Two engagements, documented

Our first published results: two engagements — one documented arc, one still in motion — both anonymized at the clients’ preference. Every claim below traces to dated documentation — Google’s own monthly reports, our own test reports, dated search captures, a recorded test. Where the record is thin, we say so instead of padding it.

About the evidence panels: the screenshots in our records show client names and personal details, so we don’t publish them raw. Each panel is recreated from that documentation — same dates, same numbers, same order — with identifying details removed.

November 2025 – June 2026

Case study: a Sacramento window-tinting shop

The situation

A mobile window-tinting shop with over a decade in business and 5.0-star reviews was fading on Google. Its November 2025 Business Profile report — Google’s own scorecard, emailed to the owner monthly — showed 145 profile views, down 43% from the month before, and zero calls for the month.

What we found

The shop had two Google Business Profiles. Google emailed two separate November reports for the same business — one for the listing customers actually used, and one for a listing almost nobody did. The second drew 58 profile views that month and produced nothing: no calls, no website visits, not one interaction.

Entity gap

One business presenting as two. Profile views split across the pair, and anything reading them — a customer or an AI assistant — had two versions of the same shop to reconcile. This is the gap we clear before any other.

Business Profile report — Nov 2025 · main listing

name withheld

Profile views
145 −43%
Calls
0 −100%
Website visits
6 −53%
Interactions
6

Business Profile report — Nov 2025 · second listing

name withheld

Profile views
58
Calls
0
Website visits
0
Interactions
0
Recreated from the two November 2025 Google Business Profile report emails in our records; identifying details removed.

What we fixed

A Concierge-scope engagement: we made the fixes ourselves rather than handing them off. We consolidated the shop’s identity down to one listing and built out the profile that customers and assistants actually read — services, service area, hours, photos, and details kept consistent everywhere they appear. We published location pages for the nearby markets the shop serves — Folsom, El Dorado Hills, Rocklin — and installed rich-results schema on the site, so its facts are stated in a form machines can read.

What changed

The monthly reports turned around. February 2026: 158 profile views and 4 calls. March 2026: 283 profile views — up 79% in a month — with 7 calls and 18 total interactions, three times the November count. The March report reached us the way evidence should: forwarded by the client, dated, straight from Google.

Monthly Business Profile reports, main listing

Monthly Business Profile reports, main listing
Google reportedNov 2025Feb 2026Mar 2026
Profile views145158283
Calls047
Interactions61218
Recreated from the client’s monthly report emails. December and January reports aren’t in our records, so they aren’t in this table.

The search page moved too. In captures from March 19 and April 14, 2026, “mobile window tinting in sacramento” put the shop in the top three of Google’s local Businesses panel — 5.0 stars alongside competitors with hundreds more reviews.

Google “Businesses” panel — “mobile window tinting in sacramento” — March 19, 2026 capture

  1. Competitor A · 4.9 stars (519 reviews)

    Window tinting service · 7+ years in business

  2. name withheld · 5.0 stars (25 reviews)

    Window tinting service · 10+ years in business

  3. Competitor B · 4.9 stars (312 reviews)

    Window tinting service · 7+ years in business

Recreated from a March 19, 2026 search capture; competitor and client names removed. An April 14, 2026 capture of the same query shows the shop third of three, at 5.0 stars with 27 reviews.

Then the part we were hired for. On June 25, 2026, Google’s AI Overview answered that same query by naming this shop first.

Named by the assistant

June 25, 2026 — the AI Overview for “mobile window tinting in sacramento” opens its list of top-rated local providers with this shop, ahead of every competitor on the page.

Google AI Overview — June 25, 2026

name withheld:“Highly rated by Sacramento locals for bringing premium ceramic tint directly to homes, offering flawless, bubble-free installations that block out intense summer heat.”

The first provider named in the AI Overview for the query. Recreated from a dated screenshot; identifying details removed.

Honest limits

December and January reports aren’t in our records, so the timeline above skips them. Search results shift by day, location, and searcher. And nobody can promise what an assistant will say — this page shows a documented sequence, not a guaranteed outcome.

November 2025 – August 2026

Ongoing case study: a Fargo-area healthcare practice

This engagement is still in motion, so we publish it the only honest way we can: as an ongoing case study — what’s documented, what’s still moving, and what we can’t yet claim.

The situation

The practice’s search results mix it with similarly named businesses. Our initial audit, prepared in November 2025, led with exactly that: the practice’s own site introduced its practitioner by courtesy title and surname only — and that surname is shared with other practitioners and practice names, some elsewhere. An assistant that can’t tell which practitioner it’s reading about tends to recommend no one rather than risk being wrong.

The recorded test

The clearest artifact in the record is a recorded test from May 2026. In a natural, consumer-style conversation about booking a routine appointment in the area, the assistant listed two practitioners at the practice: the real one, and a second who does not exist. The phantom name traced back to a single patient review that had misspelled the real practitioner’s name. When the tester said they’d contacted the practice and no such practitioner worked there, the assistant withdrew the claim and acknowledged the error.

Recorded AI test — summary of the exchange

The setup
A consumer-style conversation with an AI assistant about booking a routine appointment in the area, recorded on video.
What the assistant said
It listed two practitioners at the practice — the real one, and a second who does not exist.
Where the phantom came from
A single patient review that misspelled the real practitioner’s name.
How it ended
Told that the practice had been contacted and no such practitioner existed, the assistant withdrew the claim and acknowledged the error.
Summarized from a recorded test on file; names and identifying details removed. We’re preparing the clip for publication.

One wrong source

This is the kind of error that gets called a hallucination, as if it were unknowable. It wasn’t. A lightweight model pulled a name from one bad source and repeated it — an accuracy gap feeding an entity gap. One wrong source, one wrong answer. And that’s fixable: correct the source, and the answer follows.

What got fixed

Two entity gaps, closed. The phantom practitioner was traced to the review that created it. And the audit’s leading recommendation — the practitioner’s full name, stated wherever the practice introduces her — is now live on the practice’s site, implemented by their web team. The confusion between similarly named practitioners has not recurred in our testing since.

On July 27, 2026, we walked the client through the recommendations still open. Over the following week, implementations began appearing on the practice’s site — and our test rounds were running on both sides of that week.

What changed

Each test round puts ten questions to five AI assistants and keeps every answer. Five of the ten are discovery questions — asked the way a customer asks before they know any names. On July 31, no assistant named the practice on any of the five. On August 6, two of the five produced an answer naming it — both from assistants that search the web before answering.

The citations moved a little earlier, which is what they do. Citation Share— whose sources assistants read while answering — tends to move before the answers themselves, because a fix changes the reading list before it changes the recommendation. In the July 31 round, the practice’s site was cited once across the seven questions that never say its name. In the August 6 round: four times.

AI visibility test — July 31, 2026

name withheld

Discovery questions naming the practice
0 of 5
Assistants naming it at least once
0 of 5
Site cited on unnamed questions
1

AI visibility test — August 6, 2026

name withheld

Discovery questions naming the practice
2 of 5
Assistants naming it at least once
2 of 5
Site cited on unnamed questions
4
Recreated from our July 31 and August 6, 2026 client test reports — ten questions to five assistants per round — with identifying details removed.

Honest limits

Five questions can’t settle a naming rate — our own client reports print wide ranges around these counts, and the July and August ranges overlap. A week of movement is an early signal, not a settled result. Some recommendations also remain open, and the record reflects that partial adoption. This engagement first published here as a field note — one recorded test, not enough to call a case study. It now has the start of its before-and-after record, and it keeps the ongoing label until the record is finished.

See what AI says about your business

On a short call, we show you what three assistants answer about your business today, which sources they cite, and one obvious gap — free, no obligation.

Free AI Visibility Check