The shortlist just got shorter

What happens to a local business when a machine answers the question “who should I call?”

A homeowner in the northwest suburbs finds a wet patch on the ceiling. Ten years ago they opened a phone book. Two years ago they opened Google and got a page of options to sort through. This month, a lot of them will type the question into an assistant and get four or five names — with reasons attached.

That is not a search result. It is a shortlist, and for most people it is the entire shortlist.

So in August we ran the obvious test. We asked ChatGPT and Google AI Mode fifteen ordinary homeowner questions about roofers in Palatine and Arlington Heights, and checked the answers against the roofing companies we had on our list for these suburbs.

What actually changed

The change isn’t that the internet got smarter. It’s that the middle step disappeared.

A page of search results is a menu. You skim ten of them, open three, form your own opinion. An AI answer is a recommendation. It picks the names, puts them in an order, and — this is the part people underestimate — tells you why each one is good. “Family-owned since 2002.” “Known for honest storm-damage assessments.” “Praised for meticulous cleanup.” The homeowner doesn’t just get the name. They get the reason, pre-formed, before they’ve clicked anything.

And the list is short. Not ten links: about five names. Two or three by the time the assistant says “if it were my house, I’d call these first.”

For a business, that turns a familiar problem into an unfamiliar one. You are used to competing for attention. Now you’re competing for inclusion. If you’re not in the answer, you don’t lose the comparison — you were never in it. Nobody scrolled past you. Nobody decided against you. You simply weren’t part of the conversation, and neither you nor the customer will ever know it happened.

Three things anyone would notice

The two assistants barely agreed. Between them they recommended thirty-seven different companies. Only eleven came up on both. Ask one assistant and you get one town’s roofers; ask the other and you get a substantially different set. There is no single “the AI says” — there are several machines with different reading habits.

Neither one ranked by reviews. This was the biggest surprise. Companies on our list with hundreds of glowing Google reviews were absent from every answer. One with more than 500 reviews and a rating just shy of five stars was never named a single time. Meanwhile a Palatine company with fewer than three dozen Google reviews was recommended in ten of the thirty answers.

Both quoted the companies back to themselves. The flattering reasons in the answers — the years in business, the certifications, the specialities — were, over and over, the companies’ own words from their own websites, repeated as though they were findings. The assistants weren’t judging these businesses. They were relaying them.

So why do some show up?

Nothing exotic. There is no trick, no hidden setting, no secret code to paste into a website. Google’s own published guidance says exactly that — there is nothing special to do to appear in its AI features, and no special markup to add — and our results agree. The assistants leaned on the same public, checkable things any careful neighbour would look at. The difference was in how available those things were.

They existed in more than one place. Every company that came up repeatedly had a life beyond its Google listing — a review profile somewhere else, a trade directory, a manufacturer’s contractor listing, a chamber of commerce, a local Facebook thread. One assistant leaned heavily on Google profiles and Yelp; the other leaned on the Better Business Bureau, Angi and “best of” roundups. A company living on only one of those was invisible to at least one machine.

They said, in writing, where they work. The most-recommended companies all had a page on their own website for each town they serve — not a dropdown, not a list in the footer, an actual page. And those exact pages turned up as the assistants’ cited sources. A company that serves Arlington Heights but never writes the words “Arlington Heights” anywhere a machine can read them is, as far as the machine is concerned, not an Arlington Heights roofer.

They repeated themselves consistently. Same name, same address, same story, everywhere. When the details agree across five sources, a machine treats them as fact. When they conflict, it hesitates — and hesitation looks exactly like absence.

Their own words became the recommendation. Since the assistants relay what a business publishes, what a business publishes is what gets said about it. That cuts both ways: the ones with a clear, specific, honest description of what they do got described clearly, specifically and honestly. The ones with a website full of stock phrases had nothing worth quoting.

Reviews aren’t worthless in any of this — nobody gets recommended with bad ones. But reviews turned out to be the floor, not the lever. Excellent reviews got these companies nothing on their own.

One more thing worth saying plainly, because it surprised us: we checked the “best roofers in town” roundup that one assistant cited most often, and it did not contain most of the companies that assistant actually recommended. Being on the list wasn’t the mechanism either.

The three names that kept coming back

Across the thirty answers, three companies came up again and again: Holda Construction in Palatine, named first more often than anyone else; Smart Roofing in Arlington Heights, recommended in seventeen of the thirty answers and named for both towns; and W. Brothers Roofing in Palatine — the one with fewer than three dozen Google reviews, recommended ten times, mostly by Google AI Mode, which cited its website in ten of its fifteen answers.

AI answers change constantly — this is what the assistants said during our study window (August 22, 2026), not a permanent ranking, and not our endorsement of any company’s workmanship.

The contrast between that last one and the 500-review company that nothing mentioned is the whole study in a sentence. Reputation and visibility are not the same asset, and one does not automatically buy the other.

What this means if you run a local business

Not a roofer? It doesn’t matter. Nothing above is about roofs. It’s about how a machine decides which handful of businesses to say out loud when a person asks a question in your category, in your town.

The honest summary is short. Be findable in more than one place. Say plainly and in writing what you do and where you do it. Keep those details identical everywhere they appear. And accept that the reasons a machine gives for recommending you will be your own words — so it’s worth having some worth repeating.

None of that is new advice. What’s new is the cost of ignoring it. A business that skipped this five years ago ranked a bit lower. A business that skips it now can be left out of the answer entirely, in a market where the answer is five names long.

What this study is not

One trade, two towns, two assistants, one day, and one list of fifty-six companies — our list, not a count of every roofer in these suburbs. The answers drift: ask the same questions next month and some names will move, which is exactly why the date is printed on everything here. We found things the recommended companies had in common; we can’t prove those things caused the recommendation. And we looked only at what the assistants said, not at who got called afterwards.

Everything above is from public sources, viewed the way any customer would view them. We didn’t contact a single one of these businesses.

If your company is named here and you believe a fact about it is wrong, tell us at hello@trycentron.com — we’ll check the source and correct it.

We study how contractors get recommended.
— the Trycentron team