Same Prompt, Different HVAC Picks: Our Gemini Search Test

From A Chat with Chat, episode 2, recorded September 30, 2026. Watch the full episode or listen to the audio.

Would an AI tool recommend your HVAC company when a homeowner asks for help? A single answer can show whether you appeared in that moment. It cannot establish a permanent ranking or explain why the system chose you.

In episode 2, Aaron Husak and ChatGPT explored that distinction through repeated Gemini searches for an HVAC company in Phoenix to help with an AC problem. Aaron compared answers in Flash and Pro and examined the visible sources. The conversation began with a hypothesis that different models might draw on different kinds of material. The resulting sample raised useful questions, but did not prove a general rule about how either model selects contractors.

Recurring names did not mean identical answers

Some companies recurred across the Flash runs discussed in the episode. The answer format, company count, order, and visible sources also changed. That is evidence of overlap within those runs, alongside variation.

It is tempting to turn a recurring shortlist into a ranking report. But the observation is narrower: certain names appeared repeatedly for this question under the conditions of that session. We did not establish that another homeowner, location, date, or phrasing would produce the same result.

A source link is a clue, not a selection formula

Examining linked pages helped us see what information was available about the recommended businesses. Service coverage, AC repair information, and statements about urgent help were relevant to the question.

That does not prove those page features caused a recommendation. A company's own statement about fast service is also a claim to evaluate, rather than independent evidence of response time. Likewise, an answer without a visible source panel does not establish that no web information informed it.

The practical next step is to save both the answer and its visible citations. Keep a business mention separate from a link to that business's website. A recommendation and a citation can occur independently.

The pricing question remained unanswered

We also asked whether published pricing might explain the results. The Pro-linked pages discussed in the episode did not provide a standard AC repair price. The test therefore did not support a conclusion that price drove those recommendations.

Publishing accurate diagnostic fees or explaining how estimates work can help customers make decisions. It should be evaluated on that basis, with clear terms. This episode does not demonstrate that adding a price will improve AI visibility.

Make the next comparison more disciplined

Our recording was an exploratory conversation, not a controlled experiment. A stronger follow-up should hold the question and target market consistent, run each model multiple times, and record the conditions. Comparisons across different tools should also acknowledge their different search experiences.

The working question asked Gemini to find an HVAC company in Phoenix for an AC problem. The recording documents the September 30 session and the selected Flash and Pro modes, but does not establish a version-controlled or independently reproduced benchmark. Use this recap to plan a better test, rather than as a competitive ranking report.

1. Use an actual customer question, including the service need and location.
2. Record the tool, selected model, date, and relevant account or location conditions.
3. Save every response, including runs where your business does not appear.
4. Log recommended companies and cited sources separately.
5. Compare patterns over time, then investigate what changed.

Repeated tests can make observations more useful. They still do not reveal a proprietary selection formula, and a visibility trend is not proof of additional jobs booked.

What contractors should improve now

Make your real service coverage explicit. Explain which AC problems you handle, how someone contacts you, and what urgent-service availability actually means. Support those statements with accurate qualifications, original work examples, and consistent information across your website and profiles.

When a customer says AI sent them, capture the question they used and how they chose your business afterward. Those details connect visibility research to the customer journey.

Watch episode 2, read our introduction to GEO and SEO, or discuss your own AI search measurement. The aim is useful evidence about discovery, followed by verified inquiries and jobs booked.