How AI picks which contractor to recommend
By CJ Jewell
When someone asks who to hire, the assistant searches the live web, reads what it finds, and summarizes the businesses that several trustworthy sources agree on. Companies that are described consistently everywhere, reviewed recently, and mentioned by sites other than their own are the ones that survive that process.
It helps to drop the idea that ChatGPT holds a directory of businesses in its head. It does not. When a homeowner asks about contractors in a specific town, the assistant goes and looks, the same way you would, and then writes up what it found.
Once you picture it as a very fast researcher rather than a database, the whole thing gets easier to reason about. So: what does that researcher actually weigh?
It reads the live web, not a private list
The assistant runs searches, opens pages, and pulls out what it needs. Which means the same forces that decide what surfaces in ordinary search decide what it sees in the first place. If nothing about your company ranks for anything, there is nothing for it to read.
This is the single most useful thing to understand, and it is why treating AI visibility and search visibility as two separate projects is a mistake. They start from the same place.
It prefers claims it can corroborate
A researcher writing a recommendation does not want to stake it on one source, and neither does the software. Your own website saying you are the best roofer in town is a claim. Two hundred Google reviews, a chamber listing, a supplier’s certified-installer page, and a local news mention all pointing the same direction is evidence.
This is why businesses that are only visible on their own website rarely get named. There is nothing to check them against.
It leans hard on structured facts
Some information is written for people, and some is written for software. The second kind is a set of plain labeled facts attached to a page: this is a roofing contractor, it serves these towns, its rating is 4.9 across 212 reviews, its phone number is this.
Nobody sees that. Software reads it first because it requires no interpretation. A page carrying it is dramatically easier to quote confidently than a page where the same facts are buried in a paragraph of marketing prose, and confidence is what determines whether your name makes it into a short answer.
Reviews count, and recency counts more than people expect
Rating, volume, and how recent they are all matter. Recency does more work than most people assume, because it answers a question the assistant genuinely needs answered: is this business still operating and still good? A strong rating whose newest review is from three years ago is a weaker signal than a slightly lower rating with reviews from this month.
Consistency decides whether you are one company or three
If your name, address, and phone number appear in conflicting forms across the web, software has to work out whether those listings are one business or several, and it may fail. A recommendation requires being confident about who it is recommending. Inconsistent details undermine that quietly, and nobody ever tells you it is happening.
Specific answers get quoted, general claims do not
Look at how these answers are actually written. They contain particulars: a price range, a timeline, a service area, a qualification. Those particulars have to come from somewhere.
A page that states real numbers gives the assistant something to lift. A page that says “quality workmanship, competitive pricing, free estimates” gives it nothing, because those words are on ten thousand other contractor sites and mean nothing on any of them.
What nobody can tell you
The exact weighting is not published, it differs between assistants, and it changes. Anyone claiming to know the formula is selling something.
What is stable is the shape of it. Every one of these systems is trying to answer the same question, which is: which business here can I recommend without embarrassing myself? Being the obvious answer to that question is a strategy that survives model updates, because it is not a trick. It is just being genuinely, verifiably good and easy to confirm.