What you'll get
- 1 copy-ready blocks
- 4 min read
- Criteria, not company names
How to use it
never ask an AI assistant for a list of companies; ask it to attack your criteria. The model cannot verify that a company exists, trades or matches your filters today, but it is genuinely good at finding the holes in how you have defined your market. Fix the criteria, then fill the list from sources you can audit.
The hack: never ask an AI assistant for a list of companies; ask it to attack your criteria. The model cannot verify that a company exists, trades or matches your filters today, but it is genuinely good at finding the holes in how you have defined your market. Fix the criteria, then fill the list from sources you can audit.
The prompt
Criteria, not company names
You are helping me define criteria for a B2B target account list.
Do not suggest any company names at any point.
What we sell: [one sentence]
Our three best customers and why they bought: [one line each]
Draft criteria: industry [x], size [x], geography [x], signal [x]
Do four things:
1. Attack the criteria. Which are really proxies for something else?
What do my three best customers share that the criteria miss?
2. Propose knock-outs: cheap-to-check facts that disqualify a company
even when it matches everything else.
3. Propose observable signals of the problem we solve: things visible
from outside, like hires, tech, filings or announcements.
4. List the kinds of sources where companies matching these criteria
can be found and verified by a human.
Be specific and disagreeable. If a criterion is too loose to act on,
say so and tighten it.What comes back (illustrative)
The useful output is usually the second and third parts. "Industry: logistics" becomes "operates more than [n] sites, because your best customers all had multi-site scheduling pain, which single-depot firms do not". Knock-outs surface things a scorecard forgets: an incumbent's logo in the footer, a parent company that centralises IT, a size band where the budget owner does not exist. Signals turn a static filter into a timing filter: job adverts for the role your product replaces the pain of, a named system in careers pages, an expansion announcement.
What to do with it
- Rewrite the criteria in the Target Account List Builder, which turns them into a 25-row list framework with tier quotas.
- Fill the rows from auditable sources: your CRM's closed-won lookalikes, directories, registries, event lists, your own site's visitors. A human verifies every name.
- Score what survives with the target account selection scorecard, and cap the list at the hours you have.
Check before you trust it
- No company name from the AI made it onto the list. If one slipped in, it gets the same verification as any other source, which usually removes it.
- Every knock-out is checkable in under a minute from public information.
- Every signal is observable from outside the company, not a guess about their internals.
- The tightened criteria still fit your three best customers. Criteria that exclude your own evidence are clever, not true.
Three mistakes to avoid
- Asking for names anyway. Models produce plausible companies that are wrong in every way that costs money: defunct, mismatched, or real but failing every filter. The ban in the prompt is the hack.
- Accepting criteria-shaped flattery. If the output just restates your draft with synonyms, re-run it with "be specific and disagreeable" doing more work: name a criterion and ask what is wrong with it.
- Skipping the human verification because the criteria are now good. Good criteria speed up checking; they never replace it. The list is only as sellable-to as its worst row.
Make it a habit: re-run the prompt each quarter with your newest three wins in the evidence line. Criteria drift as the product and market do, and the list should drift with them.
Why this works on named accounts
Common questions
Can AI build a target account list for me?
It can sharpen the criteria, find knock-outs and name the sources to search. It cannot verify that a company exists or matches your filters today, so the names must come from sources a person can check.
Why ban company names in the prompt?
Models produce plausible companies that are wrong in ways that cost money: closed, mismatched or real but failing every filter. Banning names keeps the model on the part it does well.
How often should I rerun it?
Each quarter, with your newest three wins in the evidence line. Criteria drift as the product and the market do.
Read next
Related articleTarget Account List Template: How Many Accounts and How to Pick ThemRead →More from the Academy
Target Account List Template
For founders and sales heads who have chosen their accounts and need one place to work them.
You leave with a 25-row sheet, a rule for how many accounts to keep and a weekly review in five questions.
Target Account Selection Scorecard
For founders and sales leaders who have an ICP on a slide and need a ranked list of named accounts.
You leave with a one-page ICP, an 11-point weighted scorecard and three tiers.
Target Account List Builder
Turn your criteria into a 25-row target account list framework you fill in. Uses only what you enter. Runs in your browser.
Want this in your inbox?
Optional. We use your email only to send you this, nothing else.