Does AI Even Know You Exist?

Somewhere today, one of your customers asked an AI who they should hire. And you have no idea what it said about you.

Maybe it recommended you. Maybe it recommended your competitor. Maybe it has you confused with a different company that shares your name.

You do not know. And that is the problem we are going to fix.

Hi! I am Bob Michaels, and this is Evolving the Web. I have been building websites since 1994, back before the rules for how to build them had even been written.

Here is how this show works. Every two weeks I publish essays on my blog, and then I sit down and talk through them, in plain language, all connected. This episode covers the last two weeks, which were all about one question. When an AI talks about your business, what does it actually say, and how would you even know?

Let me walk you through it.

The first essay starts with a trap, and the trap is a single score.

There are tools out there that will sell you an AI visibility number. Your brand is a seventy two. Your competitor is a sixty eight. Congratulations.

Here is why that number is close to useless. Think about a checkup at the doctor. The doctor does not hand you one overall health number. They check your blood pressure, and your heart, and your lungs, each on its own. Because if something is wrong, you need to know which thing, or you cannot treat it.

Your visibility to AI works the same way. It is five separate checks, not one.

First, does the AI mention you at all when somebody asks about your category? Second, when it describes you, does it get the facts right, and is it even talking about the right company? Third, does it use your website as its evidence, or somebody else's? Fourth, when it compares you against a competitor, who wins? And fifth, when a buyer flat out asks who to hire, does it say your name, and for what reason?

A company can pass four of those and fail the fifth. One blended score hides which one failed. And if you cannot see which one failed, you cannot fix it.

The second essay is about a sneaky little lie that shows up in this work, and I want you to know it because people will try to sell you around it.

When an AI answers a question, it often points at sources. Links, citations, footnotes. It looks very trustworthy.

But here is the thing. Pointing at a page is not the same as having read it.

Imagine a student turning in a paper with a footnote pointing at a book. Impressive, until you find out the library never checked that book out to anyone. The student never opened it. The footnote is decoration.

AI answers do this. The system goes out and actually reads some set of pages, and then the finished answer points at pages, and those two lists do not always match. Sometimes the answer points at a page nobody ever fetched.

So when somebody tells you an AI cited your website, the question to ask is simple. Did it actually read the page, or did it just point at it? Only the ones it actually read count as real evidence. Everything else is decoration.

The third essay is about how you ask the questions. And this part sounds boring until you see what happens without it.

If you ask an AI the same question twice, worded a little differently, you can get different answers. So if you ask casually this month and casually again next month, and the answer changed, what did you learn? Nothing. Maybe the AI changed. Maybe your wording changed. You cannot tell.

Measurement only works if the instrument holds still. When the doctor checks your blood pressure, it is the same cuff, the same arm, the same way, every visit. That is what makes two visits comparable.

So you build what I call a query library. Real questions, written the way a real buyer would ask them. Who is the best in this category? Should we hire this company or that one? Is this company too small for enterprise work? You write them down, you lock the wording, and you ask the exact same questions on a schedule, writing down what came back every time.

One run is an anecdote. The same locked questions, asked again and again over months, that is a measurement.

And the fourth essay is where this stops being research and starts being work. Because a report that says here is where you are losing is worth nothing if it just sits there.

Every problem you find sorts into a small number of buckets. The AI has you confused with another company. The AI describes your category well but has almost nothing to say about you. The only place your claims exist is your own website, and the AI trusts outside sources more. Or your proof exists but it is locked inside a PDF or a page a machine cannot properly read.

Each of those has an owner and a fix. Get your company's basic facts stated one way, everywhere. Build the one page that states your claim plainly, with specifics a machine can lift and quote. Get one outside mention on a source the AI platforms already trust. Free the proof out of the PDF and into a real page.

Pick the smallest fix for each finding, have a human sign off on it, and ship it. Then, and this is the part people skip, ask the same locked questions again and see what moved. Not to declare victory. To find out what actually worked.

That is the whole loop. Thirty days, one pass. Then you go again.

So here is the two week arc in one breath.

Your buyers are asking AI for recommendations. Whether you show up is not one number, it is five separate checks. A citation only counts if the machine actually read your page. The questions only count if you lock them and repeat them. And a finding only counts once it becomes a fix with an owner and a deadline.

None of this is magic, and anyone selling you magic here should worry you. It is measurement, and then it is work.

Everything I just talked through is written up in full on the blog at bobmichaels dot ai, with the sources and the checklists. The episode page links every essay.

And if you want to know what the AI platforms are saying about your company right now, that is literally the first thing I do for clients. I run the questions, I bring you the answers, and we decide together whether there is work worth doing. I take that on as a consultant, or fractionally, which just means a slice of my week is yours for as long as you need it.

Thanks for listening. Go ask an AI about your own company today. Write down what it says. That is your first data point.