service-area-pages
Writes town and service-area pages for a local business that each say something true and specific about that town, folds the towns that can't into one honest areas-we-serve page, and checks the set for duplicates before anything ships. Use when a plumber, roofer, HVAC company or other service-area business needs city pages, wants to fix town pages that are all the same, or has a local SEO plan that calls for them.
How do I install this agent skill?
npx skills add https://github.com/boringmarketer/boring-marketing-skills --skill service-area-pagesIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill is a professional SEO tool for creating service area pages. It features automated research via web scraping and content validation through a custom Python script to avoid duplicate content penalties. Analysis reveals no security issues, and the skill implements quality gates that align with its stated purpose.
- Socketpass
No alerts
- Snykfail
Risk: CRITICAL · 2 issues
What does this agent skill do?
Service Area Pages
A town page earns its place only when it says something true about the business in that town that the other town pages can't. Google's spam policies name the alternative directly: "multiple domain names or pages targeted at specific regions or cities that funnel users to one page" is doorway abuse. This skill decides which towns qualify, writes those pages, folds the rest into one honest areas-we-serve page, and proves the set isn't duplicates before anyone publishes it. The method is a floor. The no-invented-facts rule and the duplicate check don't flex.
Context
Only if ./brand/, ./brands/ or MARKETING.md exists (nothing to load otherwise), run:
python3 ${CLAUDE_SKILL_DIR}/../start-here/scripts/brand_context.py --for service-area-pages
(${CLAUDE_SKILL_DIR} is this skill's folder; outside Claude Code use the folder this file was read from, e.g. .agents/skills/service-area-pages. If the script or python3 is unavailable, read the brand folder directly: ./brands/<slug>/ when ./brands/active names one, else ./brand/; then MARKETING.md, preferences.md and the files in reads:. If none exist, proceed standalone and ask only for what the task needs.)
If ./campaigns/local-seo/plan.md exists, read it first. Its market map already says which towns get a page, which get a line, and why. Don't redo that work. Use its capability ledger as the source of every claim. The ledger is input only: its rows, IDs and tiers never appear in the pages or in the handoff. Claim sources go to ./campaigns/service-area-pages/notes.md.
Modes
| Situation | Mode |
|---|---|
| No town pages yet | NEW SET, all phases |
| Town pages exist and look alike, or slid in rankings | FIX EXISTING, start with the duplicate check in Phase 6 on the live pages, then Phases 1-7 |
| One named town | SINGLE PAGE, Quick Mode: qualify that town, write it, run the duplicate check against the existing pages |
Phase 1. Qualify each town
A town gets its own page only when all three hold:
- The business really serves it. Job addresses, the Google profile's service areas and an honest drive time from the base. Google's own limit for a profile's service area is about 2 hours of driving from the base, and a page for a town the crew won't drive to is a doorway page by definition.
- There is something to win. Demand evidence for the service in that town and no page of the business winning it (from the plan, or a quick check: the town and state in the query for organic results, and "[service] near me" checked from the town's center for the map pack, since that's how most local buyers search).
- There is real local proof. At least one element from the list below. Two is a good default for a full page.
Local proof that counts:
- jobs done in the town, by type, from the job records;
- reviews that name the town or a neighborhood in it, quoted verbatim with the date;
- cleared job photos from the town;
- a crew member based there, or an honest response time from the base;
- a local condition that changes the work, from a source you fetched: the city's water quality report, its permit rules for the job, the age of the housing stock, a past freeze or flood, an HOA or utility rule, a utility rebate;
- a pricing difference (a trip fee zone, for example).
Gather the sourced conditions with one research subagent per town, in parallel (Task tool, general-purpose when the plugin's research type is unavailable), each fetching the city's own pages and returning facts with URLs, within the research budget (40 fetches, 8 minutes).
We use Firecrawl for these reads: it returns clean markdown of the city's own pages and can crawl a city site for the permit rule or the water report (free tier of 1,000 credits a month, no card). Without it, WebFetch reads the same pages as summaries, and each fact still carries its URL.
What doesn't count: "proudly serving", population, founding year, landmarks, a list of neighborhoods nobody asked about, and anything you can't source. Census trivia that doesn't change the job is padding.
A town that fails gets a line on the areas-we-serve page, listed in the output with the reason and what would unlock a page ("3 more jobs and one review naming Rowlett").
Phase 2. Plan the set
- One areas-we-serve hub. It lists every served town with one true line each and links to the town pages that exist. Unqualified towns live here.
- Town pages only for the towns that qualified.
- Service pages stay the parents. A service × town page (water heater repair in Garland) needs the plan's proof threshold for that exact combination. Never multiply services × towns because a template makes it easy.
- One page per query. The homepage should not target "[service] [town]" for a town that has its own page. Two pages chasing one query split the signal.
- Links. Town page to the services done there and to the hub. Service pages to the towns where the jobs actually happened.
Phase 3. Spec each page before writing it
Write a short spec per page in ./campaigns/service-area-pages/notes.md. The handoff carries only its URL, create-or-improve and first-question rows; the rest is working data:
| Field | Example (sample business) |
|---|---|
| URL | /areas/garland-tx/ (stable slug, no dates) |
| Create or improve | Create |
| The question it answers first | "Who can fix a leaking water heater in Garland today?" |
| Local proof | 31 water heater jobs in 12 months; 4 reviews naming Garland; attic installs in 1970s ranch homes; city permit required |
| Different from its siblings because | Older housing stock, attic gas units, the permit step, its own reviews |
| Who wins now | The three businesses observed in the Garland map pack, by name |
| CTA | Call or text, with hours |
The "different from its siblings" line is the doorway test written down. If you can't fill it with something true, the town goes back to the hub.
Phase 4. Write the page
Load references/page-anatomy.md before writing. The page copy is what the visitor reads: no ledger rows, source tags, claim IDs or confirm markers inside it. Default section order:
- Answer first. 40-80 words that answer the page's question: what, where, how fast, how to reach them, with one real fact about this town.
- The work done here. The job types from the ledger for this town, each with the substance of the work: what fails, how it's diagnosed, what the fix involves, what it costs the homeowner in time or damage if it waits. This is true in every town and specific to the trade, which makes it the safest thing to write at length.
- What's different about this town. Only sourced conditions that change the work.
- Proof. Reviews that name the town, a job story, real photos.
- Coverage and response. Honest drive times, hours, emergency availability exactly as the owner confirms it.
- Pricing. The structure (trip fee, minimum, free estimates or not). Dollar amounts only if the owner publishes them.
- Questions. Only real ones, from calls, reviews or the questions Google shows for that town's searches. Skip the section if there are none.
- Call to action. Tap to call, text if they answer texts, hours.
The filler trap. Tell a writer, human or AI, "never invent a fact" and it fills the space with procedure: what to have ready when you call, what happens after you book, questions like "What areas do you serve?". It reads as safe and says nothing. When a service business rebuilt its copied town pages and retired the rest with redirects, both drafting rounds drifted into this filler. What fixed it was pointing the writer at the substance of the work itself. If a paragraph would make sense on a page for a different trade, cut it. Enforce it mechanically across the set: at most one "when you call" block per page, two sentences or fewer; no questions about the call or the booking itself; review excerpts verbatim, 40 words or fewer, each tagged with its town and month.
Phase 5. Technical checks
Load references/technical-checks.md, which carries the commands that settle each check. In brief:
- One business entity in the markup, with the real address or none at all for a business that hides its address. Town pages say where the business works (
areaServed). Never a separate LocalBusiness per town with an invented address. - No star ratings in your own markup, and no prices, hours or claims in markup that the visible page doesn't show.
- Title and H1 name the service and the town once, plainly.
- Images sized for phones. A hero image of several megabytes on a page meant for someone with a burst pipe is a lost call.
- Every retired or renamed URL redirects in one hop to a page that loads.
Phase 6. Quality gate (before anything ships)
Run it silently. Results (overlap numbers, read-test notes, claim trace, filler count) go to ./campaigns/service-area-pages/notes.md; the handoff shows only the passing set.
- Duplicate check. Run it across the new pages and every existing town page:
python3 ${CLAUDE_SKILL_DIR}/scripts/overlap.py <pages-dir-or-files> --names "Dallas,Garland,Mesquite,Richardson"It masks the town names, then reports the share of sentences any two pages have in common. Default bar: under 30% for every pair. Over the bar, fix the template or the proof behind it, never one page by hand, because the same defect sits on every sibling. If proof can't close the gap, the town goes back to the hub. - The read test. Pick two town pages at random. Would someone who read the first learn something on the second? If not, they are one page.
- Claim trace. Every factual claim maps to the ledger, a fetched source with its URL, or the owner-confirm list. Anything else comes out. Prices, guarantees and response times appear only as the owner confirmed them.
- Filler count. Search the set for the procedural phrases the filler trap produces ("have your address ready", "what to expect", "next step", "when you call"). More than one short call block on any page fails the set.
- Markup matches the page. Nothing in the JSON-LD that a visitor can't see.
Rewrite and re-check up to three rounds. If the set still fails, ship fewer pages and say so in one line in the note. Eight real pages beat twenty where twelve are padding.
Phase 7. Publish, retire and measure
- Publish the hub and the 3-5 strongest town pages first. Add more when those are indexed and showing impressions.
- FIX EXISTING: for each old town page, rebuild it (it qualifies), merge it into the hub with a 301, or redirect it to the nearest real town page. Decide by what each page earns in Search Console: a copied page that still earns clicks gets rebuilt, not deleted; one that earns nothing redirects, with both the slash and no-slash versions of its URL covered. One hop, then update internal links and the sitemap. Town pages that are literal copies of each other tend to lose ground first when rankings drop, so fix the copies before a drop, not after.
- Record a baseline per page in Search Console (with
GOOGLE_APPLICATION_CREDENTIALSandGSC_SITE_URLset, read it directly; otherwise ask for a Performance export). After 60-90 days indexed, a page with no impressions gets more proof, a merge into the hub, or a redirect. Leaving it costs the pages that work.
Output
Write ./campaigns/service-area-pages/<YYYY-MM-DD>-pages.md, the handoff for the owner or the developer. Length: page budget, about 800 to 1,500 words of copy per page; the handoff around the pages stays short.
## SERVICE AREA PAGES for <business>, then one line with the date and what the set is for.- The qualify table: every served town, qualified or not, the reason in a few words, and the unlock for those that aren't.
- The set: one row per page (URL, create or improve, the question it answers first), the hub, the service-page links, and the redirect list for FIX EXISTING.
- Per page: the full copy and its JSON-LD block.
- Owner to confirm: every claim that needs a yes before publishing, with the page it sits on.
- Measurement baseline: Search Console for each page's query group, and the near-me map pack position from each town's center, recorded with the date and device so next month's check is comparable.
- Sources and assumptions, 10 lines at most: what the set rests on (the plan, job records, pages and reports read, with dates), and what is missing. When proof is thin, at most 3 of those lines say what was checked, what could not be, and the one gap that would unlock the next town ("three more jobs and one review naming Rowlett") with one offer to gather it. Missing inputs (no plan, no job records, no Search Console export) are one line here, never an audit.
./campaigns/service-area-pages/notes.md holds the working data: per-page specs, the ledger rows each page used, source URLs, and the Phase 6 results. Save it, don't paste it, and list it once under Files saved.
If the project has a codebase, also write the pages into it, the way the project already builds pages, and list the paths. Never publish. Append one line to ./brand/assets.md.
When the user edits your work
Apply the edit, then make it stick: save a gold case to ./brand/gold/<this skill>/ and
one to three yes/no checks to ./brand/checks.md, and confirm in one line. Before handing
over any work, answer the checks the context block loaded, silently, and fix every no.
How: ${CLAUDE_SKILL_DIR}/../start-here/references/taste-loop.md.
What's next
/first-screen-checkon each new page, on a phone, before traffic goes to it (~10 min)./google-business-profileso the profile's services and links match the pages (~20 min)./reviewsto keep fresh reviews coming in. The proof on these pages ages, and Google's rules forbid scripting what customers write (~15 min)./performance-reviewat 60-90 days to read the baseline and decide what to merge (~10 min).
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/boringmarketer/boring-marketing-skills/service-area-pages">View service-area-pages on skillZs</a>