LinkedIn ICP Builder Prompt
An ICP written in the vocabulary LinkedIn search understands, derived from deals you actually closed rather than from who you wish would buy. Most ICPs are a paragraph of adjectives: "mid-market companies that care about efficiency." That cannot be typed into a search box. This prompt converts your won and lost deals into headcount bands, industries, titles, regions and one observable trigger — the five things Sales Navigator can actually filter on — and states what it could not infer instead of inventing it.
You are a revenue operations analyst. Your job is to turn closed-deal evidence into a targeting definition that can be typed into LinkedIn Sales Navigator without further interpretation.
## Objective
Produce an ideal customer profile for {{COMPANY}} that another person — or a search tool — could act on without asking me a single follow-up question.
## What I sell
{{PRODUCT}}
## Evidence
Below are my recent deals. For each one I have given what I know; some fields will be missing.
[PASTE 10–20 DEALS. One line each: company, headcount, industry, country, buyer title, won or lost, deal size, and — if you know it — what changed at that company shortly before they bought.]
## Exclusions I already know
{{EXCLUSIONS}}
## Method
1. Separate the won deals from the lost ones before looking for patterns. A pattern that appears in both is not a qualifier, it is background.
2. For each attribute, state the RANGE the wins cluster in, not the average. "200–800 staff" is actionable; "average 470 staff" is not.
3. Identify the trigger: the observable event that preceded the wins. It must be something visible from outside the company — a funding round, a new hire in a specific role, an office opening, a product launch, a regulatory date, a job posting. "They realised they had a problem" is not observable and does not count.
4. Where the evidence does not support a conclusion, say so. An ICP with three confident attributes and two honest gaps is more useful than five confident-sounding guesses.
5. Rank the attributes by how much they separate wins from losses. The top two are the ones that go in the search; the rest are tiebreakers.
## Constraints
- Every attribute must map to something LinkedIn can filter or something visible on a company page or profile. If an attribute cannot be observed on LinkedIn, put it in the "qualify on the call" section instead of the ICP.
- Do not invent industry categories. Use LinkedIn's own industry names.
- Headcount must be expressed in LinkedIn's bands: 1-10, 11-50, 51-200, 201-500, 501-1000, 1001-5000, 5001-10000, 10001+.
- No personas, no jobs-to-be-done narrative, no empathy map. This is a targeting document.
## Output format
Return exactly these sections and nothing else.
### ICP in one sentence
One sentence. It must contain a headcount band, an industry or category, a geography and the trigger.
### Firmographics
| Attribute | Value | Confidence | Evidence |
|---|---|---|---|
Confidence is high / medium / low and must reflect how many wins support it.
### Buying committee
| Title | Role in the deal | What they are measured on |
|---|---|---|
Economic buyer, user and blocker, at minimum. Use the titles that appeared in my deals, not generic ones.
### The trigger
The observable event, how to spot it on LinkedIn, and how long the window stays open after it happens.
### Exclusions
Who to leave out and why. Include anything my evidence shows losing, not just what I told you.
### What I could not determine
List the attributes the evidence was too thin to support, and say what data would settle each one.
### Search-ready summary
A short block I can paste into the next prompt: headcount band, industries, titles, geographies, trigger, exclusions.- You are about to build your first Sales Navigator search and do not want to guess the filters.
- Your reply rate is fine on some accounts and zero on others, and you cannot articulate the difference.
- You are briefing an agency, an SDR or an AI agent and need the target definition in writing.
- You are re-doing an ICP that was written before you had customers.
- Export your last 10–20 closed deals — won and lost. Lost deals are half the signal and the half everyone leaves out.
- Paste them in place of the bracketed block. One line each is enough; the model does not need your CRM schema.
- Fill the three variables. If you have no exclusions yet, write "none known" rather than deleting the section — it makes the model ask for them at the end.
- Run it. Read the "What I could not determine" section first: it tells you whether the rest is worth trusting.
- Keep the search-ready summary. It is the input to the Sales Navigator Search Builder and to ICP.md.
The output schema it demands
The single highest-leverage part of any prompt. Without a declared shape you get prose you then have to reformat; with one, you get sections you can act on — and you can tell at a glance when the model skipped something.
### ICP in one sentence ### Firmographics (table, with confidence + evidence) ### Buying committee (table) ### The trigger ### Exclusions ### What I could not determine ### Search-ready summary
What it looks like filled in
Illustrative, not a promise — this is the shape of a good answer rather than a guarantee of one. The scenario is the same fictional company used across the whole toolkit.
Cadence — payroll reconciliation for multi-entity finance teams. 14 deals pasted: 9 won, 5 lost. Exclusion given: "not single-entity companies".
ICP in one sentence — Companies of 201–2,000 staff in Transportation & Logistics or Retail operating payroll in three or more countries, within nine months of adding an entity or replacing a finance system.
The trigger — A new entity or acquisition announced on the company page, or a "Group Financial Controller" / "Payroll Manager" hire posted in the last 120 days. Spot it via the company's LinkedIn posts and their open roles. Window stays open roughly two quarters; after the first close with the new structure, urgency drops sharply.
What I could not determine — Whether ERP matters. Three wins were on NetSuite, two on SAP, four unknown. To settle it, record the finance system on every deal from now on.
Do not do these
Each one is a real failure mode, not a disclaimer.
- Feeding it only won deals. Without losses the model has nothing to contrast against and will describe your customer base rather than your ideal one.
- Accepting a trigger you cannot observe. "They are scaling" is not a trigger; "they posted two roles in the same function in 60 days" is.
- Letting it produce more than five firmographic attributes. Anything below the top five does not change who you contact.
Attach these once and stop retyping your business
This prompt works pasted into a blank chat. It works considerably better inside a Project with these attached, because then the ICP, the product and the limits are already in context and every answer inherits them.
This is one step in a sequence
The output of the step before is the input to this one, and this one's output is the next one's input. That is the whole reason these are not independent pages.
Before you run it
How many deals does it need?
Ten is the floor for the model to find a pattern rather than describe the deals back to you, and about twenty is where extra rows stop changing the answer. Below ten, run it anyway but treat every confidence rating as low and the trigger as a hypothesis.
What if I have no closed deals yet?
Then this is the wrong prompt and the honest substitute is worse: pick the ten companies you most want as customers, write why for each, and treat the result as a hypothesis to be killed. Do not dress that up as an ICP — run the search, message twenty, and rebuild it on evidence within a month.
Why does it insist on LinkedIn headcount bands?
Because Sales Navigator does not have a free-text headcount field. "200–800 staff" has to become 201-500 plus 501-1000 before it can be searched, and doing that conversion at the ICP stage stops the same argument happening every time someone builds a list.
The reference behind this prompt
Some of the filters an ICP naturally produces — headcount growth, department headcount, job-posting signals — only exist on Advanced. Worth knowing before you build the search.
Your ICP size has to be reconcilable with about 100 invitations a week. An ICP of 40,000 accounts and one seat is a seven-year plan.
Written to a fixed structure and run through Claude and ChatGPT to confirm each one returns the output format it declares. That is not the same as a measured reply-rate benchmark, and nothing here claims one — if you run these at volume and have numbers, we want them.
Take the whole folder instead
This prompt, the files it expects and the rest of the sequence, as a zip you can drop into a Claude Project, a Cursor workspace or an agent repo. Free, no account.