Resource

Account research prompt

This is the prompt from the video, written as one block so you can paste it in one go. Fill in the bracketed parts, turn on web search, and send it. ChatGPT, Claude, and Gemini all handle it.

The prompt
You are a sales research analyst preparing me for a first meeting.

My context: I'm an account executive at a company that sells [WHAT YOU SELL]. My meeting is with the [TITLE OF THE PERSON YOU'RE MEETING] at [COMPANY], a [ONE-LINE DESCRIPTION OF THE COMPANY].

Research this company and produce a one-page meeting brief with these sections:

1. Facts: what they do, how they make money, and recent developments from the last twelve months. Give a source and a date for each item.
2. Hypotheses: what a [TITLE] there is likely measured on right now, and whether [WHAT YOU SELL] might be a priority for them. Label every item in this section as a hypothesis.
3. Five discovery questions that test those hypotheses without assuming a problem exists, ordered from safest to boldest.

Rules:
- Search the web and cite your sources for anything factual
- If you can't verify something, say so instead of guessing
- Never invent numbers, names, or quotes
- Flag anything that might be outdated
- If you find no evidence that [WHAT YOU SELL] is a priority for them, say that plainly

Everything in brackets is yours to fill in. Leave the rules alone. They tell the model to show its sources, flag what it can't verify, and say when it comes up empty. Without them the brief is a lot less useful.

Before you use the brief

A brief like this looks finished the moment it appears. It isn't. Run these three checks before anything goes into your meeting notes. They take a few minutes.

1

Open one source

Take one item from the facts section and click through to the source. Confirm the page says what the brief claims, and check the date. If it holds up, check a second one and move on. If the page says something different, or the date is two years old, reread the whole section with that in mind. A citation tells you where the model looked. It doesn't tell you whether the model read the page correctly.

2

Keep the hypotheses as hypotheses

The second section is the model's best guess at what this person is measured on and why what you sell might matter to them. Some of it will sound right. Treat all of it as unconfirmed, because it is. Turn each item into a question for the meeting, and let the person across the table tell you whether the guess was any good.

3

Sharpen one question

Read the five questions and look for one that assumes the company has a problem. Rewrite it to ask how they handle that area today. Their answer will tell you whether anything hurts. If none of the five assume anything, take the strongest one and tie it to a specific fact from the first section. It should be obvious you did the reading.

If it comes back and says it found nothing

That's the last rule working. The model looked for public evidence that what you sell is on their radar. None was found, and it said so instead of inventing a reason. You still have the meeting. Walk in with questions instead of a pitch.

From Signal Driven

Signal Driven coaches small B2B sales teams to build AI workflows like this one and run them on their own.

If you're not sure which of your team's workflows to start with, that's what the Workflow Diagnostic is for.

About the Workflow Diagnostic →