What AI Answer Engines Know About You, and What They Get Wrong

What this covers

• The Audit

• Error Type One: You Do Not Exist

• Error Type Two: The Details Are Stale

• Error Type Three: The Invented Detail

• Error Type Four: You Are Described in Someone Else’s Words

• Why the Answer Changes Between Runs

• The Gate That Makes Everything Else Pointless

• Where an Assistant Gets Its Picture of a Business

• Correcting a Wrong Answer

• What Cannot Be Promised

• Do the Audit This Week

Your customers are asking assistants about your business. Not hypothetically, and not only the young ones. The question is what those systems say back, and almost no owner has checked.

It takes twenty minutes to find out. Here is the audit, the four kinds of error you are likely to see, and what each one tells you about the data underneath.

The Audit

Open an assistant. Any of them. Ask these five questions and save the answers with today’s date.

1. What do you know about [your business name] in [your town]?

2. Who would you recommend for [your main service] near [your town]?

3. What are [your business name]’s hours?

4. What services does [your business name] offer?

5. How do people describe working with [your business name]?

Do it in a fresh session so there is no earlier conversation shaping the reply. Then repeat the whole thing in a second assistant, because they draw on different sources and frequently disagree.

Write down what came back. Do not correct anything yet.

Error Type One: You Do Not Exist

The reply is generic, or names competitors, or says it has no specific information.

This is the most common outcome for small businesses and it is not a mistake on the system’s part. It is an absence. There was not enough corroborating information across independent sources for the model to say anything confident.

What it points at: thin presence outside your own website. One website is one source, and it is you describing yourself. These systems want several sources agreeing. Directory listings, a mention in local press, a community page, a supplier’s site, reviews with actual detail in them.

Worth checking immediately: whether you have a Bing Places listing. Several assistants lean on Bing for live local lookups, and a business with no Bing listing is starting from close to nothing on that path regardless of how it performs on Google. It is free and takes under an hour.

Error Type Two: The Details Are Stale

Right business, wrong facts. Old hours, a phone number you stopped using, an address from two moves ago, a service you discontinued.

What it points at: your information exists in more places than you are maintaining. A phone number changed in 2019 still sits on a dozen directories that nobody updated, and those directories are being read.

This one matters more than it looks. Wrong hours send a customer to a closed door, and that customer does not blame the assistant.

The fix is unglamorous: find every listing carrying your details and make them agree. Name, address, phone, identical everywhere. Conflicting details also make a business look unreliable to a system that cannot call you to check, so consistency is doing two jobs at once.

Error Type Three: The Invented Detail

The assistant states something specific and wrong. A founding year, a specialty you do not have, a location you do not serve.

This is the one that alarms owners and it is worth understanding rather than panicking about. A model asked a direct question with insufficient evidence will sometimes produce a plausible-sounding answer built from patterns rather than facts.

What it points at: a gap where your own clear statement should be. The model is filling a vacuum. If your site states plainly what you do, where you do it and how long you have done it, in ordinary readable text rather than inside a graphic or behind a tab, there is less vacuum to fill.

What does not work: telling the assistant it is wrong. The correction lasts for that conversation and changes nothing underneath.

Error Type Four: You Are Described in Someone Else’s Words

The reply describes you accurately but in language that came from a review, a directory blurb or an old article, and it emphasizes something you would not have chosen to lead with.

What it points at: your third-party presence is stronger than your first-party presence, which is not a bad problem. It means the corroboration is there. What is missing is your own clear description competing for attention.

It also shows you which reviews are doing the talking, which is useful. Reviews that name the specific job, the town and the outcome get quoted. Reviews saying “great service, highly recommend” do not, because there is nothing in them to lift.

That is the practical change worth making: when a customer offers a review, ask them to mention what the job was. Most will. It alters what the evidence about your business looks like far more than getting ten more generic five-star ratings.

Why the Answer Changes Between Runs

One warning, because it causes unnecessary alarm and unnecessary celebration.

These answers are volatile. Run the same question an hour apart and you can get a different set of businesses named, or an AI summary in one run and none in the next. A single result is a snapshot, not a verdict.

Which means two things. Do not panic at one bad answer. And do not accept anyone’s claim that a single good answer proves their work is succeeding. The signal is the trend across months, recorded the same way each time, which is why the audit above is worth repeating on a schedule rather than running once.

The Gate That Makes Everything Else Pointless

There is one technical condition that overrides all of the above.

The crawlers these systems use do not run JavaScript. If your website assembles its content in the browser, which a great many modern builds do, those crawlers receive an almost empty document. Your services, your town and your phone number are not in what they see.

Test it before anything else. If that is your situation, no amount of listing and review work will register, because the system never receives your side of the story.

The measurement and repair process sits under answer engine optimization and a good share of it costs nothing. The business profile shows the market it is written from, and there is more on what assistants actually weigh. We have worked in local search here for over a decade and we run this audit on our own listing before selling it to anybody.

Where an Assistant Gets Its Picture of a Business

The sources are not the ones classic search optimization concentrates on, which is why a business can rank well and still be described incorrectly.

The last row carries a trap. A site that assembles its content in the browser can look complete to a visitor and arrive empty at an automated reader, which means the business has no voice in its own description.

Correcting a Wrong Answer

None of that is fast. Corrections propagate over weeks, and there is no support line to call. That is the argument for getting listings right before they matter rather than after somebody notices an error.

What Cannot Be Promised

No supplier can guarantee how an assistant will describe a business, because the systems are opaque and change without notice. What can be done is supplying accurate, consistent, machine readable information everywhere these systems look, then checking the answers periodically to see whether it took.

Anyone selling a guaranteed outcome here is describing a control that does not exist.

Do the Audit This Week

Twenty minutes, five questions, two assistants, one date written at the top.

Whatever comes back, it is better to know. Most owners discover error type one, which feels discouraging and is actually the easiest to fix, because building a presence from nothing is more straightforward than correcting a wrong one.

A last practical note. Check what is currently being said about the business before assuming anything is wrong. Ask two or three assistants the question a customer would ask, in the wording a customer would use, and write down what comes back. Most owners have never done this and are surprised in both directions: sometimes the description is better than expected, and sometimes the business is confidently described as something it stopped doing years ago.