How to write an ICP a local search can answer.

What a local ICP has to contain

A local ICP has to contain a vertical a directory already uses, a metro you can point at, a public problem you can detect, and an offer you can fulfill. Those four fields are the hunt. Everything else is optional commentary.

The buyer of this page is the marketer. The dentist is not writing an ICP. You are writing who you will hunt so that a search, a spreadsheet, or Scout can answer without a brainstorm. If a sentence cannot become a query, it is a brand exercise. Keep it off the profile.

B2B ICPs talk about titles, headcount, and tech stack. Local ICPs talk about listings. Apollo is better at the first corpus. This guide is the second. If you catch yourself writing VP of Operations at multi-location groups, you have slipped into a people graph. Come back to the shop on the corner and the leak on its profile.

Write the profile in plain fields, not a paragraph. Category: general dentistry. Metro: Austin, Texas. Problem: no online booking. Offer: a booking block on a one-page site. Exclusions: DSOs, mobile-only clinics, shops you already lost. That block can live in Notion, in a sheet header, or as an ICP profile in Nerxus One. The test is the same. Can someone else run it tomorrow without calling you.

Role is a fifth field when you know it. Owner, office manager, service writer. It is useful for the first line of an email. It is not a search filter on Google Places. Do not block the hunt because you have not named a persona. You can learn who answers after the first three conversations.

How to write the vertical so a directory can match it

Write the vertical in the words the directory already prints on the pin. Dentist, HVAC contractor, Mexican restaurant, independent auto repair. Those strings match Places types and GBP categories. Boutique experiential wellness does not.

Start from a category the shop would choose for itself, then tighten only if the directory lets you. General dentistry is tighter than health. Pediatric dentistry is tighter than general dentistry if that is the only work you fulfill. If you sell the same booking DEMO to every dental office, stay on the parent category and let exclusions do the rest.

Nerxus One keeps a deeper taxonomy, on the order of hundreds of local categories, so the product already knows dentists are not the same hunt as roofers. You still pick the row. You do not invent a new row because your positioning doc used a prettier noun. If Scout cannot see the noun, the hunt is a paragraph again.

When two categories overlap, pick the one that owns the leak. A med spa that is also a dentist is a dental hunt if you are selling chair-hour booking. It is a reputation hunt if you are selling review collection. One profile, one leak. A second profile can exist next month.

Check the category against three live listings before you lock it. If those three listings use a different parent type than the word you wrote, change your word. The directory wins. Your mood board does not.

How to write the metro as a place, not a vibe

Write the metro as a city, a county, or a named metro the map already understands. Austin is a place. Greater Central Texas creative corridor is a vibe. Directories geocode places. They do not geocode brand poetry.

Pick a radius you can actually work. If you will drive for the install, use the drive. If the work is remote, you still want a city name so the listing language, the phone area, and the competitors stay coherent. A nationwide ICP is a people-graph habit. Local search decays into junk when the place field is the country.

One metro per profile. If you sell Austin and San Antonio, that is two profiles or one profile you clone. Mixing them in a single filter makes LeadScore and your own eyes lie, because a thin Austin list plus a fat San Antonio list looks like a healthy hunt. It is two hunts glued together.

Neighborhoods are legal when the directory supports them and when the volume is real. East Austin coffee is a hunt if the map returns enough shops with the leak you sell. East Austin lifestyle brands is not a neighborhood. It is a slide.

Write the exclusion for places you will not serve. Islands, towns you cannot invoice, cities where you already have a conflict. Exclusions keep you from feeling busy. They also keep Relay off a row you would refuse after the reply.

How to name the problem as a detector

Name the problem as something a fetch can see. No website, no SSL, no booking link, unclaimed listing, rating under a number you choose, review count under a number you choose, a service they perform that the listing omits. Those are detectors. Bad brand energy is not a detector.

Cheap detectors ride along at search time because they come from the same listing fetch. That is why they belong on an ICP you will run every Monday. Expensive detectors wait for a deeper pass: a full site crawl, a permit match, an answer-engine probe. You can mention those in a note. Do not require them before the first search or the ICP will never run.

Pick one primary detector for the profile. Secondary tags can rank the pile. The primary detector is the offer. No-booking dentists is an ICP. Dentists with any of twelve issues is a junk drawer. LeadScore can rank a junk drawer. It cannot tell you what to build by Friday.

Service-gap is a legal primary detector if you already know the taxonomy of that vertical. It means they do a job they forgot to list. It is a stronger page later and a harder Monday. If week one is your first week, start with no website or no booking. Graduate to service-gap when you can name the missing service without guessing.

Write the detector in the same words you will use in the email. If the ICP says digital presence gap and the email says your listing has no site, you will rewrite every letter. The ICP language should be the finding language.

How to attach an offer you can fulfill this week

Attach the offer as a deliverable, not as a category of marketing. A one-page site with their services and a book button is an offer. Local growth is not an offer. If you cannot demo the deliverable, it does not belong on the ICP.

The offer has to map to the detector. No website maps to a site DEMO. No booking maps to a booking block. A collect-page gap maps to a page with public review buttons. A missing privacy policy maps to a draft labeled professional review recommended. Mixing those maps is how you sell a site to a shop that needed a collect page and then wonder why the call felt confused.

In Nerxus One, many findings already carry a fulfillment action id so Forge opens the matching builder. That seam is the ICP in software form. Outside the product, write the builder name in the profile: Webflow, a page host you know, Echo’s collect pattern, a form tool. If the builder field is blank, the ICP is a hunt with no job.

Price is not required on the ICP. A FEE range you will actually say is useful so you do not hunt shops you will resent. Do not put a modeled DEV number on the profile. DEV is a later memo with assumptions. The ICP is the hunt, not the invoice.

If you have two offers, write two ICPs. Website plus reputation is a retainer you sell after they are a client. It is a bad first filter because the search cannot tell you who wants both. Hunt one leak. Bundle later.

What to exclude so the list stays sellable

Exclude the shops you will not take even if the detector fires. Chains, franchises with a national site, businesses you already serve, businesses a partner already serves, categories that need a license you do not want to touch, owners who told you no last quarter. Exclusions are how an ICP stays kind to your future self.

Size bands are optional and easy to fake. Employee counts on local listings are often missing or wrong. Prefer exclusions you can see: multiple locations listed, a .com that is clearly a corporate template, a Facebook page that only posts national creative. If you cannot see it, do not filter on it.

Exclude detectors you cannot fulfill. ADA findings are real and they are drafts, not a law practice. If you do not have a reviewer, keep ADA off this ICP. Medical pages that would imply HIPAA coverage you do not have should stay off the hunt. Honesty in the exclusion list is cheaper than a bad first call.

Exclude sources that are unknown. If you have no Yelp key, do not write an ICP that requires a Yelp-only signal. Unknown is a first-class state in this product. Pretending the data arrived is how a profile looks complete and a list looks empty.

Review the exclusion list monthly. A no from March is still a no in September unless they asked you back. A chain that opened a truly independent location might belong. Exclusions rot in both directions. Watchtower-style freshness applies to your rules, not only to their rating.

Why quality businesses is not a filter

Quality businesses is not a filter because no directory has that field. You cannot search it. You cannot rerun it. You cannot hand it to Scout. It is a compliment you will either earn in the meeting or invent to avoid picking a detector.

People write quality when they are afraid of ugly listings. Ugly listings are often the ones with the leak you sell. A four-star shop with no site is a week-one conversation. A polished five-star shop with a national agency is a long cycle you do not have on Monday. Your taste is allowed after the search, when you save or skip a row. It is not allowed in the query.

Replace quality with three observable checks you will run by hand if you must: they still operate, the phone rings a human, the leak is still public. Those checks are a review step, not an ICP field. LeadScore can rank cheap tags. It will not print a taste score, and you should not either.

If a partner asks for your ICP and you send a paragraph about ambitious owners who value partnership, you have sent a brochure. Send the four fields. They can decide whether they want that hunt. You can decide whether you want that partner.

The product exists so you stop retyping the brochure. ICP profiles turn vertical, metro, exclusions, and an offer into a filter Scout can run. If you never buy, the same four fields still beat the paragraph. That is the whole point of this page.

How to test the ICP with one search

Test the ICP by running one search and reading the first ten listings against the four fields. If you cannot decide save or skip in thirty seconds per row, the detector is fuzzy or the category is wrong.

Count three numbers and stop. How many listings came back. How many showed the primary detector. How many you would actually take this week. You do not need a percentage from a study. You need to know whether Monday is a working day or a rewrite.

If the list is huge and the detector is rare, keep the metro and change the detector, or keep the detector and shrink the metro. If the list is tiny and every row is a yes, you may have a good ICP in a thin market. Clone it to the next city before you invent a second offer.

Save the test as a dated note: query, source, counts, three example names, what you would change. In Nerxus One the saved leads are that note. The UI should not invent a finished list if the queue is still sitting. Your sheet should not invent one either.

Do not contact anyone during the test. The test is a read. Contact is a later module with a gate. Mixing them is how an ICP demo becomes a spam story you then blame on the filter.

When to write a second ICP

Write a second ICP when the offer changes, the metro changes, or the detector changes, and not when you are bored. Boredom is a sequence problem or a fulfillment problem. It is rarely a missing persona.

A second metro with the same vertical and detector is the cleanest clone. Copy the profile. Change the place. Run the test. That is how a solo marketer grows a book without becoming a generalist overnight.

A second vertical is a new fulfillment stack until you prove otherwise. Roofers and dentists do not share a DEMO. They may share a collect page. If the only common offer is reputation, the second ICP should be a reputation hunt, not a vague home-services-and-health hunt.

A second detector on the same vertical is allowed when the first hunt is empty or when leftover findings after close become the next product. Service-gap after no-website is a classic second profile. Do not run both as one filter on day one.

Retire a profile that you have not run in a month. Dead ICPs clutter Mission Control the same way dead boards clutter a CRM. Ten named campaigns that nobody opens are not a strategy. They are costume.

How to write exclusions for chains without guessing employee counts

Write exclusions you can see, because employee counts on local listings are often missing or wrong and a size band you guessed is a fake filter. Chains, franchises with a national site, a .com that is clearly a corporate template, a Facebook page that only posts national creative, businesses you already serve, owners who told you no last quarter. Those are exclusions. Headcount is a people-graph habit.

Exclude detectors you cannot fulfill. ADA findings are real and they are drafts, not a law practice. If you do not have a reviewer, keep ADA off this ICP. Medical pages that would imply HIPAA coverage you do not have should stay off the hunt. Honesty in the exclusion list is cheaper than a bad first call. Unknown sources stay unknown. If you have no Yelp key, do not write an ICP that requires a Yelp-only signal.

Review the exclusion list monthly. A no from March is still a no in September unless they asked you back. A chain that opened a truly independent location might belong. Exclusions rot in both directions. Watchtower-style freshness applies to your rules, not only to their rating. Write the exclusion for places you will not serve. Islands, towns you cannot invoice, cities where you already have a conflict.

Exclusions keep you from feeling busy. They also keep Relay off a row you would refuse after the reply. In Nerxus One the profile can carry them. In a sheet header they still work. If your ICP cannot become a Scout query, it is a mood board. Write it again. Quality is not an exclusion. Quality is a feeling after the meeting.

What to do when two directories disagree on the category

When two directories disagree on the category, pick the word the directory you will search already prints on the pin, then check three live listings before you lock it. Dentist, HVAC contractor, Mexican restaurant, independent auto repair. Those strings match Places types and GBP categories. Boutique experiential wellness does not. The directory wins. Your mood board does not.

Start from a category the shop would choose for itself, then tighten only if the directory lets you. General dentistry is tighter than health. Pediatric dentistry is tighter than general dentistry if that is the only work you fulfill. If you sell the same booking DEMO to every dental office, stay on the parent category and let exclusions do the rest. Nerxus One keeps a deeper taxonomy, on the order of hundreds of local categories. You still pick the row.

When two categories overlap, pick the one that owns the leak. A med spa that is also a dentist is a dental hunt if you are selling chair-hour booking. It is a reputation hunt if you are selling review collection. One profile, one leak. A second profile can exist next month. If Scout cannot see the noun, the hunt is a paragraph again.

If the search returns nothing, the ICP is too narrow, the metro is too small for that detector, or the category name is not what the directory uses. Change one variable. Do not add a lifestyle segment to feel clever. B2B ICPs talk about titles. Local ICPs talk about listings. If you catch yourself writing VP of Operations at multi-location groups, you have slipped into a people graph. Come back to the shop on the corner.

How an offer field keeps you from hunting ADA when you sell sites

An offer field keeps you from hunting ADA when you sell sites because the offer has to map to the detector and a missing policy is a draft with a reviewer, not a one-pager with a book button. A one-page site with their services and a book button is an offer. Local growth is not an offer. If you cannot demo the deliverable, it does not belong on the ICP.

No website maps to a site DEMO. No booking maps to a booking block. A collect-page gap maps to a page with public review buttons. A missing privacy policy maps to a draft labeled professional review recommended. Mixing those maps is how you sell a site to a shop that needed a collect page and then wonder why the call felt confused. In Nerxus One, many findings already carry a fulfillment action id so Forge opens the matching builder.

Outside the product, write the builder name in the profile: a page host you know, Echo's collect pattern, a form tool. If the builder field is blank, the ICP is a hunt with no job. Price is not required on the ICP. A FEE range you will actually say is useful so you do not hunt shops you will resent. Do not put a modeled DEV number on the profile. DEV is a later memo with assumptions. The ICP is the hunt, not the invoice.

If you have two offers, write two ICPs. Website plus reputation is a retainer you sell after they are a client. It is a bad first filter because the search cannot tell you who wants both. Hunt one leak. Bundle later. If Friday's DEMO took six hours, the ICP's offer is too big or your builder is too slow. Shrink the deliverable on the profile. Do not silently hope next week is faster.

How to clone an ICP to a second metro without changing the story

Clone an ICP to a second metro by copying the profile and changing only the place, because a second metro with the same vertical and detector is the cleanest growth a solo marketer can run without becoming a generalist overnight. Austin and San Antonio are two profiles or one profile you clone. Mixing them in a single filter makes LeadScore and your own eyes lie, because a thin Austin list plus a fat San Antonio list looks like a healthy hunt. It is two hunts glued together.

Write the metro as a city, a county, or a named metro the map already understands. Greater Central Texas creative corridor is a vibe. Directories geocode places. They do not geocode brand poetry. Pick a radius you can actually work. If you will drive for the install, use the drive. If the work is remote, you still want a city name so the listing language, the phone area, and the competitors stay coherent. A nationwide ICP is a people-graph habit.

Neighborhoods are legal when the directory supports them and when the volume is real. East Austin coffee is a hunt if the map returns enough shops with the leak you sell. East Austin lifestyle brands is not a neighborhood. It is a slide. Test the clone the same way you tested the first: one search, first ten listings, three counts. How many came back. How many showed the primary detector. How many you would actually take this week.

A second vertical is a new fulfillment stack until you prove otherwise. Roofers and dentists do not share a DEMO. They may share a collect page. If the only common offer is reputation, the second ICP should be a reputation hunt, not a vague home-services-and-health hunt. Retire a profile that you have not run in a month. Dead ICPs clutter Mission Control the same way dead boards clutter a CRM.

What Nova can and cannot map from a five-step builder

Nova can map categories, metros, roles, detectors, and offers that exist in the tenant catalog from a five-step builder, and it cannot select an unrelated category because a generic word sounded close. That mapping is why an ICP in this product is a filter a search can run, not a paragraph about quality businesses. If you never buy, the same four fields still beat the paragraph. That is the whole point of this page.

Nova routes. It does not scrape listings or write the email itself. Ask it to find dentists in Austin with no online booking and it routes to Scout. Ask it to email them and it routes to Relay. Supervised is the default. Mixing those jobs is how an ICP demo becomes a spam story you then blame on the filter. Do not contact anyone during the test. The test is a read.

Generic words cannot select an unrelated category. If your ICP cannot become a Scout query, it is a mood board. Write it again. Role is a fifth field when you know it. Owner, office manager, service writer. It is useful for the first line of an email. It is not a search filter on Google Places. Do not block the hunt because you have not named a persona.

When you join a waitlist or an app later, you are buying the rerun, the detectors, and the handoff of the same business id into Radar and Pulse. You are not buying a new theory of ICP. The theory is the four fields. The software is the part that refuses to let the paragraph sneak back in. If you cannot remember the four fields without opening this page, write them on a card over the monitor. Vertical. Metro. Problem. Offer.

How Watchtower uses the same ICP without becoming a new hunt

Watchtower uses the same ICP without becoming a new hunt by re-scanning the businesses you already saved and dating the diffs, so Monday is the same query plus what changed, not a new niche because you were bored. New listings, a rating that dropped, a site that appeared, a claim status that flipped. Timing is a finding. A Monday alarm plus last week's sheet is the manual version.

Do not refresh the same metro until it feels lucky. Lucky is how people burn a sender on stale rows. If the detector is gone, the row is gone. If the detector is still there and you already emailed, the row is a follow-up with a reason, or it is parked. Boredom is a sequence problem or a fulfillment problem. It is rarely a missing persona.

A second detector on the same vertical is allowed when the first hunt is empty or when leftover findings after close become the next product. Service-gap after no-website is a classic second profile. Do not run both as one filter on day one. Keep the offer field honest as you learn. If Friday's DEMO took six hours, shrink the deliverable on the profile.

Share the profile with a teammate as fields, not as a vibe channel. Two people running dentists / Austin / no booking / booking DEMO will produce comparable piles. Two people running quality Austin clinics will produce an argument. When you have a seat, Scout still never contacts anyone. Watchtower still never contacts anyone. They feed Pulse and Relay. The four fields stay the ICP.

How to reuse the ICP next week without rewriting it

Reuse the ICP by running the same query, then diffing what changed. New listings, a rating that dropped, a site that appeared, a claim status that flipped. Timing is a finding. Watchtower is the product name for that re-scan. A Monday alarm plus last week’s sheet is the manual version.

Do not refresh the same metro until it feels lucky. Lucky is how people burn a sender on stale rows. If the detector is gone, the row is gone. If the detector is still there and you already emailed, the row is a follow-up with a reason, or it is parked.

Keep the offer field honest as you learn. If Friday’s DEMO took six hours, the ICP’s offer is too big or your builder is too slow. Shrink the deliverable on the profile. Do not silently hope next week is faster.

Share the profile with a teammate as fields, not as a vibe channel. Two people running dentists / Austin / no booking / booking DEMO will produce comparable piles. Two people running quality Austin clinics will produce an argument.

When you join a waitlist or an app later, you are buying the rerun, the detectors, and the handoff of the same business id into Radar and Pulse. You are not buying a new theory of ICP. The theory is the four fields. The software is the part that refuses to let the paragraph sneak back in.

If you cannot remember the four fields without opening this page, write them on a card over the monitor. Vertical. Metro. Problem. Offer. The card is the ICP. Everything else is optional until the card works.