Your prompt

Two answers in, one complete brief out.

Describe what you sell and who you think it is for. The page assembles a full research-and-reasoning brief around your answers — nothing is sent anywhere, and no AI runs here. You paste the result into ChatGPT, Claude, Gemini or whichever assistant you use, and that tool writes the ICP.

What do you sell, and what problem does it solve? Describe the product or service, the problem it solves and the outcome customers pay for.

Who do you think it is for? Add industries, company size, markets or roles if you know them. ‘Not sure yet’ is fine.

Your website optional A link for your AI tool to use if it has browsing. This page does not read the website.

Add evidence or constraints — optional

Which customers have been a good fit? Share one or two examples: what they bought, why, and what improved. Add a poor-fit example if useful. You can anonymize names.

What should the ICP respect? For example: countries served, minimum contract size, required tools, delivery capacity, sectors to exclude or sales-cycle limits.

ICP output language Choose the language of the answer your AI tool should produce.

Answers stay in this tab and clear when you refresh. Nothing you type is sent to us or to any AI service by this page.

Your ICP prompt

Build the prompt and it appears here, ready to copy. It asks your AI tool for five things:

Company fit: Which accounts qualify, and why
User persona: Who uses the solution or operates the workflow
Buyer persona: Who approves, funds or influences the decision
Buying signals: What suggests a reason to act, and how to verify it
Activation: Exclusions, targeting fields and next validation steps

An ICP is a working hypothesis. Use customer conversations, won and lost deals, and delivery outcomes to improve it.

Framework notes

Inspired by the ICP framework shown in Jan Rasmussen’s arrel.ai diagram. Revenue Puzzles adapts it into a prompt workflow with evidence checks, exclusions and validation steps.

01 · How it works

Add context. Build your prompt. Copy it into your AI tool.

Add context. Your offer and the audience you have in mind are enough. A website link, a couple of good-fit customers and any hard constraints make the result sharper, but they are optional. ‘Not sure yet’ is an accepted answer for the audience.
Build your prompt. The page wraps your notes in a fixed, versioned brief: how to treat the evidence, what not to invent, and the six sections the answer must contain. Identical answers always produce an identical prompt.
Copy it into your AI tool. Paste the prompt into the assistant you already use. If it can browse, it may read your website; if it cannot, it says so and works from your notes. Either way it produces a first ICP with its assumptions labelled.

Scope

The result is a prompt. Your AI tool creates the ICP when you paste it there. This page does not read your website, score accounts or store what you type.

02 · What the prompt asks for

Six sections, evidence labelled, unknowns kept visible.

The brief tells the AI to act as a practical B2B go-to-market strategist and to deliver a working first version — roughly 1,200 to 1,600 words with concise tables — in this order:

Section	What it must contain
1 · ICP hypothesis	A one-sentence definition, the primary segment and why it fits, the buyer’s desired outcome, and whether the conclusion rests on evidence or is a hypothesis.
2 · Ideal company criteria	An account-fit table: criterion, proposed definition, must-have / preference / disqualifier / to validate, why it matters, evidence status and how to verify. Observable criteria, no “innovative companies”.
3 · User and buyer personas	Side by side across seven rows: who they are, jobs to be done, day-to-day work, what makes it hard, operating and buying environment, research and evidence, change and intent.
4 · Signals and purchase timing	Four to six signals with what each may indicate, where to observe it, how fresh it must be, the alternative explanation and the next action. Fit, problem, change event and buying intent kept apart.
5 · Who not to target	Three to five concrete poor-fit conditions, each marked as an explicit exclusion, a disqualifier to validate, or simply wrong timing.
6 · Make the ICP usable	A target-account checklist, five to eight CRM or research-sheet fields, one message angle per persona, a validation plan, and exactly five follow-up questions.

The four evidence labels

Provided — it is in your notes.
Sourced — supported by a source the AI actually consulted, cited beside the claim.
Hypothesis — the AI’s proposed inference, to be validated.
Unknown — not enough information; the prompt asks how to find out rather than allowing a guess.

The prompt also forbids invented revenue, headcount, budget or sales-cycle ranges, treats hiring, funding and technology use as weak signals until corroborated, and refuses to assume the user is a salesperson or that the goal is pipeline.

03 · The framework behind it

Company, user, buyer, signals — adapted to any offer.

The brief rests on a four-part ICP structure: ideal company criteria (firmographics, stage and market, fit), a user persona (who operates the workflow), a buyer persona (who decides), and the signals that show intent now. The original diagram was drawn for outbound sales teams, so several of its examples are about generating pipeline. The prompt keeps the structure and generalises the examples.

In the source	In the prompt
Company size, expansion, revenue evidence, sales capacity	Scale, maturity, ability to pay and capacity to adopt the offer. Sales capacity only where relevant.
Growth constraint, CRM and data environment	The problem limiting customer outcomes, and the tools or data the solution needs to work.
Industrial, software and services fit	Examples, not defaults. No segment is included unless your context gives a reason.
User persona: “who generates pipeline?”	“Who uses the solution or operates the workflow?” — the seven persona dimensions stay.
Buyer persona: “who makes the decision?”	Kept, with beneficiary, champion, budget owner, approver, evaluator and blocker told apart when they differ.
Hiring, engagement, tech stack, website visits	Signals to investigate, never proof of intent on their own.

What Revenue Puzzles adds: explicit disqualifiers, economic and delivery fit, an evidence status on every claim, visible unknowns, practical targeting fields, and a validation plan built from real wins, losses and delivery outcomes.