Work
Case studies.
Two client engagements and two products of the company’s own. Each entry says what was there at the start, what was built, and what can be measured now.
Client · since July 2024, continuous from May 2026
Intertwine House Plans
A residential designer in Charleston sells her house plans through her own Shopify store and through the national plan marketplaces that carry them. At the start, the store drew little search traffic of its own, its tracking could not say where a sale came from, and its product data carried few of the specifications search engines and AI assistants read.
What was built
- Plan names, titles, descriptions and collection pages reworked, a new theme launched in June 2026, and product structured data extended for house plans, which have no type of their own in schema.org.
- A measurement method for AI answer engines: question panels across ChatGPT, Perplexity, Gemini and Claude, read against dated versions of the site.
- A Pinterest feed built in Liquid from the store’s own data, replacing a vendor feed that advertised a higher price tier than the entry price on most rows. All 66 plans accepted.
- Tracking repaired: a conversion pixel that had never fired, a tag migration that double-counted the funnel, and the tagging that now separates paid from organic Google at the order.
- Google Ads rebuilt on the store’s own copy and imagery, with automatically generated assets switched off and every change logged with how to roll it back.
- A $75 study set offered as the first purchase and credited against the plan, and a save-for-later list whose sign-ups carry the plan the visitor was viewing.
- A study-set pipeline that turns 36×24 construction sets into 11×17 sets and checks its own crops at build time, and Karpo, the revenue ledger read from six years of email.
Results
| Visits | 3.5× store sessions in Shopify, June to September 2026 against the same months of 2025. Pinterest supplies more than half |
| AI search | Monthly sessions referred by ChatGPT rose from 30 to 172, July to September 2026 |
| Search | Google clicks roughly doubled from late May to August, mostly on searches for the store and its plans by name. Broad category searches are the next piece of work |
| Speed | Largest Contentful Paint of 1.5 s on phones in Chrome field data, Core Web Vitals passing |
| Site health | Semrush reads 97 with zero errors in every scan since August 2026 |
Client · since August 2026
Bethel Lutheran Church, Roseville
A nineteen-page church website on a legacy platform, ahead of a planned rebuild. Before any change, domain ownership was confirmed separately from the hosting account and a backup was put on file. A site-wide fault had search engines treating the www and non-www version of every page as two pages, and it was fixed. Every title and description was rewritten from the page’s own text, with lengths checked by script. The rebuild is scoped and has not started.
Own product · launched October 2, 2026
Atlas of Plant Based
A plant-based travel and food publication built from the founder’s 2021 Pacific Crest Trail thru-hike: 129 days in 26 legs, mapped from his own GPS log, receipts and photographs. It also serves as a live test of the company’s publishing and search methods, started from no existing audience.
- A prerendered Astro site on Cloudflare Pages, with a sitemap, llms.txt, structured data and a robots policy open to search and AI crawlers.
- A water atlas built from 56,265 hiker reports across 313 sources, 2013 to 2026, and a rated library of 2,411 trail photographs.
- A production desk across three models: ChatGPT drafts from a voice guide built from the founder’s own writing, Claude checks facts against primary records and builds the page, and Codex reviews code, photos and copy.
- An answer-engine baseline taken in the first week, so later citations in ChatGPT and Perplexity can be measured against it.
Own product · alpha
Atlas Ascent
A lifting app that runs offline in the phone browser. It builds each session from the equipment in the room, can read that equipment from a photo with Claude, and sets the next session’s weight from an easy, medium or hard rating after each lift. Programs include a six-month strength plan and a hiking program that counts back from a trip date.
- Fifteen builds so far, with outside reviews by other models. A review of build 12 found eleven problems, and build 13 fixed all eleven.
- Strong and Hevy workout import and a consent screen before any Claude request. An App Store release is planned.
- An automated test suite runs in a headless browser.