Say you run a two-truck HVAC shop in Chandler, Arizona. Call it Summit Heating & Air. Tonight, somewhere in your service area, a homeowner's AC dies and instead of googling, she asks ChatGPT: "who's a reliable AC repair company in Chandler?"
It answers in about four seconds. It names the 40-truck outfit across town, plus two others. It does not name Summit.
Nothing about that was random, and almost none of it was about company size. AI assistants build recommendations from a short list of sources they can crawl and verify, and most shops Summit's size are invisible on half of them. That's fixable, and right now it's fixable cheaply, because almost nobody in the trades is doing the work.
This is what people mean by GEO: generative engine optimization. SEO earns you a spot on a results page with ten links. GEO earns you a spot in an answer that names three companies and ignores everyone else.
A results page has a page two. An AI answer doesn't. You're named, or you don't exist.
What the assistant read before it answered
When an assistant recommends an HVAC company, it's drawing on a few sources it trusts because it can check them:
- The company's website, read as data. The 40-truck outfit's site tells crawlers, in machine-readable schema markup, exactly what it does, where it operates, and when it answers the phone. Summit's site says "Comfort is our business!" over a stock photo. One of these can be quoted in an answer. The other forces the AI to guess, and it doesn't guess — it moves on.
- The review corpus. Not volume. Recency and specific language. "Fixed our AC" is filler; "diagnosed a failed capacitor and had the part on the truck" is the kind of checkable detail an assistant repeats when someone asks who's good with older units. Summit's five-star reviews all say "great service!!" — which reads as enthusiasm, not evidence.
- Consistency across the open web. Same name, same phone number, same service list on the website, the Google Business Profile, Yelp, and the trade directories. If Summit's Google profile says "Summit Heating and Air LLC," closes at 5, and lists "AC repair," while the site says "Summit HVAC," closes at 6, and lists "cooling services," every mismatch lowers the AI's confidence that it knows who Summit even is. Low confidence means no mention.
Notice what's missing from that list: ad spend. There is currently no way to buy your way into the organic answer. That won't stay true forever, which is a reason to move now, not later.
Fix one: say what you are in a language crawlers read
The highest-leverage afternoon in this post is adding LocalBusiness schema to the website. Here's roughly what Summit's would look like — swap in your own details and hand it to whoever built your site:
{
"@context": "https://schema.org",
"@type": "HVACBusiness",
"name": "Summit Heating & Air",
"telephone": "+1-480-555-0142",
"url": "https://example.com",
"address": {
"@type": "PostalAddress",
"streetAddress": "1234 W Example Rd",
"addressLocality": "Chandler",
"addressRegion": "AZ",
"postalCode": "85225"
},
"areaServed": ["Chandler", "Gilbert", "Tempe", "Mesa"],
"openingHours": "Mo-Sa 07:00-18:00",
"makesOffer": [
{ "@type": "Offer", "itemOffered": { "@type": "Service", "name": "AC repair" } },
{ "@type": "Offer", "itemOffered": { "@type": "Service", "name": "Furnace repair" } },
{ "@type": "Offer", "itemOffered": { "@type": "Service", "name": "HVAC installation" } }
]
}
That block does one job: it turns "Comfort is our business!" into facts an AI can verify and repeat. Every field is something an assistant would otherwise have to infer — and inference is where small shops lose to big ones, because big shops leave less to infer.
Fix two: ask for evidence, not stars
Change what you say after a job. Stop requesting "a quick review" and ask one question instead: "Would you mention what we actually fixed?" Ten reviews that say things like "replaced the blower motor same day" beat a hundred that say "great service!!" — the AI is looking for evidence, not enthusiasm, and specific reviews are how a two-truck shop out-credentials a 40-truck one on the exact jobs it's best at.
Fix three: one version of the truth, everywhere
Pick the canonical name, hours, phone, and service list. Then enforce them on the site, the Google Business Profile, Yelp, Nextdoor, and every directory that mentions you. Boring, yes. It's also the difference between an assistant that vouches for Summit and one that hedges and names somebody else.
Then re-run the homeowner's question monthly. Ask ChatGPT, Perplexity, and Google's AI mode who they'd call for AC repair in each city you serve. Screenshot the answers. That screenshot log is your GEO scoreboard, and it's more honest than any ranking report.
FAQ
Do people actually hire businesses an AI recommends?
A growing share do, and they behave differently: someone who got two names from an assistant calls both, decides fast, and rarely comparison-shops further. Fewer, warmer leads. The channel is small compared to Google today, but it's compounding, and early position is cheap.
Is GEO just SEO with a new name?
They overlap about halfway. Clean structure, real content, and consistent business data help both. GEO adds a bias toward being quotable: assistants prefer sources that state checkable facts plainly over pages optimized to rank. A page written to be cited reads differently than a page written to be clicked.
Can I just pay someone to "do GEO" for me?
You can, and most of what's sold under that label is the three fixes in this post at agency prices. Do them first. If you want the deeper layer — entity consistency at scale, structured data beyond the basics, content built around the questions assistants actually get — that's the point where help earns its fee.