← Blog

Sell ·

ICP and Prospect Scoring Prompts for Founders

Five copy-paste prompts that build your ideal customer profile from real past buyers and score real prospects, with honesty rules inside every prompt.

By , Founder, Baljia AI

The short answer

An ideal customer profile is worth exactly as much as the real buyers it came from. The prompts below build yours from your actual past customers, then score a list of 20 real prospects against it, with a reason for every score.

What makes this pack different: the honesty rules live inside the prompt text. Every prompt instructs the AI to mark unknowns as unknown, never invent a person, company or number, and count only unique real entries. Most packs skip that, and the AI fills every gap with something plausible.

We make Baljia, an AI employee for founders and small businesses; it appears briefly near the end. The prompts work in any AI assistant you already use.

Gather this before you paste anything

Ten minutes of homework, or the prompts have nothing real to work with.

Your best customers, 5 to 10 of them. Best means fastest to say yes, still with you, and least painful to serve. For each, write down what you actually know: industry, company size, who signed, what they did before you, how they found you, how long the sale took, and anything they said about why they bought. Where you do not know, write "unknown". That word is load-bearing in everything below. Add 2 or 3 customers who churned or stalled, with what you know about why; prompt 2's disqualifiers need them.

Your prospect list, up to 20 rows. Real people you are considering contacting: name, title, company, what the company does, size if you know it, and one line of evidence for why they might fit. If you cannot write the evidence line, the row is not ready; the method behind that line is in how to get customers from LinkedIn. No customers at all yet? Run the one-day exercise for finding people asking for your product first and come back.

One privacy rule: never paste prospects' emails or phone numbers into a third-party AI tool. Names, companies and public evidence are enough.

Prompt 1: extract the pattern from your best customers

You are helping me find the pattern in my best customers. Below are my real
customers, with the facts I know about each.

Rules you must follow:
- Use only the customers and facts I pasted. Do not add, merge or invent any
  company, person or number.
- If a field is missing or unclear, treat it as unknown. Never fill a gap
  with a guess.
- Count only unique real entries. If the same company appears twice, count
  it once and tell me.
- When you state a pattern, name the pasted customers that support it. A
  pattern needs at least three supporters.
- If the data is too thin to support any pattern, say exactly that.

My best customers:
[Paste 5 to 10 customers with the facts you gathered. Keep your "unknown"
entries in; they matter.]

Task: list the strongest patterns you can actually see in this data, grouped
by company traits, the trigger that made them look for a fix, the buyer who
signed, and the words they used about the problem. For each pattern, list
the supporting customers and a confidence of high, medium or low based only
on how many pasted entries show it.

Prompt 2: write the ICP one-pager

Using only the pattern analysis above and the customer facts I pasted,
write my ideal customer profile as a one-pager.

Rules:
- Every line must trace back to the pasted customers. If a section would
  need information I did not give, write "unknown - needs research" instead
  of filling it in.
- Do not invent market sizes, statistics, personas with made-up names, or
  example companies.
- Plain language. No marketing voice.

Sections:
1. Company: industry, size range and stage, as seen in my customers.
2. Buyer: the title that signed, and who else was involved if I said so.
3. Trigger: the events that made these customers look for a fix.
4. Problem words: exact phrases customers used, quoted from my paste only.
5. Disqualifiers: traits of the customers I said churned, stalled or were
   painful to serve.
6. Open questions: the three facts most worth checking with my next five
   customers, chosen from the unknowns you marked.

Prompt 3: score 20 prospects, 1 to 10, with a reason each

Score the prospects below against my ideal customer profile.

Rules:
- Score only the prospects I pasted. Do not add names, drop names, or merge
  two people into one. Count the unique entries and tell me the count you
  scored.
- Use only the evidence in each row plus the profile. If a row lacks the
  evidence a criterion needs, mark that criterion unknown rather than
  assuming.
- Never invent a fact about a company to justify a score. If you are
  reasoning from the company name alone, say so.
- Every reason must point at evidence in the row. "Looks like a good fit"
  is not a reason.

My ideal customer profile:
[Paste the one-pager from the previous prompt.]

My prospects:
[Paste up to 20 rows: name, title, company, what it does, size if known,
and your one line of evidence.]

Task: for each prospect, give a 1 to 10 fit score, a one-sentence reason
tied to the evidence, and the single missing fact that would most change
the score. Then list them in score order and flag any score that rests
mostly on unknowns.

Prompt 4: the disqualify prompt

The most useful prompt in the pack: the expensive mistake is not missing a good prospect, it is spending three weeks on one who was never going to buy.

Now play the skeptic on the same list.

Rules: same as before. Use only the pasted evidence, mark unknowns as
unknown, invent nothing, and point at the row whenever you make a claim.

Task: find the prospects who look good but probably are not, using these
tests:
1. Right company, wrong person: the company fits but this title does not
   feel the problem or cannot sign.
2. Flattering evidence: the "why they fit" line describes something
   impressive about them, not a sign they have the problem.
3. Dressed-up mismatch: they match one loud trait of my best customers but
   miss the trigger that made those customers buy.
4. Wrong size: outside the range where my pasted customers actually closed.

For each flagged prospect, say which test caught them and what evidence
would move them back onto the list. Do not soften the verdicts.

Prompt 5: turn the unknowns into a research list

Build me a research list from the scoring above.

Rules: work only from the unknowns and missing facts you already flagged.
Do not introduce new prospects or new claims about existing ones. When you
point me at a place to verify, name the type of source only; do not state
what it says, do not invent URLs, and mark unknowns as unknown.

Task: for each prospect scoring 6 or higher, and for every prospect flagged
in the skeptic pass, list the one missing fact that most changes the score,
where I could verify it from public information in under five minutes
(their website, their job posts, their public profile), and what answer
would confirm or kill the fit. Order the list so the checks that affect the
most prospects come first.

How to read the output

Trust the unknowns more than the scores. A prospect scored 8 on full evidence is not the same as an 8 built on guesses the prompt refused to make; the flag in prompt 3 tells them apart. Unknown-heavy rows are homework, not pipeline.

Check the count. Prompt 3 reports how many unique prospects it scored; if you pasted 20 and it reports 19 or 21, something merged or split, so fix the paste first.

A pattern with two supporters is a hunch. The three-supporter rule in prompt 1 is deliberate: if your customers produce no such pattern, you do not have an ICP yet, you have early customers. Write the one-pager anyway and mark it provisional.

The disqualify list is not a trash can. Most flagged prospects fail on missing evidence, not evidence against them; prompt 5 names the check that settles each one.

What to do next

Take the top five fully-evidenced scores and start conversations, one at a time, using the message rules in how to get customers from LinkedIn. Before anything goes out, run it through the sales message review prompts. Re-run the whole pack after every five new customers; the profile sharpens as the data grows.

Where Baljia fits: it is an AI employee for founders and small businesses. It finds prospects and drafts LinkedIn and email outreach, and nothing is sent until you approve it or switch sending on; LinkedIn campaigns come with the Growth and Scale plans. The AI sales outreach page covers details and limits, and pricing lists the plans.

Frequently asked questions

How many customers do I need before an ICP is worth writing?

Enough that a pattern can have three supporters, so five is a reasonable floor. Below that, write the one-pager anyway but treat it as a hypothesis sheet: every section is provisional, to be tested with the next few sales rather than used as a hard filter.

Can I just ask AI to write my ICP without pasting customers?

You will get a document, but it will be generic advice about your industry, not a profile of your buyers. An ICP's whole value is that it comes from people who actually paid, and without pasted customers the AI has nothing real to extract.

What counts as evidence in a prospect row?

Something observable that connects the person to the problem: a job post, a public complaint, a tool change, a question they asked somewhere, a trait shared with your closed customers. "Big company, nice title" is a description, not evidence; if the line is empty, research first.

Why do the prompts keep repeating the honesty rules?

Because assistants drift: a rule stated once fades as the chat grows, and the model reverts to filling gaps helpfully. Restating the rules in every prompt keeps them active, at the cost of a few pasted lines.

Score your own twenty

Run the pack this week: one hour, your real customers, your real list. To see how Baljia handles this kind of work, start the 7-day free trial, no credit card, at https://baljia.ai/ and give it one focused task. LinkedIn campaigns come with Growth and Scale, not the trial.

Sources