hand hedging. Forces a conclusion you can actually act on or argue with.

"Rank by impact."

Turns a flat list into a prioritized one. The AI decides what matters most — and usually gets the ordering right.

"Under [N] words."

A hard length cap is the simplest cure for rambling. It forces the AI to compress to the point.

Now Take It To The Floor

You've got the method. Every scenario on every floor is a GCES prompt you can copy, use, and learn from. Pick your floor and start practicing.

Choose Your Floor →