How to check in one afternoon whether AI knows your business exists

You can do this today, for nothing, in about three hours. What you find will probably surprise you.

By Aaron Stalberg · 22 September 2026

There is a version of this you can run yourself without buying anything. It takes an afternoon, it is mildly tedious, and almost nobody does it, which is part of why the findings are so consistently surprising.

You need a browser, a spreadsheet, and the willingness to write down answers you do not like.

Before you start

Sign out of everything, or use a private window. Assistants personalise, and your account history will flatter you. Use the same method for every assistant so the comparison is fair.

Step 1: write ten buyer questions

Not "who is the best [your category]". Real questions, in a customer's words, about their problem. Include the awkward ones: the geographic version, the cheap version, the version where they are comparing you to a named competitor.

Ten is enough. Fewer than eight and you are measuring noise.

Step 2: ask them in four assistants

Two conversational, one with live search, one that cites heavily. Forty answers total. Paste each one into your spreadsheet with the date.

Step 3: record three things per answer

Are you named? Yes or no. Nothing subjective.
Who is named instead? Build the list. This is your actual competitive set, which is often not the one you would have written down.
Which sources are cited? Copy every domain. This is the most valuable column and the one people skip.

Step 4: read the answers about you

Where you are described, check every fact: what you do, who you serve, how long you have been going, anything stated as current. You are looking for two specific things.

Is anything wrong? Trace each error back to the cited source. It is usually traceable, and usually older than you expect.
Is anything described as current that has ended? This is the common one, and the hardest to spot, because a finished thing described in the present tense reads as ongoing.

Expect two findings the first time. You are named less often than you assumed, and the sources driving the answers are not the pages you have been optimising.

Step 5: build the source list

Count the domains across all forty answers. You will end up with a short list, often under twenty, and frequently dominated by directories and trade press rather than anything you control.

That list is your work plan. Each entry is either something you can fix, something you should ask someone else to fix, or something worth becoming.

Step 6: check your own pages say the obvious thing

Read your homepage as though you had never heard of the company. Can you tell what it sells, to whom, and where it operates? If not, that is the cheapest fix on the list and it should go first.

Step 7: repeat in two weeks

One reading is a snapshot. Two readings tell you whether the position is stable or drifting. After that, monthly is enough.

What this method misses

This is a manual version of a measurement problem, and it does not scale past one category or one company. It also will not hold still: answers change as models update. But it will tell you, in an afternoon and for nothing, roughly where you stand, which is more than most businesses know.

ShowUp Labs runs this continuously, across assistants and questions, so you are not doing it by hand.

ShowUp Labs →