Same questions, a month apart: ChatGPT, Gemini and Perplexity went from naming them 6% of the time to 19%.
Our client is a local service business on the SF Peninsula. We asked AI assistants the same 21 questions its customers would ask, before our changes and again a month later.
Uses three of the five surfaces.
Figure 1 · ChatGPT, Gemini and Perplexity together
19%of answers named the business on Oct 12, 2026
On the same questions, a month apart, ChatGPT, Gemini and Perplexity named the business 19% of the time, up from 6%.
How to read this
Uses three of the five surfaces. Share of answers that named the business: 11 of 189, then 36, then 36. AI answers vary by person and session, so these are measured rates, not a guaranteed position.
Our own measurement for a local service business on the SF Peninsula: the same 21 questions on each date, through each vendor’s API with web search on. Bars are drawn from 0 to 40%.
- Client
- a local service business on the SF Peninsula. Its name is withheld.
- Result
- The share of answers that named the business rose over the month; Figure 1 shows the three readings.
- Surfaces
- Three of the five we audit: ChatGPT, Gemini and Perplexity.
- Questions
- 21 questions, each run 3 times per assistant, 63 answers per assistant on each date.
- Dates
- Sep 10, 2026, Oct 2, 2026 (interim) and Oct 12, 2026 (30-day check).
- Not published
- Google AI Overviews results.
- Read it as
- A measured rate, not a guaranteed position. It cannot prove which change caused the rise.
On Oct 12, 2026 we also ran a wider check: 66 questions across all 10 cities it serves. It was named in 37% of those answers. That is a first reading with no earlier one to compare, so it shows where things stand, not how far they moved.
About Claude: we measured it in September, when it named the business 0 times in 63 answers. We left it out of the October checks only because it costs several times more to measure than the other three assistants combined. The October numbers on this page are ChatGPT, Gemini and Perplexity.
By Jonah Samarron, Founder
1 · What we measured
21 questions in plain customer language (urgent needs, everyday situations, specific services, and “who is best” comparisons), in and around the cities the business serves. None of them names the business. Each question ran 3 times on each assistant, because AI answers change from one ask to the next, through each vendor’s API with web search on. We counted an answer only if it named the business.
The original questions left some of the cities it serves out. So on Oct 12, 2026 we added a wider set: 66 questions covering every city it serves and every service it offers, asked the same way during its business hours (594 answers).
Why Claude is not in the October numbers: our audit covers ChatGPT, Claude, Gemini, Google AI Overviews, and Perplexity, and Claude was part of the September measurement. It named the business 0 times in 63 answers. For a client check-in, Claude costs several times more to measure than ChatGPT, Gemini and Perplexity combined, so the October checks left it out. That was a cost decision, not a way to hide a weak number: the September zero is reported here.
2 · What we worked on in between
- We worked on how the business is presented to search engines and AI assistants: on its own website, on its Google Business Profile, and on the other places they read.
- The specific methods stay with our clients. Each client sees exactly what we change and why.
3 · Tracked since Sep 10, 2026: the same 21 questions
Share of answers that named the business on Sep 10, 2026, Oct 2, 2026 (interim) and Oct 12, 2026 (63 answers per assistant on each date):
Figure 2 · By assistant
Gemini kept rising and ChatGPT held. Perplexity is the one assistant that went backwards.
ChatGPT
Gemini
Perplexity
63 answers per assistant on each date, the same 21 questions. All three together: 11 of 189 (6%), then 36 (19%), then 36 (19%). Bars are drawn from 0 to 40%.
- Links to its own website: ChatGPT linked to the business’s site in 8, then 17, then 17 of 63 answers
The Oct 2, 2026 gain held at the 30-day check. Gemini kept rising and ChatGPT held. Perplexity slipped from 6 to 3, the one assistant that went backwards. For the three assistants together, the rise is larger than the run-to-run noise we see between repeat runs; for ChatGPT alone it is not yet clear of that noise. The mentions are concentrated: on Oct 12, 2026, 10 of the 21 questions produced every mention, and the rest never named it.
By the city each question was set in, Sep 10, 2026 vs Oct 12, 2026:
| City | Sep 10, 2026 | Oct 12, 2026 |
|---|---|---|
| Los Altos | 1 of 18 | 11 of 18 |
| Menlo Park | 1 of 36 | 6 of 36 |
| Atherton | 1 of 9 | 5 of 9 |
| Palo Alto | 6 of 63 | 10 of 63 |
| Its home city | 1 of 9 | 4 of 9 |
| Sunnyvale | 0 of 18 | 0 of 18 |
| Cupertino and Stanford | 1 of 36 | 0 of 36 (outside its service area, not a clean comparison) |
4 · First measured Oct 12, 2026: all 10 cities it serves
This set has no earlier reading. Oct 12, 2026 is its starting point, so these numbers show where things stand today, not how far they moved. Share of answers that named the business:
Figure 3 · A first reading, no earlier one
On a wider set of 66 questions across all 10 cities it serves, the business was named in 37% of answers.
How to read this
Uses three of the five surfaces. A first reading with no earlier one to compare, so never read it as the next step of the 6% to 19% series.
218 of 594 answers, Oct 12, 2026. The three repeat runs gave 37%, 38%, 35%. 54 of the 66 questions named it at least once. Bars are drawn from 0 to 80%.
Perplexity is by far the weakest of the three. By the city each question was set in:
| City | Share of answers | Named |
|---|---|---|
| Portola Valley | 57% | 31 of 54 |
| Atherton | 46% | 25 of 54 |
| Its home city | 44% | 36 of 81 |
| Menlo Park | 39% | 21 of 54 |
| Los Altos Hills | 39% | 21 of 54 |
| Los Altos | 35% | 19 of 54 |
| Woodside | 35% | 19 of 54 |
| Palo Alto | 35% | 28 of 81 |
| Redwood City | 26% | 14 of 54 |
| Sunnyvale | 7% | 4 of 54(a city it serves where it was rarely named) |
5 · What this does and does not prove
It shows a measured rise on the same questions that held for a month, and where the business stands across every city it serves. It does not prove which change caused the rise: AI assistants also refresh what they know on their own schedule, one assistant (Perplexity) moved to a newer model between September and October, and each date is one round of runs. The wider check has no earlier reading, so it cannot show movement. It covers AI answers only, not Google Maps rankings or Google search rankings, and says nothing about calls or revenue. AI answers vary by person and session, so these are measured rates, not a guaranteed position. This is one client, in a different industry from medspas, so your results will differ.
A similar measurement, for your medspa
Our medspa audit measures ChatGPT, Claude, Gemini, Google AI Overviews, and Perplexity, and we re-measure after changes the same way. Email Jonah@sunnyvaleaeo.com or request your audit at the intake form.
Let’s audit your business