Commerce

Ask the Assistant What a Customer Would: A Small Firm's Guide to Checking AI Answers

Owners of local service companies can learn a good deal by putting plain buying questions to AI assistants, but only if they ask more than once and keep notes.

The American Intelligencer standing plate
From the pages of The American Intelligencer.

Shoppers have long leaned on tested recommendations before spending money. CNN Underscored, the network's shopping and reviews section, published a piece titled a bidet buying guide, with a headline saying that bidets "save you money and reduce waste" and that its reviewers tested the options. The product is beside the point here. What matters is the habit it reflects: people want someone else to narrow the field. Increasingly, they ask an AI assistant to do it, including for services such as a plumber, an electrician, a roofer or a heating technician.

For the owner of a small local company, that raises a practical question. When a customer types a buying question into an assistant, does the business come up, and how is it described? The answer is not hard to find out, though it takes some discipline.

Start with the customer's words

The most useful questions are the plain ones a neighbor would ask. "Who should I call for a water heater that stopped working in my town?" "What should I look for when hiring a company to replace a roof?" "Which local companies do drain cleaning, and what do people say about them?" Avoid questions that include the company's own name at first, since most customers do not know it yet. Add the town or metro area, because local intent is the whole point.

Use a logged-out browser or a private window. Assistants can tailor responses to an account's history or saved preferences, and a business owner who has searched for his own firm many times may get a result no customer would see. Logging out does not make the test perfectly neutral, since location and settings can still matter, but it gets closer to what a stranger would encounter.

Ask more than once, and write it down

A single answer is an anecdote. Ask each question several times, on different days if possible, and try more than one assistant. Keep a simple log with the date, the assistant, the exact wording, which companies were named, in what order, and how each was described. Note whether the answer cited sources, and whether any claim about your firm, such as service area, hours or specialties, was wrong.

Paper or a spreadsheet will do. The value lies in the pattern across entries, not in any one line.

Why the answers move

Owners are often surprised that the same question yields different companies from one try to the next. This is expected. These systems generate text rather than retrieve a fixed list, so wording and selection can vary between attempts. Some assistants also consult live web sources in certain modes and not in others, so what they say can shift with what they were able to find. Slight changes in phrasing, the town named, or the product version in use can alter the result as well.

That variability is the reason for repeating the test. If a company appears in most attempts, that tells you something. If it appears once in ten, that tells you something different. Neither is a verdict on the quality of the work.

What a handful of checks can and cannot show

A modest set of checks can reveal real, fixable problems. An assistant may list the wrong service area, describe the company as doing work it does not do, or lean on an outdated business listing. It may repeat a complaint from a review the owner never answered. It may consistently name competitors for a service the firm actually offers, which suggests the public information about the firm is thin or unclear.

What a handful of checks cannot show is market share, trends over time, or how many customers saw any given answer. Twenty attempts are not a statistical sample, and no owner should treat them as one. They also cannot prove why a competitor was named. Without access to how a given assistant weighed its sources, any explanation is a guess. The honest conclusion from a small test is usually modest: here is what a customer might plausibly be told, here is what looks wrong, and here is what to look at next.

Becoming as routine as reading reviews

Home-service owners already know the drill with online reviews. They read them, respond to them, and correct the record where they can. Checking AI answers is shaping up as a companion chore, for a similar reason: the first impression of a firm is often formed before anyone calls it. A quarterly check, or a monthly one in a busy season, takes under an hour with a prepared list of questions.

The fixes tend to be unglamorous. Make sure the company's website states plainly what it does, where it works and how to reach it. Keep business listings consistent across directories. Answer customer reviews factually. Some firms bring in outside help, and there is a growing body of AI search work for local service companies, but an owner can do the first round of checking alone and learn a great deal.

No one can promise that any change will make an assistant recommend a particular firm. The sensible goal is narrower: know what is being said, correct what is verifiably wrong, and keep the public record of the business accurate. Customers increasingly ask a machine for a shortlist. A careful owner can at least find out what the machine says.