Most HVAC websites are invisible to AI search. The top 7% aren't.

What we did

In May 2026, we audited 207 HVAC websites across 25 US cities — the businesses that actually answer the phone when a furnace dies — to see how visible they were when a homeowner asks ChatGPT, Perplexity, Gemini, or Google AI for HVAC repair near them.

Here’s what we found.

What “the average HVAC site” actually looks like

There isn’t one. There are three groups.

Score	Sites	Share	What it means
80–100	15	7%	AI search can identify them and recommend them
30–79	163	79%	Coin-flip visibility — found sometimes, missed sometimes
0–29	29	14%	Invisible. AI search can’t see the business at all

The 7%. When a homeowner asks ChatGPT “best HVAC repair near me,” these sites can be cited. Their pages declare — in a form a machine can read — what they do, where they are, when they’re open, and why customers love them.

The big middle. Some signals present, key signals missing. The page renders fine for humans. AI search picks them up on some queries and misses them on others, and there’s no analytics dashboard that tells you which.

The 14% invisible. Nothing for a machine to read. The website looks fine to a human visitor. It does not exist for an AI. These businesses are losing customers they will never know about, because the customer never landed on the site to begin with.

What separates the 7%

The top 15 sites are doing the same 9 things. The bottom 29 aren’t doing any of them. There is almost nobody in between.

The gap isn’t gradual. It’s binary. You either do the 9 things or you don’t. They’re all in the playbook — with copy/paste code — grab it from the form at the top if you haven’t already.

The basics are fine. The AI layer isn’t.

92% of HVAC sites have the basics that mattered to Google in 2010 — page titles, descriptions, the homework your web person learned a decade ago. They’ve nailed the test AI search doesn’t grade anymore.

Reviews in a form AI can actually read? 13% of sites. Services described so AI can pick them up? 23%. Social profiles linked so AI knows “yes, that Facebook page is this business”? Almost nobody. The things AI search cares about — the things that turn into “here are the best HVAC repair companies near you” — are the things HVAC websites do worst.

The form is correct. The substance is absent.

The weird part

5.8% of HVAC sites paid someone to add the AI-readable code. It doesn’t actually work. The capitalization is wrong. The labels are wrong. It compiles fine and AI search reads it as nothing. They paid for code they don’t have.

That’s roughly 1 in 17. Not an edge case. The fix takes about two minutes once you know what to look for. (We put that in the playbook too.)

Don't be one of the 17.

Download the free playbook — 9 steps, copy/paste code, and the 5 mistakes to check your own site against.

Email addressDownload the playbook

Free, no fluff. We'll save your email so we can let you know when we publish new research.

What you can actually do

Three real paths:

Hand-write it yourself. A few hours for a developer who’s done it before. Hours of frustration if not. Easy to ship; easy to ship broken. (See: the previous section about paying for things that don’t work.)

Use a platform that generates this by default. Launch10 does. So do a couple of others. The same data that defines the top 7% comes out as a default — not as a feature you have to remember to configure.

Hire an agency. $1,500–$5,000/month and the quality varies wildly. Pre-engagement question to ask: “Show me the structured data on three of your existing HVAC client sites.” If they don’t have an answer, they’re the wrong agency for what this report is about.

Methodology, briefly

207 single-location and small-multi-location HVAC operators across 25 US cities, sourced from Google Maps + BBB + ServiceTitan directory backfill. National chains (Roto-Rooter, ARS, Service Experts, et al.) excluded to keep the sample focused on operators small businesses actually compete with. Each site fetched as raw HTML — what AI search crawlers actually see, since most don’t run JavaScript before they read. JSON-LD blocks were parsed and validated against the canonical schema.org type vocabulary (933 types, snapshotted May 5, 2026). The scoring rubric is open-source and the raw data is available on request — email [email protected] with the subject “HVAC 2026 raw data.”

This report measures readiness, not citation outcomes. We’re not claiming these specific top-scoring sites are the most-cited HVAC sites in Perplexity yesterday. We’re measuring whether the site has the signals AI search needs to cite it at all. Readiness is the prerequisite. Citation is what readiness compounds into.

---

This is the first of a planned series of vertical AI search readiness audits from Launch10 Research. Plumbing, roofing, dental, and legal are next. If your industry isn’t on that list and you’d like it audited, get in touch.