A Rhode Island hauler was earning search impressions on ten pages. Ten weeks after the rebuild, he was earning them on seventy-three.

Complete rebuild on Next.js + React, Tailwind, Vercel. 73 pages, written page by page rather than templated.

The obvious move on a local service site is templated city pages — same paragraph, swap the town name. We refused it. Templated city content is the exact pattern Google devalues and AI retrieval systems skip.

• 28 city pages, each naming real neighborhoods, landmarks, and road names — 5 to 12 genuine neighborhoods per page, not a radius list
• 30 sub-service pages, each with a unique intro, items-handled list, and 4–5 FAQs
• Rural towns given full pages on purpose — Little Compton, Rehoboth, Dighton, Berkley, Freetown. Low competition means a real page is a cheap number one
• Full internal linking web with breadcrumbs and descriptive anchor text

The AI-search layer

LocalBusiness JSON-LD on the home page and all 28 city pages with full NAP, hours, geo, social profiles, and aggregate rating. Service schema on service pages, FAQPage on the FAQ page, Review schema on highlighted reviews. Canonical URLs, OG and Twitter Card meta sitewide, one H1 per page, descriptive alt text and filenames carrying service plus city.

The conversion layer

A sticky mobile Call / Text / Quote bar on every page. A trust bar under every hero — Same-Day, Licensed & Insured, Family-Owned, 10% Military and First Responder. Full pricing published rather than hidden behind a form — a full rate card, 11 line items. Navigation collapsed from 30+ items to 6, with /services as the master directory.

On navigation

The client’s instinct was to cut the sub-service pages because the menu was overwhelming. That would have thrown away the entire asset. Thirty dedicated service pages are the citation surface — when someone asks an AI assistant who removes pianos in Newport County, the model needs a specific page with location signals, pricing context, and FAQ schema to cite. Nav bloat is a menu problem with a menu fix. Keep the pages, fix the menu.

Search footprint, year over year

Traffic, same window

July, head to head

Terms the old site could not reach

Head terms the old site already held, tightened

Did not rank at all in 2025

New pages, current position

Indexation

The sitemap contains 73 URLs — 30 service, 28 city, 15 core. 72 earned impressions in the first 68 days. All 58 service and city pages earned impressions — 13 in the top 10, 40 in the top 20, median position 15.3.

Source

All search figures from Google Search Console for clarksjunkremoval.com, exported August 9, 2026. Lighthouse from a mobile audit run August 9, 2026, Lighthouse 13.2.0.

Launch date

The new site replaced the Wix build at the end of May 2026. The first city page impression is dated June 7, 2026, with zero impressions on any /service-areas/ URL in the preceding 16 months — that is the date the new architecture entered Google’s index.

Why year-over-year

Junk removal demand is seasonal. Comparing June–August 2026 to the preceding spring would overstate the result. Comparing it to the identical 2025 window does not. Every figure here comes from two exports filtered to identical 68-day calendar windows. Nothing is blended across the launch date.

What the old site already did well

Clark’s ranked position 2.3 for “junk removal portsmouth ri” and inside the top ten for Newport, Middletown, and Tiverton before the rebuild. Those gains are reported as tightening, not transformation. The transformation is in footprint.

On average position

Against the immediately preceding 68 days, average position reads worse (13.2 → 17.6). That is expansion, not decline — impressions rose from 4,648 to 14,856 as 58 new pages entered the index and began ranking for queries the old site never appeared for. A site ranking for far more queries shows a higher average position number. The 30.9 shown in Search Console’s 16-month header is the blended lifetime average — dominated by the old site — and sits between the 2025 window (36.0) and the 2026 window (17.6).

Query counts

Non-branded top-ten and top-twenty counts are raw Search Console counts and include some queries outside Clark’s service area. Both years are counted identically, so the comparison is valid, but 319 should not be read as 319 commercially relevant rankings. Individually named queries are all in-market.

Impression sampling

Search Console anonymizes rare queries, so query-level impression totals don’t sum to the performance chart totals, and the gap is larger in 2026 because the query tail is longer. All click and impression totals come from the daily performance chart, which is not sampled.

Regressions

“junk removal jamestown ri” ranked 7.7 in the 2025 window and does not appear in the 2026 top 884. The Jamestown city page exists and ranks, but lower. A cluster of long-form variants such as “junk removal newport rhode island” also dropped out while their shorter equivalents entered the top ten, at zero clicks either way — Google consolidating variants rather than lost visibility.

What is not claimed

This case study reports search performance only. Lead volume, cost per lead, booked jobs, and revenue are not reported because lead attribution was not instrumented during this window. When it is, this gets a V1.1 with those numbers or it doesn’t get them at all.