When someone asks ChatGPT, Gemini or Perplexity what to use in your category, a handful of products get named and the rest are never considered. Here is exactly how we find out which side you are on.
What happens
Paste your website. We work out what your product actually is and what a buyer would type when looking for it, then turn that into the real questions people ask: "best X for Y", "alternatives to Z". Your own product name never goes in the question, because a question containing your name guarantees an answer containing your name and teaches you nothing.
What you get
The exact questions we asked, shown to you. You can copy any of them and run it yourself.
What happens
Each question gets one live web search, and the answer comes from ChatGPT, Gemini and Perplexity. Never from training memory, which is where a small product simply does not exist. Whether you were named is decided by looking for your name in the text, not by asking a model whether it thinks you were mentioned.
What you get
How many of the questions named you, which of the three assistants did, and which competitors were named instead.
What happens
We record every page each answer was built from. This is the part that turns a score into an instruction: assistants write from what they retrieve, so a product absent from those pages is absent from the answer however good it is.
What you get
The ranked list of domains deciding your category, and a roadmap ordered by how much each move would change the answers above.
We measured this rather than assuming it. Across the 39 of 40 buying questions where all three answered, only 30% of products were named by all of them, and 47% were named by exactly one.
Checking one assistant shows you 58% to 65% of the products named across the three, depending on which one you pick.
The failure that causes is specific: a founder checks ChatGPT, sees their name, and stops worrying. Gemini, which their buyer probably met inside a Google search without choosing it, might not name them at all, and nothing in that founder’s process would ever reveal it. The full method and numbers are in the study.
This category is full of scores with no denominator. Ours always has one, and you can see the questions behind it.
Real questions
Generated from what your product is, never seeded with your own name, and shown to you in full so you can check them.
Real answers
Live web queries every time. If a question fails, it is reported as failed rather than quietly dropped to make the score look better.
Real sources
The pages each answer was actually built from, not a guess about what AI might be reading.
A count, not a feeling
Named in 3 of 10 is a measurement. "Improve your AI presence" is not. When nothing has run yet we say so instead of printing a zero.
The line we won’t cross: we will not tell you why a model chose what it chose. Nobody outside those companies knows, and anyone selling you a list of “AI ranking factors” is selling a guess. What can be measured honestly is the question asked, the pages retrieved, and the names that came back. That is what we measure, and it is enough to act on.
I’m Oğuzhan, a solo dev. I shipped products that were good and completely invisible to the people they were made for. Asking an AI for the best tool in my own category returned other names, every time, and there was nowhere to see that happening. No impression, no click, no line in any dashboard. A buyer decides against you inside somebody else’s chat window and you never learn it happened.
So I built the thing that looks. It is built in public on @oguzhanglotier, numbers and all, and I run the exact same playbook on Glotier that it runs for you. Assistants already refer real people to us, and we publish the method. I would rather ship less and tell the truth than overpromise.
Paste your website. No account. You get the questions, the answers, and every source behind them.