What ChatGPT says about your business partly comes from your customers

What ChatGPT said when she asked about the retailer

ChatGPT told a search consultant to shop somewhere else. She had asked it about a retailer she was working with, by name, and the answer came back with reasons attached: product quality, inconsistent materials, returns and refunds. Every one of those started as something a customer wrote down.

The consultant is Jessica Bowman, who has spent more than 20 years doing search work inside large organizations. She told the story to Search Engine Land in an interview published on 9 September. This is one consultant's account of her own client work, with no study or sample behind it, and what follows is what she watched happen.

She had been trying to get ChatGPT to recommend that retailer and couldn't. Asking about the company head-on is how she found out why. Bowman called it completely operational. The things ChatGPT named happen at a loading dock and a service counter, on days nobody in marketing hears about.

Bowman describes two jobs these systems do: repeating back what they already know about a company, and weighing whether that company suits the particular person asking. Her phrase for the second job is "the world's most sophisticated buyer advisor," she told Search Engine Land. An advisor reaches for whatever it can find about you.

I read these answers most days at Vyzz, where visibility comes down to whether an assistant says your name when a customer asks. On audits for dental practices, HVAC companies and vet clinics, what comes back about a small business is a lot shorter than what Bowman got, and it usually still carries a line the owner never wrote.

What it reads besides your website

Bowman says she has watched these systems draw on customer complaints, software release notes, earnings calls, vendor case studies and employee reviews when describing a company. That list belongs to big companies. For a practice or a shop with one location, the same idea comes down to what other people have already written about you in places you don't control. That means reviews on the platforms, a thread on a local forum, a complaint post, or whatever a former front-desk hire wrote about working for you.

None of that was written by you, and none of it can be edited by you. A review from four years ago sits there in plain text with a date on it, and ChatGPT can read it as easily as it reads your services page. It reads that review flat. To ChatGPT it's a report from somebody who dealt with you, and the context you'd add if you were standing there never makes it into the answer.

Bowman says the systems turned this pattern up on their own, by reading whatever people had already put online about a company. In some niche business-to-business markets she has watched AI draw conclusions where relatively few comments about a company or product exist at all.

How often an assistant names anybody at all

One study in January put a number on how rarely a local business gets named at all. SOCi looked at performance data covering nearly 350,000 locations belonging to 2,751 companies with a lot of locations, and Search Engine Land wrote it up on 28 January 2026.

SOCi's January estimate is that getting named by an assistant runs three to 30 times harder than showing up on Google the normal way, in the map results and the blue links. Those are January figures, measured on companies much bigger than yours. Read them as a fair warning about what you are likely to get back.

There's a decent chance the answer comes out thin, or names other businesses and never reaches you. A blank is a different job from a complaint, and it starts with whether the plain facts about you are published anywhere a machine can read them. When an assistant does name a business, it describes that business in some detail, and the description is built out of whatever it found.

The words it uses about your competitors

Bowman ran one comparison for a client that's worth copying. She looked at how AI described her client next to how it described that client's rivals, and the two answers did not read the same. Her client came out positive. The rivals got the stronger language, the sentences about where they were good and why.

That gap is hers. You only see this kind of thing by asking about both. If you check what an assistant says about you and stop there, you've got half the picture. Ask the same question about the two competitors you keep losing work to, then read the adjectives it uses about them.

The two-minute test

Open ChatGPT in a window where you're not signed in, type your business name and your town, and ask whether it's a good choice. The signed-out window matters, because your own account has heard plenty about your business from you and it will be kind about it. Read the answer slowly and mark every place it hedges.

Answers move around. Bowman says they change from person to person and inside a single chat, and a company recommended early can drop out as the assistant learns more about the buyer. Her example is shopping for a HEPA vacuum. ChatGPT steered her away from the model with the strongest reported suction, because it knew from an earlier conversation that she'd dealt with an illness that left her fatigued, and it judged the heavier machine a poor fit. "It shows how it's really considering your context," she said.

What comes back is one answer, on one day, for a customer the assistant knows nothing about. Run the check a few times over a month before you trust any pattern in it. Doing that by hand across seven assistants is the tedious version, and it's what vyzz.io automates, though the ChatGPT check on its own takes two minutes.

Some of what comes back is your own website talking. Your services, your hours, the paragraph you wrote about yourselves. The part worth your morning is the rest of it.

Whatever it names as a reason to hesitate is something a person already went through and wrote down somewhere. A woman wrote that nobody called her back. One review says the quote changed after the work started.

If the answer never reaches you at all, write that down too.

Write each hesitation on a list. Next to it, put the part of the business that produced it: who picks up the phone, and who handles it when a job goes wrong. That's a Monday meeting for a dental practice or an HVAC company, and nobody had to sell it to you.

One honest limit before you go and do it. I can't promise you, and I'd be selling you something if I did, that fixing callbacks changes what ChatGPT says about you next quarter. Newer reviews land every week, and an assistant reads whatever is sitting there on the day somebody asks. Watch the wording month to month.

Bowman gives three questions to run every month. "Are we visible when people ask? Are we mentioned? How are we described?" she told Search Engine Land.