I have a client AI has never picked once. Not once.

We focus on GEO first, and that is simply does AI accurately know who you are and what you do. AI cannot pick you unless it knows you and AI knows our client. We raised his GEO from 3% to 86%. When we ask ChatGPT, Claude, and Perplexity the questions his customers would actually ask. Who should I hire. Who is the best in Kansas City. Hundreds of asks. Not one of them said his name.

What bugged me is that the AI knows him. Ask it straight out and it tells you who he is, what he does, what city he works in. It just will not put him in the answer.

So I quit staring at him and went and looked at everybody it picks instead.

I pulled every business the AI named. Then I went and found where the AI got them from.

They all had the same thing.

Not better websites. His website is good, way better than some of the others. Not more content. No tricks.

Every one of them was on somebody else’s list.

Industry rosters. Directories other people run. Association pages. Somewhere a third party had already written their name down and said this one belongs here.

The AI was not reading their website and deciding they were good. It was reading other people’s lists and repeating who was on them. Somebody asks who is the best, and the machine does not sit there and form an opinion. It goes and finds where somebody already answered that question, and reads it back.

My guy was not on any list. So there was nothing to read him off of.

Now the part I have to tell you about, because we got it wrong first.

If nobody else vouches for you, the shortcut is obvious. Go be the third party yourself. So we did. Last year we built encyclopedia entries for our clients. Every one of them factual. Every one accurate. It looked like a credential and it sat in a place machines read.

I want to be clear about something. We were not trying to pull anything. Everything we do now and forever is white hat. This is a hill we will gladly die on being something anything other than white hat in our work. AI demands this because the rules truly can change any time.

We read the research. The research said an entity record like that was one of the strongest signals you could hand an AI. So we built them carefully, by the rules as they were written down.

In July the volunteers who run that encyclopedia deleted every single one.

Not because we lied. Because the businesses were not notable enough to be there, and because the guy creating the entry was the guy getting paid by the business. They were right on both counts. We did not know any better at the time because we did not do our own research.

Here is what I actually learned, and it is bigger than a few deleted entries.

Reading the research is not the same as doing the research.

The rules for this work are not written yet. Everybody quoting a best practice for AI right now is quoting something somebody guessed at a year and a half ago, when the models were different. If you do not go test it yourself, on your own clients, with your own data, you are repeating a guess with more confidence than the person who made it had.

We were following rules that were never really set. That is not an excuse. That is the whole lesson. This work makes you do your own research or you find out the hard way, in public, on somebody’s site who trusted you.

And here is the part that matters most to me.

A machine did not catch this. No dashboard went red. No alert fired. Every automated check we had said that client was fine.

I caught it because I was in his account and something felt off, and I kept pulling the thread until I found a link on his site pointing at a page that did not exist anymore. It had been sitting there almost a month. That is on me, not on him, and I told him that in those words.

There is a whole industry right now selling AI visibility as a thing you can automate. Set it, let it run, check the dashboard. I do not believe that, and I have the receipts.

The software is genuinely useful. I built most of ours myself. It counts, it watches, it flags things I would never have seen. It has never once told me why. Why is a person. Why is somebody sitting there at eight at night going that number looks wrong, and then finding out it was.

Nobody has a playbook for this. I say that out loud to clients on the first call. Anybody telling you they have it figured out is selling you certainty they do not have.

So back to the lists, because the fix is not complicated. It is just slow.

You cannot put yourself on the list. That is exactly what makes a list worth anything. The second you can add yourself, the machine has no reason to trust it, and neither does a person. Every shortcut into a credential kills the thing that made it a credential.

What works is boring.

Get on the lists that already exist in your line of work. The rosters, the directories, the association pages, the ones that were there before you and will be there after you. Somebody else decides who gets on. That decision is the entire value.

Get the credentials you can actually earn. Licenses. Memberships. Awards you win instead of buy. The local paper writing about you because you did something worth writing about.

Put something out with your name permanently attached in a place that keeps its own catalog. A book gets a number from a system that does not care who you are. Nobody can delete that.

And get your identity in one stable place. Same name spelled the same way, pointing at the accounts you actually control, so when a machine goes looking for who you are it finds one answer instead of four. That last one I can do for somebody without ever touching their website, and it is what I spent today doing for the client I have been telling you about. It is not the finish line. It is a starting place nobody can take away from him.

None of this is fast. I keep wanting there to be a fast version. There is not one.

The businesses beating you in AI answers are not better at AI. Most of them have never thought about AI one time in their life. They are just on lists you are not on, put there by people who are not them.

Somebody else has to say your name first. Your job is to be worth saying.

Nathan. Lefty Media Co, Kansas City.

Next Tuesday: I counted the robots reading one of our sites and felt pretty good about the number. Then I found out how many of them were lying about who they were.