Whether you appear
The share of answers in which the model names your brand, per question and per engine.
AI visibility
Find out whether AI models recommend your company
Buyers increasingly ask a model instead of scanning search results. We ask those questions for you, on a schedule, and show whether you appear in the answer, who appears next to you and which pages the model cites.
Product under construction. The packages and limits below describe what is being built, not something you can buy today.
Four things that can actually be counted
We do not sell a single score, because a single score never tells you what to fix. Every number keeps its numerator and denominator so you can take it apart.
In what order
Where you appear relative to the competitors you track. This is the order in the text, not a ranking the model produced, and we call it that.
Who appears beside you
The competitors you name, counted exactly the way you are. Without them you cannot tell your own drop from a shift across the whole category.
What it cites
The domains the model leans on. Being named builds recognition, being cited brings traffic, and those are two different things.
How we measure
The method is part of the product, not small print: it decides what the numbers on the chart mean.
- We ask through the providers' official APIs
- We do not impersonate a user in someone else's app and we do not scrape answers off a screen. An answer obtained through an API can differ from what the same person sees in the app, and we say so plainly.
- Every question in a fresh session
- No history, no personalisation, with the country set explicitly. That is what makes two measurements comparable instead of dependent on what someone asked earlier.
- A trend, not a single shot
- The same prompt yields different answers on different days. We show the run over time together with how many answers the result came from, so a market shift is distinguishable from noise.
- We did not ask is not the same as you did not appear
- Answers we failed to collect stay out of the result. Counted into the denominator, they would lower your visibility precisely when our own infrastructure was the thing that failed.
Site readiness audit
Before a model can cite you, it has to be able to read your site at all. We check that separately, without asking any model.
Who you let in
Your robots.txt rules, separately for the bots that gather training data and for the ones that fetch pages at answering time. Blocking the latter rules you out of citations.
Whether the content is visible
Whether your page carries content in raw HTML or only paints it with a script. Some bots run no scripts and see an empty page.
Whether you can be understood
Structured data, headings, question and answer sections, sitemap, llms.txt. Every finding comes with a reason and a specific fix.
The audit reads only domains you point us at, with the same bot that identifies itself in the header and honours robots.txt. We do not fetch anyone else's pages: competitors are recognised by name and by domain in citations, never by visiting them.
Three things you can do right away
None of them asks a model anything and none has an external provider, so they work regardless of when the measurement starts.
An llms.txt from your knowledge base
A list of your most important pages that models read instead of guessing your site structure. We build it only from what you published as public, and we check by hash that the file really sits on your site.
schema.org structured data
An Organization, LocalBusiness or FAQPage block to paste into the head section. It is built only from what you type: no model adds anything here.
A report with the numbers
What changed in the period, which questions you are missing from, and which pages the model cites instead of you. The database computes the numbers, not a model, so the report exists even when the description comes straight from the data.
The site audit, the llms.txt file and the structured data need no subprocessor at all: the whole job happens between us and your site.
In progress
What is still being built
The measurement is written and tested against stubs, but it does not yet ask real engines: every provider is a new subprocessor and the list carries its own notice period. The parts below are under construction and we promise no dates until they work.
Measurement on real engines
We start with OpenAI models with web search. Every further provider is a new subprocessor, so they arrive one at a time and only under a data processing agreement.
See the subprocessor listMention sentiment
Whether the model recommends you, mentions you neutrally or advises against. Judging that takes a second model call, so it will be switched on separately.
Ask for detailsWriting content
Articles and question and answer sections assembled from your knowledge base, always as a draft for a human to approve.
Ask for detailsWant to know where you stand today?
Tell us what your customers ask. We will show you what the models answer, and what can be improved about it.