Summary
When work began with a residential mold-testing company in 59 US cities, three of four AI assistants had never named it and 24 markets were invisible on every tested question. By August 2026 it was named in 46% of 265 graded answers and cited as a source in 60% of answers that showed sources. In its strongest cities it appeared in over 70% of answers; nationally, 9.8%.
A buyer’s question the ad budget can’t reach
A homebuyer in Austin, three days from closing, types into ChatGPT: “Do I need a mold inspection before closing, and who does it near me?” The assistant answers in a paragraph and names two or three companies.
No ad appears. If your company isn’t one of the names, the $20,000 a month you’re spending on Google Ads never enters the conversation.
That was this company’s situation. It had 59 city pages and a decent ad budget, and when I asked four assistants about mold testing in its markets, three of them didn’t know it existed.
Measuring the invisible
The citation work got its own phase and its own dataset, because nothing could be fixed until it could be seen. I asked four assistants 30 national category questions, 120 answers. Then 165 location-specific questions across all 59 markets, 522 answers, with the negatives stored alongside the positives because “not mentioned” is a finding too.
I checked which markets triggered a Google AI Overview and who it cited. I crawled 168 competitor location pages to see what the assistants were quoting instead of us, and pulled the 33 competitor pages drawing the most traffic from the keywords we wanted.
Every answer was saved word for word. A later run had to be comparable to an earlier one, not a fresh opinion.
Becoming the answer, one city at a time
The content changes were about shape more than volume. Where an AI Overview already fired for a question, we rewrote the page through my Article Factory so the direct answer sat at the top in a block an assistant could lift whole. Each market page got FAQ sections that answered the specific questions buyers in that city ask.
The company got a public entity record so assistants could disambiguate it from every other mold company. Structured data was standardized across 276 indexed pages. Video went on the pages that book.
The delivery was as important as the content. The whole plan, produced by the growth-plan layer of my Audit Engine, shipped on July 6, 2026, as 59 execution cards, one per market, each carrying its own verified structured data, drafted FAQ additions, title rewrites and a teardown of the local competitor.
Every figure in the plan deep-linked to its row in an evidence workbook, 1,147 links in all, and a 93-check quality gate ran over the whole package before it went out. The client’s own developer installed it from a single zip.
The gap the panel found
The results by intent contained one number I could have hidden.
When buyers asked assistants about mold inspections against a real-estate deadline, the company was named in 78.3% of answers. When buyers asked assistants to compare two named providers, the question a buyer asks immediately before booking, the company was named in 0 of 23 answers.
The zero went into the report as the next quarter’s work.
What the panel showed
From the August 2026 panel, 265 graded answers across 25 buying prompts:
- Named in 46% of AI answers. Cited as a source in 60% of answers where the assistant showed sources.
- Google AI Overview citations went from 4 to 6 of 59 markets between July 6 and July 15, same method both times.
- By engine, reported separately because the spread is the point: Gemini 64.0%, Grok 56.0%, Perplexity 30.7%, ChatGPT 22.5%.
- By market: West Palm Beach 76.2%, Sacramento 72.7%, Austin 72.4%, New York 72.4%. Nationally, 9.8%.
That last pair describes the strategy. AI assistants answer local questions with local sources, so a company that’s a household name in Sacramento and a stranger nationally is exactly the shape a 59-market business takes, built one city page at a time.
What it was worth
The citation panel in my Optimization Loop is how we saw it. The CRM is how we counted it.
Over the same twelve weeks, organic leads grew to 134 a month, organic cash to $17.2K a month, and organic collected $32,538 in 90 days while the ad budget was cut 96%. The company’s decision-maker, on a recorded call, said he was happy to see us “doing a great job” with the search work, and contrasted it with a previous agency under which, in his words, traffic got bombed.
At a glance
- The niche: Residential mold testing, 59 US city markets.
- The problem: No AI visibility. Three of four assistants had never named the brand; 24 of 59 markets invisible on every tested question.
- What was built: A citation dataset of 642 verbatim answers with negatives kept; per-market AI Overview checks; 168 competitor pages crawled; answer-shaped rewrites, FAQ layers, a public entity record, structured data across 276 pages; 59 execution cards with 1,147 evidence links and a 93-check quality gate.
- Implementation: Its own phase and dataset; the plan shipped July 6, 2026, as a single installable package; the Optimization Loop’s panel re-run on fixed prompts.
- Results: Named in 46% of 265 graded AI answers; cited as a source in 60% where sources were shown; AI Overview citations from 4 to 6 of 59 markets in nine days; 72 to 76% in the strongest markets against 9.8% nationally.
- Built with: my Audit Engine (growth plan), Article Factory and Optimization Loop (citation panel).
