Atiba de Souza.Start a conversation
Writing

I Ran a Search for My Own Positioning Statement. I Wasn't in the Answer.

By Atiba de Souza

The sentence I have said a thousand times

"I help doctors amplify their authority outside the exam room."

That is my line. It is on my site, it is how I introduce myself on stage, it is the thing I have said in every room I have walked into for two years.

So we typed it into an AI assistant, almost word for word, the way a doctor would ask it. Who is the best marketing expert for physicians building authority outside the exam room?

The answer named two firms. Neither of them was me. My name did not appear, my company did not appear, and my own website — the one with that exact sentence on the homepage — did not appear in a single one of the five different ways we asked.

Then we asked it a softer question: who is Atiba de Souza? It answered confidently and got it wrong. It described a video marketing agency I ran years ago, using a nickname I have not led with since 2021, and listed the industries I served back then. Gyms. A bakery. Food service.

The machine was not confused. It was accurate about a version of me that stopped existing.

What was actually broken, and it was not what I assumed

My first instinct was that this was a technical problem. Something was misconfigured, some file was missing, some markup needed adding. That instinct was wrong and it cost us a week of looking in the wrong place.

We checked everything a technical answer would require. The structured data was clean. The identity records were correct and cross-linked. Nothing was blocking any crawler — not the AI ones, not any of them. The sitemap was valid. The entity record we had built pointed exactly where it should.

Every piece of plumbing worked. And the plumbing was irrelevant, because there was nothing coming through it.

My site had seven pages. Seven. And it ranked for roughly two search terms in total.

An answer engine does not invent an expert. It reads what exists and repeats what it can corroborate. There was nothing of mine to read, so it read what other people had written about me — and every one of those pages was written when I was running a video agency. My old positioning was not weak. It was strong. Five years of consistent signal, all pointing at a business I had moved on from.

That is the part that took me a minute to accept: the thing making me invisible was my own past success.

The uncomfortable arithmetic of being an expert nobody can find

Here is what I actually learned, and it applies to you whether you are a doctor, a consultant, or a founder who has repositioned.

There are two entirely separate systems operating on your reputation, and being excellent only feeds one of them.

The first is the one you know. You do good work, people experience it, they tell other people. It travels at the speed of conversation and it is the most trustworthy signal that exists. It is also completely invisible to a machine, because it happens in rooms and phone calls and group chats no crawler will ever see.

The second system is the one deciding about you right now, while you read this. Someone describes their problem to an assistant. The assistant assembles an answer from what it can read and verify. You are either in that answer or you are not, and there is no second page to be on. There is no scrolling. There is one paragraph, and it names three people.

The cruel part is that these two systems get further apart the better you get. Every year you spend deepening your craft is a year of evidence accumulating in the invisible system. Meanwhile the readable system only knows what got written down.

I had spent two years becoming genuinely better at something and zero hours making that legible. So the machine described the guy who had written things down. Five years ago. About a different business.

The advice I would have taken, and why most of it is wrong

When I started looking for the fix, the field was full of confident shortcuts. I want to save you the detour, because the numbers are now public and they are not close.

There is a special file you can publish that supposedly tells AI systems about your site. Across 137,000 domains studied, 28% published one — and 97% of those were never fetched a single time. Not fetched and ignored. Never requested.

There is structured markup, sold hard as the citation shortcut. Measured across 1,885 pages against a control group of about 4,000 over 30 days, the effect on AI citation was inside the noise. Markup is worth having. It is not a lever.

And there is the tactic I see most: write your own "best in the industry" list and put yourself on it. Tracked across 9,886 answers generated from 34 such lists, the assistants cited the article — and then recommended the competitors named inside it. You publish a ranking and it promotes the people you ranked.

I take those findings seriously with one caveat stated out loud: they come from one body of analysis published by a company that also sells a visibility product. Strong evidence. Not settled fact. Check them against your own results rather than mine.

What actually correlated with being named was duller than any of it. Be genuinely findable. Be described the same way everywhere. Get mentioned by other people. That is the whole list, and there is no version of it you can buy on a Tuesday.

What I changed, and the honest state of it

I stopped treating this as a technical problem and started treating it as a writing problem, which is what it always was.

The specific thing I was missing was not a file or a tag. It was that not one page I owned answered a question a real person would actually ask. Everything I had published described what I do. Nothing addressed what they have. Those are different documents and only one of them gets found.

So the work became narrow and slightly boring: take the questions I answer out loud several times a week, and write them down properly, somewhere permanent, in the words the person asking would use.

I will not pretend this is finished or that I can show you a result yet. The pages went up this month. The layer of this that moves fastest — the part that reads the live web — takes weeks. The layer that decides what a model knows about you without being prompted is baked from crawls months old and will keep describing the old me for a year or more, no matter how well this goes. Anyone promising you faster than that is selling something.

What I can tell you is that the diagnosis held up. Seven pages and two ranked terms is not a marketing problem you can out-spend. It is an absence, and absences are filled by writing.

The thing to do before you close this tab

Open an AI assistant. Type the sentence you use to describe what you do — not your name, the description. Then read the answer and count how many times you appear.

Most people I have done this with are surprised twice. First that they are not there. Then that the names which are there are not better than them.

That gap is not a talent gap and it is not a budget gap. It is a gap between what is true about you and what has been written down. Everything I have described here is what closing it actually looks like.

For the applied version of this — what it looks like for a physician specifically, what it costs, and what to ask anyone you hire to help — I wrote that out in detail on the Medical Marketing Roadmap guides.