I Measured Whether AI Knows My Site Exists. It Doesn't.

On 7 August 2026 I ran a mention search across ChatGPT responses for three things: my domain, my name, and my product. All three came back empty. Not low, not a handful. Nothing.

I am publishing that number because a baseline is only worth anything if it is recorded before you start working on it. Everyone writing about AI visibility right now is describing a problem. Very few of them will tell you their own number, and there is a reason for that.

What exactly did I measure?

Three separate mention searches against a dataset of AI assistant responses, run on the same day, looking for any occurrence of each term. This is the same measurement the research engine behind my own product runs for other people’s brands.

Two honest caveats, because a number without its limits is a claim rather than a measurement. First, this is a sample of AI responses and not every response ever generated, so the correct reading is “does not appear in the sampled set” rather than “has literally never been shown to anyone”. Second, “Vantage” is a common word and a product name used by other companies, so a zero there is if anything the most surprising of the three and the one I would most expect to move first for reasons that have nothing to do with me.

Why is it zero?

Because nothing has told these models that I am a source on anything. That is the short version and it is not mysterious.

The supporting numbers point the same way, and they are not flattering either.

A model builds its sense of who is credible on a topic from the same web everyone else reads, and on that web I have been publishing quietly and linking to nobody who links back.

None of that is a technical fault. The site is fast, it is indexed, the markup is clean. I mention that because the instinct when a number comes back at zero is to go looking for a broken thing to fix, and there usually is not one.

Is zero actually bad?

It is a starting line rather than a verdict. Absence from AI answers means you have not been established as a source on those questions yet, which is a different problem from being judged and rejected.

It is also, briefly, an advantage. A zero is unambiguous. There is no argument about whether the trend is real, no seasonality to adjust for, and no debate about attribution. Any movement at all is signal. Most measurement problems in marketing are hard because the baseline is noisy, and mine is not noisy at all.

What am I changing?

Publishing on the specific questions I want to be the answer to, structured so the answer can be lifted out of the page, and getting the same claims to appear somewhere other than my own domain.

That last part is the one I underrated. A model gets comfortable naming you when the same claim from the same person shows up in more than one place, and a personal site alone is a single source with an obvious interest. So the plan is not only to write here. It is to say the same things on the channels where I already have an account and have been neglecting, and to earn a mention or two somewhere I do not control.

The structural change is smaller than it sounds. Ask the question in the heading, answer it completely in the first two sentences underneath, then explain. Not because a model rewards the format for its own sake, but because a passage that answers a question on its own is a passage that can be quoted without dragging the paragraph around it. Google says much the same from the other side, that structured data is not required for generative AI search and that content people find useful will influence presence there in the long run more than anything else it suggests. That is convenient for me, because useful writing is the only asset on the list I already have. The reasoning and the four metrics that measure the outcome are in LLM visibility and how to measure it, and the method that produced today’s number is in how to run an AI visibility audit.

What would count as working?

The first non-zero reading, on any of the three searches. That is the whole success criterion for now, and deliberately so.

I am not measuring this in rankings. Several of the terms in this space sit at a difficulty my domain cannot reach on Google for a year or more, and if I judged these pages on position they would look like failures while doing exactly what I built them for. Being cited in an answer does not require the authority that ranking tenth requires. Those two things come apart, and this is the year they came apart.

When do I report back?

Monthly, on roughly the same date, with the same three searches. The next reading is due in early September 2026.

I will publish it whether or not it moved. That is the part that makes this worth reading: a public baseline that only gets an update when the news is good is not a measurement, it is a campaign. If it is still zero in September, the September post will say so and will be shorter.

There is an obvious thing to say here about the product, so I will say it plainly and once. Vantage exists to run this kind of comparison for other people, against their competitors, and it returned zero for me. I would rather tell you that than quietly benchmark myself out of the sample. The reports are AI created and you should read them with your own judgement, as I read mine. Your first analysis is free, and getting your own number is the only part of any of this that cannot be argued with.

Mine is zero. Dated, published, and about to be tested.