AI prompts

Account Brief Skill File: A Claude and ChatGPT Instruction File for Named-Account Briefs

For sellers who brief themselves on large accounts again and again.

You leave with a ready skill file, three ways to load it, a document list, a five-step test and a rule for keeping it honest.

By Intandemly team · Updated 28 Oct 2026 · 9 min read · Free, no sign-up

Jump to the first copy block ↓

Preview

On this page

What you'll get

  • 2 copy-ready blocks
  • 3 tables
  • 9 min read
  • Account brief skill file
  • Test steps

How to use it

  1. Step 1: Copy the skill file in Part 1 into a plain text document and change the bracketed lines.
  2. Step 2: Load it using one of the three routes in Part 2.
  3. Step 3: Gather the documents for one account from the list in Part 3.
  4. Step 4: Run the five-step test in Part 4 on a company you already know well.
  5. Step 5: Only then use it on a real account, and keep the fact check in Part 5.

Most sellers who use an AI assistant for research retype a prompt each time, and each time it comes out a little different. A skill file fixes that. It is a short set of standing instructions that you save once and use for every account. This kit gives you one for a one-page account brief, tells you where to put it in Claude or ChatGPT, and shows you how to test it before you trust it. The file makes no calls and finds nothing on its own. It only works on the documents you give it.

Part 1: The skill file

Copy everything between the lines. Replace the lines in square brackets with your own.

Account brief skill file

Account brief skill file
Name: Account brief
Description: Use when I ask for a brief on a company I plan to sell to. Builds a one-page account brief only from documents I give you, and says what it could not find.

About me: I sell [what you sell] to [who buys it]. Keep this in mind when you decide what matters, but never state it as a fact about the company.

Rules
1. Use only the documents I paste or attach in this chat. If I have given you none, ask for them and stop.
2. Do not use your memory for any fact about the company: figures, names, dates, products or customers.
3. Put a tag after every fact: [document name, page or section]. A fact without a tag does not go in the brief.
4. If the documents do not say it, write "Not found". Do not guess and do not fill the gap with something likely.
5. Keep what the documents say apart from what you infer. Inferences go in section 5 only and each starts with "Inference:".
6. Do not write messages to the company. This file produces a brief and nothing else.

Output, in this order
1. The account in one line: what the company sells and to whom.
2. What has changed: up to five facts from the last two years that show direction. Each one tagged.
3. Priorities in the company's own words: up to three short quotes, tagged.
4. People named in the documents: title, name if given, and what the document says they own. Nobody the documents do not mention.
5. Where we might matter: up to three inferences, each tied to a tagged fact above.
6. Not found: what I would need to know that the documents do not say.
7. Check first: the five facts a person should check against the source before anyone uses them, ranked by what a mistake would cost.

Style: short sentences, plain words, British spelling. No figures unless tagged.

Rules 3 and 4 do most of the work. A tag lets you check a fact in seconds. "Not found" stops the assistant from writing something that sounds right.

Part 2: Three ways to load it

Tools change their menus often, so follow your tool's current help for where each option sits. The idea is the same in each.

RouteWhat you doBest when
Saved skillIn Claude, a skill is a folder holding a file named SKILL.md. The Name and Description lines go at the top, the rest followsYou use Claude and want the file loaded only when you ask for a brief
Project instructionsPaste the file into the instructions of a Claude project or a ChatGPT project, then keep each account's documents in itOne person works one set of accounts
Custom assistantPaste the file into the instructions of a custom GPT or similarA team wants one shared assistant with the same rules

Whichever you pick, check what your plan does with the documents you upload. Lesson 5 of AI for Account Research has a paste list for this. Public documents are the safe start.

Part 3: The documents to give it

The file is only as good as what you feed it. Give it documents that come from the company itself or from a source you can name.

DocumentWhere it comes fromWhat it gives the brief
Annual report or results statementThe company's investor pagePriorities, spending, risks
Latest press releasesThe company's news pageWhat has changed
Careers page, saved as textThe company's websiteWhere it is hiring, which hints at direction
Leadership page, saved as textThe company's websitePeople and titles
Your own call notesYour notesWhat a person told you, tagged as notes

Paste the text, or attach the file. Name each document before you paste it, such as "Annual report 2025", so that the tags make sense. Five good documents beat twenty random ones.

Part 4: The five-step test

Run this on a company you already know well, so you can tell if the brief is right.

Test steps

Test steps
1. Give it the documents for the company you know and ask for a brief.
2. Check every tag. Open the document at the page given. Does it say that?
3. Look at section 6. Did it write "Not found" for something you know is missing? If it filled the gap, tighten rule 4.
4. Start a new chat with no documents and ask for a brief. It should ask for documents and stop.
5. Ask it to write an email. It should decline, because rule 6 says so.

If any step fails, change the rule and run the test again. Do not use the file on a real account until all five pass.

Part 5: Keeping it honest

A skill file is a rule for the team, so treat it like one.

  • One owner. One person edits the file. Everyone else suggests changes.
  • A version line. Add "Version 1, [date]" to the bottom and change it when you edit, so nobody uses an old copy.
  • Section 7 is not decoration. Before any message goes to a Tier 1 account, a person checks the five facts it lists against the source.
  • Keep it short. Every extra rule is another chance for the assistant to ignore one. If the file passes three screens, cut it.

Lesson 4 of AI for Account Research gives the routine for a full fact check.

Worked example (illustrative, not a real company)

Illustrative example

Quaywise sells warehouse software. Sam Carter loads the skill file, then pastes two documents for Northfield Logistics: "Annual report 2025" and "Press releases 2026". All names and facts are invented.

SectionWhat the brief saysWhat Sam checked
1Northfield moves goods for retailers across three regions [Annual report 2025, p. 4]Page 4 says so
2Opened a second distribution site in 2026 [Press releases 2026, item 2]Item 2 says so
3"Reduce handling time at every site" [Annual report 2025, p. 11]Quote is exact
4Dana Whitfield, Head of Operations, named as owner of the site programme [Press releases 2026, item 2]Title matches
5Inference: the new site may need better stock visibility, tied to the second site [Press releases 2026, item 2]Marked as a guess
6Not found: current warehouse software, budget, who signs offTrue. Sam adds these to his call plan

Sam spent five minutes on the checks. The brief told him what he did not know, which is the part he wanted most.

Five mistakes to avoid

  1. Asking for a brief with no documents. Without them the assistant fills the page from memory. Rule 1 exists to stop that.
  2. Trusting a tag without opening it. A tag can point to a page that does not say the thing. The test and section 7 are there for that reason.
  3. Letting the file write the email. A brief and a message are separate jobs. Use the brief, then write to a person.
  4. Piling on rules. Ten more rules do not make a better brief. They make a file the assistant follows less well.
  5. Using it on every company. For a list of thousands this is too slow. It belongs on the accounts where one sharp message is worth the time.

Why this works on named accounts

A prompt you retype for every account drifts. One day it asks for sources and the next day it forgets. A saved instruction file does not drift, so every brief on your list comes out in the same shape with the same rules. That matters most on named accounts. When you work forty companies, the person who reads your message may know their own company better than the assistant does, and a brief that guessed will show. A file that says 'use only these documents, tag every fact, write Not found when it is missing' is how forty briefs stay as careful as the first.
How account-based selling works →

Common questions

What is a skill file in Claude?

In Claude, a skill is a saved set of instructions, kept as a folder with a file named SKILL.md at its centre. The Name and Description at the top tell the assistant when to use it. Check Claude's current help for the exact steps, as the product changes.

Does it work in ChatGPT?

Yes, in the sense that the rules are plain text. Paste the file into the instructions of a project or a custom GPT. The Name and Description lines are only labels there, so they do no harm.

Can the assistant search the web for the brief?

You can let it, but the file as written says to use only the documents you provide. If you allow web search, change rule 1 and ask for a link as the tag for each fact. Then check each link yourself.

More from the Academy

AI prompts8 min read

15 AI Prompts for Account Research

For sellers who use Claude, ChatGPT or Perplexity to prepare for an account.

You leave with a set-up prompt, 15 research prompts in working order, a fact check that catches invented details and a sample of its output.

AI prompts5 min read

AI Prompt Pack: Read an Annual Report in 10 Minutes

For sellers researching a large account before first contact.

You leave with 7 prompts, a 10-minute timetable, a one-page brief format and one brief written out.

AI prompts6 min read

Fact-Check AI Research Prompts

For anyone who sends AI-researched claims to real buyers.

You leave with three prompts (extract the claims, demand the sources, cross-examine the brief), the risk tiers and a five-minute routine per account.

ToolRuns in your browser

Buying Committee Mapper

Map the people behind a decision by role, stance and influence, and see who to approach next. Runs in your browser.

Want this in your inbox?

Optional. We use your email only to send you this, nothing else.

Want this run on your key accounts?

Tell us which accounts you need to open. We will show you how we would work them.