How to tell whether AI assistants recommend your product

Someone asks an assistant to recommend a tool in your category. It names three companies. You want to know whether you were one of them.

You cannot find that out from Google Search Console, because no search happened there. You cannot find it from your analytics, because an answer that mentions your name without linking to you sends no traffic. The entire interaction is invisible to the measurement stack most teams run.

Why this is a different problem from ranking

Search gives you a list and you pick. An assistant gives you an answer and you mostly take it.

That changes what winning looks like. In search, position three on a page of ten still gets clicks. In an answer that names three products, being the fourth-best candidate gets you nothing at all. There is no scroll.

It also changes the unit of measurement. Rank is a number per keyword per location, and it is stable enough to track daily. An assistant's answer is generated fresh, varies between runs of the same question, and differs by model. There is no position to record. What you can record is whether you were mentioned, and what was said about you.

What to actually measure

The useful question is not "how do I rank in AI". It is "for the questions my buyers ask, how often am I named, and alongside whom".

That means starting from questions rather than keywords. Not "ai seo tool" but the things a person types when they have a problem:

What is the best tool for writing SEO content automatically? Which SEO tools work for a solo founder with no budget for an agency? What should I use instead of hiring a content writer?

Then you run those questions, repeatedly, and record three things.

Whether your brand appeared. A yes or no per question per run, which over enough runs becomes a rate rather than an anecdote.

Which other brands appeared. This is the most useful output and the one people skip. The set of companies named alongside you is your real competitive set, and it is frequently not the set you track. We have watched a product discover that the competitor it benchmarked against never appears in answers, while a company it had not heard of appears in most of them.

Which sources the answer drew on. When an assistant cites, the citations tell you which pages are feeding the answer. That is where the opportunity is: the page being cited is usually not yours and usually not even a competitor's. It is a comparison article, a Reddit thread, a listicle on a site you have never written for.

What moves the number

Being cited is mostly downstream of being the clearest source on a specific question.

Write pages that answer one question completely. Assistants synthesise from passages. A page covering eight loosely related subjects gives a model nothing clean to lift. A page that answers one question, with the answer stated plainly near the top, does.

Say the thing rather than building up to it. A lot of SEO content withholds the answer for four paragraphs to keep people scrolling. That format was already tiresome for readers and it is actively harmful here, because the passage a model extracts is the one that contains the answer. If your answer is in paragraph nine, surrounded by throat-clearing, something else gets quoted instead.

Be specific enough to be quotable. Numbers, versions, prices, concrete tradeoffs. Vague competence reads as filler to a reader and extracts as nothing to a model.

Get named on the pages that already get cited. If a comparison article is feeding answers in your category, your absence from it is the gap. This is closer to PR than to SEO, and it is slower, and it works.

Mark up what you are. Organisation and product structured data, a clear statement of what the product does in plain language on the page rather than only in a hero image, an llms.txt if you want to be explicit. None of this is magic. It removes ambiguity about what your company is, which is the cheapest win available.

The honest caveat

Nobody can promise you a position in a generated answer, and any tool that does is selling you something it cannot control. The output varies between runs. It varies between models. It changes when the model changes, which is often.

What you can do is measure it over time instead of guessing, see which questions you are absent from, and look at who is being named instead. That is a report, not a ranking, and it is the right shape for a thing that genuinely does vary.

Visorank runs that measurement on a schedule: your questions, repeated runs, mention rate, the brands named alongside you, and the sources the answers drew on.