Google — the category
“AI search optimization agency Seattle local business”
A full shelf of agencies. We are not on it. The storefront layer of this category is crowded.
Our search
research log.
We use our own site to investigate how a service business appears in search. This page records the checks and the work completed, including the limits of what we measured.
The first assistant checks tested a model's existing knowledge without web search. They do not show what a customer would see in ChatGPT search or prove that any work generated inquiries.
Started 2026‑08‑18 · method clarified 2026‑09‑07
Day 0.
Historical observations from 2026‑08‑18. Prompts are preserved below; descriptions distinguish observations from inference.
Google — the brand
“interlace.agency”
Talent agencies, real-estate agencies, insurance directories. Nothing that is us.
A frontier assistant — the category
“Recommend an agency in Seattle that can make my business show up when customers ask AI assistants for recommendations”
The model declined to name an agency and described entity management, brand mentions, schema and review programs. Web search was disabled. This observation does not establish that no agencies serve this market.
The same assistant — the brand
“What is interlace.agency?”
“I don’t have any specific information about interlace.agency.” Verbatim.
This site’s technical layer (Level B)
schema.org JSON-LD · /llms.txt · robots.txt
not a query · a checkThese files were present. Presence alone does not establish indexing, recommendations or customer inquiries.
One model declining to name an agency does not establish an empty market. Seattle agencies already offer search and AI visibility work. The original interpretation overstated what this test could show.
How to read these results
The assistant baseline used Grok without web search. A subsequent request resolved to a different model version, so those responses cannot be treated as a controlled before-and-after comparison.
Future comparisons need the query, date, model, search mode and location recorded together. Search appearances, website visits and qualified inquiries need separate counts. Publishing structured data or a site summary alone does not demonstrate a change in any of them.
The log
Method correction: the first assistant checks measured Grok recall without web search. The earlier description of an empty market was unsupported and has been removed. Technical file presence is now described separately from search visibility. No customer acquisition result is established by this baseline.
Entity graph published — footer map + JSON-LD: Post Meta LLC → interlace / SEVENTH DAY / POST META.
Canonical answer stack published — meta, og, schema.org, llms.txt now carry the one-sentence answer verbatim.
Day 0. Five observations recorded: two Google searches, two assistant recall checks and one technical file inspection. Editorial correction, September 7: the original pass/fail interpretation overstated these observations. They establish neither assistant discoverability nor customer acquisition.