Key Takeaways
- AI models don't read your service radius. They read text. If a town name isn't written down somewhere a crawler can reach, you don't serve that town as far as the model is concerned.
- Four signals carry a service area business: town specific pages, matching service areas on your business profile, third party citations that name the towns, and reviews that mention where the job happened.
- A town page only counts if swapping the town name would make it wrong. Thirty templated pages get discounted. Five specific ones get quoted.
- Never open a mailbox address in another town to get a second listing. It gets flagged, and the split records take months to clean up across every directory that copied them.
- Verify it yourself: fresh session, ask the customer question with the town name, screenshot the answer. Do it before work starts and monthly after.
Your Address Is Not the Boundary
A restoration company in Wisconsin covers four counties. Its trucks run forty minutes in every direction on a normal day. Ask ChatGPT who to call for water damage in the town twenty minutes north and it names three competitors, none of them closer.
That's not a ranking problem. It's a description problem. There's no text anywhere connecting that company to that town, so the model uses businesses that do have it.
Google Maps knows your service radius. The models don't read your service radius. They read words.
What the Models Actually Read
Four sources do most of the work for a service area business:
- Your own site, with the town named in the copy and not just the footer.
- Your business profile, specifically the service areas field.
- Third party citations: directories, association pages, supplier sites, local press.
- Reviews, where a customer mentions where the job happened.
Notice what's missing. No radius setting. No coverage map upload. No field you tick. All four are text a crawler can read, which means if the town doesn't appear in the text, it doesn't exist to the model.
Town Pages That Survive
This is where most of it is won and lost. The failure mode is obvious once you've seen it: someone generates thirty pages, each the same four paragraphs with a different town dropped in. Search engines have been discounting those since before AI search existed, and the models are worse for you, because when the answer gets generated there's nothing quotable. The page says the same thing about one town as the next.
A page that works contains things that would be wrong on any other page:
- Response time from your base to that town, stated honestly.
- The neighbourhoods or ZIP codes you actually reach, not the whole county.
- A job you did there. What happened, roughly what it cost, how long it took.
- Something local that changes the work. Older housing stock. A permit office that takes a week. A flood plain.
Simple test: swap the town name. If the page still reads correctly, it isn't a town page. Delete it and write five real ones instead. This is also the point where most owners find out whether doing this themselves with ChatGPT is realistic, because the model can format a page but it can't know your response time.
Make the Profile Match the Pages
Your business profile has a service areas field. Fill it with the same towns you built pages for, and only those.
Two things people get wrong. First, they list forty towns to cast a wide net, then build pages for two. The mismatch reads as noise. Second, they leave the field empty because the address is set, assuming the radius covers it. It doesn't, not for AI.
If you serve customers at their location rather than yours, hide the street address on the profile and let the service areas carry it. That's the supported setting for a service area business, and it stops the model anchoring you to one point on a map.
Citations That Name the Towns
Your own site saying you serve a town is a claim. A third party saying it is evidence, and the models weight the two differently.
The list depends on the trade. Restoration and trades: contractor directories, insurance and supplier partner pages, the local chamber, the trade association member listing. Cleaning: the same directories plus any local business network. Whatever the vertical, find the ten to twenty places that already list businesses like yours, and make sure your entry names the towns rather than just the head office.
Consistency beats volume here. Same business name, same phone number, same spelling of every town, everywhere. If an agency tells you they handle this, it's one of the five things worth checking before you believe them.
The Second Listing Trick
Someone will suggest a mailbox address in the next town so you can open a second listing. Don't.
It works for a few weeks and then it doesn't. The listing gets flagged, often by a competitor who noticed. You end up with a suspension on the fake one and, more expensively, name address phone data that disagrees with itself across every directory that scraped it. Now the models see two versions of you with different addresses and trust neither.
Cleaning that up afterwards takes months and touches every directory that copied the bad record. The honest version, town pages plus matching service areas plus real citations, costs about the same effort and doesn't blow up later.
Which Path Is Yours
| Approach | Cost | What AI does with it |
|---|---|---|
| Five to ten real town pages | 6 to 10 hours, or $150 to $400 per page | Quotable text per town |
| One page listing thirty towns | 1 hour | Weak association, nothing to quote |
| Bulk generated doorway pages | A weekend | Discounted, rarely cited |
| Second address in another town | $30 a month plus the risk | Suspension and split records |
Build the pages yourself if you have five to ten towns, someone in the business who can describe a real job in each of them, and about ten hours spread over a month. That's the cheapest version that actually works.
Fix the profile and citations first if your service areas field is empty or wrong, or your phone number appears three different ways across directories. Two hours of work, and it caps everything else until it's done. Do it before you write a single page.
Hire it out if you have more than fifteen towns, or you've tried and the pages read like templates because writing isn't your thing. Budget $800 to $3,000 a month for pages, profile work, citations and the monitoring that tells you it worked.
Do none of it yet if you can't name which towns send you profitable jobs. Pull last year's invoices, sort by town, start with the top five. Building coverage in towns you don't want work from is a way to stay busy without getting anywhere. It also decides whether the leads arrive as real jobs or price shoppers.
Verify It Yourself in Ten Minutes
Every claim above is checkable, including the ones we make. Open a private window so there's no history in the session. Ask the question a customer would ask, with the town name in it. Run it on ChatGPT, Perplexity, Google AI Overviews and Gemini. Screenshot the answers. That's your baseline.
Do it for every town before anyone starts work, then monthly after. You're watching two things: whether you appear at all, and whether what the model says about you is accurate. The second one catches problems the first one hides.
Want the baseline run for your towns?
Name the towns and we'll run the queries live on the call. You keep the screenshots either way. Book a call.
If you'd rather do the work yourself, the outline above is the whole method. When writing ten distinct town pages a month becomes the bottleneck, that's when hiring it out starts to make sense.
Frequently Asked Questions
How does AI decide which towns a service business covers?
It reads whatever is written down. The models pull from your website, your business profile, and third party sources that mention you. If your site only names your home city, and your profile lists a radius nobody can see, there's nothing to work with for the town twenty minutes away. Businesses that get named in surrounding towns have those town names in page copy, in profile service areas, and in citations a crawler can reach. It isn't a radius setting you tick. It's text in enough places that the model treats the connection as a fact instead of a guess.
Do I need a physical address in every town I serve?
No, and you shouldn't fake one. AI search doesn't require an address per town. It requires evidence you work there: town specific page content, service areas on your profile that match those pages, and outside mentions naming those towns. Businesses that open mailbox addresses to chase coverage end up with duplicate listings, suspended profiles, and name address phone data that disagrees with itself across directories. That inconsistency does more damage than the missing address ever did, because the models start seeing two versions of you and trust neither one. Serve the town honestly and say so clearly.
What makes a town page different from a doorway page?
A doorway page is the same paragraph with the town name swapped. Search engines have discounted those for a decade, and the models skip them because there's nothing distinct to quote. A real town page answers something only a local would ask. Response time from your base. Which neighbourhoods or ZIP codes you actually reach. A job you did there, described in specifics. A local quirk that changes how the work goes. If you can't write two hundred words about a town that would be wrong for the next town over, don't publish the page.
How many town pages should I build?
Start with the towns you'd actually drive to today, which is usually five to twelve. Build them one at a time and publish only when the copy is genuinely town specific. A restoration company covering four counties doesn't need eighty pages. It needs the ten towns that send real jobs, written properly, then a second batch once those hold. The mistake is bulk generating forty pages in a weekend and watching none of them get cited because they read as templates. Volume isn't the signal. Distinctness is. Ten pages a model can quote beat forty it can't tell apart.
How much does this work cost?
DIY costs your time. Roughly six to ten hours to write five town pages properly, plus two hours fixing profile service areas and directory listings. Software cost is zero. If you hire it out, town page copy runs about $150 to $400 per page from a freelancer who researches the area, and $80 to $250 per page from a content mill that'll hand you doorway pages. Full service AI search retainers covering the pages, the profile work, the citation network and the monitoring run roughly $800 to $3,000 per month, depending on how many towns you want and how contested they are.
How long before AI names me in the next town over?
Expect four to twelve weeks for the first movement, and it varies with how contested the town is. The page has to be crawled, the profile change has to propagate to the directories that feed the models, and the citations have to get picked up. A town where three competitors already have real local content takes longer than one nobody has bothered with. Run baseline queries before you publish anything, then rerun the same queries every two weeks. If nothing has shifted by week twelve on an uncontested town, the problem is the page content, not the timeline.
Can I use one page listing all my towns instead?
A single service area page listing thirty town names is better than nothing and worse than real pages. Models can read the list, so you get a weak association. What you don't get is anything worth quoting when somebody asks about one specific town. To generate an answer, the model needs a sentence about that town. A list of names doesn't give it one. Keep the overview page for humans scanning your coverage, then build individual pages for the towns that matter commercially. The two do different jobs and you want both.
How do I check whether this is actually working?
Open a fresh session with no logged in history and ask the way a customer would. Water damage repair in that town, who to call. Run it on ChatGPT, Perplexity, Google AI Overviews and Gemini, then screenshot whatever comes back. Do this for every town on your list before any work starts, so you have a baseline you can point at. Then repeat monthly. You're watching two things: whether you get named at all, and whether the description matches what you actually do. If an agency won't run this live on a call, they can't show you the result either.
About the author
Matthew Johnson is the founder of Pleiades Consultancy. He previously scaled his own marketing agency to multiple six figures before serving as CMO of an Amazon agency, where the client base tripled from 15 to 45 active clients during his tenure. He worked with some of the largest names in e-commerce, including Ridge Wallet, HexClad, BK Beauty, The Woobles, Walkize, Lonely Planet, and Obvi. He now works with local businesses to maximize their client acquisition and visibility through AI search with ChatGPT, Claude, Gemini, Perplexity, and Bing Copilot.