Repositioning a website without losing its search history

The problem

In September 2026 wildrunai.com changed what it sells: from an AI receptionist product to a local search studio. The site carried dozens of indexed pages, comparison pages and blog posts about the old offer. Deleting them would have broken inbound links and thrown away the relevance Google already credited them with; keeping them would have told every search engine and AI assistant the wrong story about the business.

What was done

• Redirect map by topic. 48 exact and 9 prefix rules in one file, each retired URL sent in a single hop to its closest live page or matching article, query strings kept. Withdrawn pages answer 410 so they leave the index faster. Tests fail the build if a live page links to a retired one.
• Content prune. 3 posts removed (410), 61 off-topic posts set to noindex, follow and taken out of every listing, reversible per post. The blog index went to 27 guides.
• One canonical host. www now 301s to the apex for GET and HEAD only, so form posts still aimed at www keep their data.
• Sitemap fixes. Nine malformed dates found in the live sitemap; serialization fixed and covered by regression tests. The sitemap and llms.txt are generated from the same list of live pages.
• Structured data. One Organization + ProfessionalService entity with a stable @id, city-level address, the verified Business Profile in hasMap and sameAs; Service, BreadcrumbList and FAQPage only where the content is visible on the page.
• Readable by AI systems. AI crawlers allowed, an llms.txt index, a Markdown version of every page, and a read-only MCP server listed in the official MCP registry so AI agents can query the studio’s services and articles.
• Speed and accessibility. A design audit at phone and desktop widths fixed small type, contrast and heavy images: 127 blog images given WebP variants. Lighthouse mobile on the homepage went from 88 to 99 performance and LCP from 3.7 s to 1.8 s (7 October 2026), with 100 for accessibility, best practices and SEO. An earlier pass ran 23 automated accessibility scans with zero violations.
• Indexing hygiene. Sitemap resubmitted in Search Console and IndexNow pinged after each release.

Guardrails that keep it fixed

Search regressions usually come back with the next release. The site ships through a deploy script that runs the type checker and a test suite of about 3,000 tests first. Those tests assert one H1 per page, a canonical URL, a unique title and description, the expected schema types, that every live page is in the sitemap exactly once, and that no live page links to a redirect or a removed URL.

What it does not prove yet

A clean technical foundation is the start, not the result. Competitive Bend searches take time, and the studio does not publish traffic or ranking claims it cannot show. This page reports work done and how it is guarded, nothing more.

← All work