AI Search Is a Conversation, Not a Keyword

A facility manager with a leaking gym roof doesn't open ChatGPT and type "roofer for schools." Nobody does. What she types is closer to this:

"We need to replace the roof on our middle school gym over the summer. It has to be finished before teachers come back in August, and the crew can't be on the roof during summer camp hours. Who are commercial roofers near Charlotte that have done school work?"

That's one question with five requirements in it. A keyword tool would boil it down to "commercial roofer Charlotte" and throw the other four away. The AI doesn't throw them away. It goes looking for a company that matches all five.

I tell clients this all the time. In AI, people don't search "Concord plumber." They search something like, "I've got a leak under my shower, I'm in a two-story home in Concord, give me the best good-value plumber who can come out today." It's a long, specific question. The more specific information your site has, the more of that question you can answer.

The short version

People search AI tools by describing their situation, not by typing a keyword. Google says early users of AI Mode wrote queries two to three times as long as a normal search. Those extra words are qualifiers: the kind of building, the deadline, the constraint, the town. The contractor who gets named is the one whose website, profiles, and reviews already answer those qualifiers in plain language. Keyword research still tells you which topics have demand. It just doesn't tell you what the question sounds like anymore.

What the question actually looks like

Here's the same need, written two ways.

The left column is a category. The right column is a job, with the details that make it this job. That second homeowner isn't asking for a plumber. She's asking for someone who understands hard water, will tell her honestly whether tankless makes sense, and will still be around in year three to flush it.

This isn't new behavior, by the way. It's how people have always described a job on the phone. What's new is that the thing on the other end is now reading your website to decide whether you're the answer.

What happens on the other end

When a question like that comes in, the AI doesn't search it as one string. Google calls it "query fan-out": the tool breaks the question into smaller searches (school roofing experience, summer scheduling, commercial roofers in the Charlotte area) and runs them all at once. Then it looks for a business that shows up in the answers to most of them.

So each qualifier is a test. Have you done school work, and does anything on the internet say so? Do you mention working around an occupied building or a fixed deadline? Do your reviews describe a job like this one? Does your map profile list commercial roofing as a category, in this area?

A company that does all of that work but never writes any of it down fails the test. Not because it's worse. Because there's nothing to match.

Which of those places the AI checks, and in what order, is its own topic. I've laid that out in the four places AI checks before it names a contractor. This piece is about the other half: what the question looks like, and what your site has to say to answer it.

It happened to me

I can tell you this works because it's how a good part of my own pipeline shows up now.

In September, a co-owner of a home improvement company booked a strategy session with me. I asked how she'd found me. She'd been hearing a lot about one of the big contractor agencies, and she asked ChatGPT, in her words, who else was "specific for contractors, maybe more smaller boutique style." It gave her two names. I was one of them.

Look at what she didn't type. She didn't type "SEO agency." She typed three qualifiers: contractor-specific, smaller, boutique. Those three words match things I've written down on purpose, over and over. I only work with contractors. I keep it to ten clients. You'll never talk to an account manager, because it's just me. The AI didn't need a keyword. It needed someone who matched the description.

She was the second prospect in six weeks who found me that way.

What to write instead

You don't need to guess every question someone might ask. You need your site to say, plainly and specifically, the things that tend to show up as qualifiers. For most contractors that's five things.

1. Who you work for. "Homeowners" is a category. "Older homes in the historic district," "property managers with 20-plus units," "school districts and universities" are qualifiers. If you have a type of customer you're good with, name them.
2. Which jobs you want. Not a list of 40 services in a grid. A real page for each job you want more of, written the way you'd explain it on a site visit. If you want to be named for a whole-home repipe and not for a $90 drain call, your site has to talk about repipes more than drains. If you’re not sure which jobs to pick, that’s a good first conversation for a free strategy session.
3. The constraints you handle. Occupied buildings. Tight deadlines. Hard water. Steep lots. HOA approvals. Lender letters. These are exactly the details people put in a question, and exactly the details most contractor sites leave out because they feel too small to mention.
4. Where you actually work. The towns, by name, with jobs you've done there. Not a paragraph listing 30 cities. Proof in a few.
5. What customers say about it. Reviews that describe the job ("they replaced our gym roof in three weeks and worked around camp") match a question in a way that "great job, very professional" never will.

None of this is a trick. It's the stuff a good salesperson says on the first call. The change is that it now needs to be written down somewhere a machine can read it before anyone calls at all.

Where keywords still matter

I'm not telling you to throw out keyword research. Plenty of people still type "water heater repair near me" into Google, and the map pack still answers them. Search volume is still the best way to know which topics have real demand, and which jobs are worth building a page for in the first place.

What's changed is what you do once you've picked the topic. A page written to repeat "tankless water heater installation Hendersonville" six times reads like it was written for a machine from 2015. A page that explains when tankless makes sense on hard water, what it costs to install and maintain, and what you'd tell a neighbor who asked, answers the keyword search and the conversation. You don't have to choose. You just have to write for the person.

What this changes on Monday

If you want to see where you stand, try this. Write down the last three good jobs you sold. For each one, write the question that customer would have asked an AI, in full sentences, with every detail they gave you on the first call. Then read your website and see how many of those details it mentions.

Most contractors find that their site answers the category and none of the qualifiers. That's the gap. It's fixable, and it's mostly writing. It follows the same order you build it in for any search work: the right pages first, then the profiles that confirm them, then the reviews that describe them.

And since 45% of consumers now use AI tools to find local businesses, up from 6% a year earlier, according to BrightLocal's 2026 consumer survey, it's worth doing before your competitor does.

If you'd rather have someone run that audit with you, that's what my AI search optimization work starts with.

FAQ