Platform · AI Assistants
Mentient puts real buying questions to ChatGPT, Perplexity, Gemini, Claude and Grok, several times each, and shows you the answers — including the ones that name your competitors and skip you.
5
AI engines queried
3x
samples per question
100%
of answers stored, not just yes/no
Ask ChatGPT the same question twice and you can get two different shortlists. That is how these models work. Any tool showing you a single stable "AI ranking" is hiding that variance behind a number, and the first time your score moves you will not know whether anything actually changed. Mentient asks each question several times and reports how often you appeared. Two of three runs is a real finding. A position is not.
Every question runs three times against every model. You get "named in 2 of 3 runs", which is what the underlying data supports. Ordinary model variance stops looking like a trend.
Including when you are not in it. "Not found" tells you nothing you can act on. Reading what the assistant said instead tells you exactly what you are up against.
We read the companies named in each answer rather than matching a competitor list you had to write from memory. The rival you have never heard of is the one worth knowing about.
When a model states something about your pricing, features or policies that contradicts your own knowledge base, it is caught and shown with the contradiction spelled out.
A local firm is measured on questions that name its city. A global product is measured on questions that do not. You choose the scope; we write the mix.
Every result records the exact model build that produced it. When a provider quietly ships a new version behind the same name, your history stays interpretable.
How it works
Name, website, and anything you want treated as verified fact. Takes a minute.
Realistic buying questions for your category, written from your own data and never naming you — because a question that names you proves nothing.
Five assistants, three passes each, with web search enabled where the model supports it.
Appearance rates, the answers themselves, the competitors named, and any claims about you that were wrong.
Who it's for
A defensible read on whether you appear when buyers ask AI for options, tracked over time instead of spot-checked by hand.
See which engines cite you, which cite a competitor instead, and what they say, so content work is aimed at a specific gap.
Category questions asked on a schedule across five engines, so a shift in how AI describes you shows up as a trend rather than an anecdote.
AI answers are not rankings and we will not present them as if they were. The same question can return different vendors on different days, because these systems sample rather than retrieve a fixed list. A single check is therefore close to meaningless. Mentient puts each question to ChatGPT, Perplexity, Gemini, Claude and Grok several times and reports how often you appear, which turns a noisy signal into a rate you can actually track. It is a less impressive-sounding number than a rank, and a far more honest one.
FAQ
No, and we would rather say so than sell you a number that behaves badly. AI answers vary between identical runs. What Mentient gives you is a diagnostic: how often you appear, what gets said, and who wins the questions you lose. That is genuinely actionable. A position that swings for no reason is not.
Because asking once cannot tell the difference between "never recommended" and "recommended, but not this time". One sample of a non-deterministic system is a coin flip you are reading as a measurement.
No. When you set a market, your location is written into the questions we ask — "best agency in Austin". The AI providers do not expose a way to appear as a user in a given place, so nobody can honestly claim to do that. What you get is how you appear for location-qualified questions, which is a narrower claim and a true one.
Perplexity, ChatGPT, Gemini, Claude and Grok, through a single provider so the set can grow as new assistants matter.
Read the answers you lost. The companies named instead of you are being cited from sources the models trust — reviews, comparisons, directories, community threads. That tells you where to earn presence, which is the work. The rate then tells you whether it worked.