A few weeks back I told you about two lies.
The first was the impressive number your dashboard hands you. The second was asking AI whether you are brilliant, because it will always say yes.
Here is the third one. I did not see this one coming, and it happened to me, in my own house.
I was digging through the code of my own website last week and found five questions and answers sitting in the part of a page a person never sees. The part only machines read. Questions like “what services does Lefty Media Co offer for website development.”
Nobody wrote those. Not me. Not anyone who works with me. My own AI wrote them and published them, quietly, on four pages.
And it described us as a web development agency.
If you have been here more than a couple weeks you know that is the one description I have spent a year and a half climbing out of. I do build websites. That is not the thing I sell and it is not the thing I am. My own machine put it in writing anyway, and pointed it at every AI crawler on the internet.
So I went looking for who typed it. Nobody typed it.
What happened is simpler and worse. The part of my setup that writes those descriptions had our business name, our city, a short list of services, and nothing else. No page content. It failed to read the site and kept going anyway. And then it was told to state what the business is.
Give a machine a name and no facts and it will reach for the most average answer available. Lefty Media Co. Something about websites. Something about AI. Web development agency. Every single time.
That is the same thing I wrote about last week. I asked an AI that did not know me for the best coffee in Kansas City and it named a chain. Not because the chain is better. Because the chain was what it had. My own AI called us a web development agency for the exact same reason. It had our name and nothing behind it, so it reached for the average and said it with total confidence.
An AI is a mirror. When there is nothing in front of it, it does not say “I don’t know.” It makes something up that sounds about right.
Now here is the part that actually got me.
I did not catch this alone. I have three AIs. One builds, one thinks, one writes. I put two of them on the problem and let them audit each other for most of a day. They found real problems. Every round turned up something true. It was good work and I was proud of it.
And in four straight rounds of checking each other, neither one of them noticed that we had completely walked away from the thing I asked for at the start of the day. The actual assignment. Gone, hours earlier, and both of them were too busy being right to look up.
A person caught it. That person was me, and it took one sentence.
I had been treating it as proof when two of them agreed. It is not proof. They are built the same way. They think in the same grooves. They are wrong in the same places. Two of them agreeing is not two opinions, it is one opinion said twice, a little louder.
That is worth sitting with if anybody is running AI on your business right now, and somebody probably is. The thing checking the work cannot be the same kind of thing that did the work. It will nod. It will nod convincingly, with citations, for hours.
I did not build an assistant. I built a witness. And last week the witness needed a witness, and it turned out to be a guy in Kansas City who remembered what we were trying to do in the first place.
That is the job. Turns out it has always been the job.
Next Tuesday I want to show you something I found while all this was going on. I pulled the numbers on a client AI has never once recommended, and looked at every business it recommends instead. They all have one thing in common. It is not their website, and it is not a trick.
See you next Tuesday.
Nathan. Lefty Media Co, Kansas City.