AI Visibility
See exactly what ChatGPT says about your brand, and whether it recommends you or a competitor when a buyer asks.
Check My VisibilityChatGPT is the assistant most of your buyers reach for first, which is why it carries the highest weight in our combined score. A ChatGPT brand check means asking it the questions your customers ask, in their words, and recording where your name appears in the answer, how it is described, and which competitor is named when you are not. This page covers how to do that and what the results mean. Live ChatGPT coverage inside our own check is in progress; Gemini is live today.
Last updated
Understanding which one is failing determines what you should do, and most advice ignores the distinction entirely. If ChatGPT names you from training, your brand was mentioned often enough and consistently enough across the web that it settled into the model. That is slow to build and slow to lose. If it names you from browsing, it ran a search and read the pages it found. That is fast to influence and fast to lose. A brand that is invisible in both needs authority. A brand visible in browsing but not training is newer than the model, which is fine and simply takes time.
The category of question changes who wins, and most brands are strong in exactly one. Comparison prompts favour whoever has published the comparison, which is very often a competitor writing about you. Best-of prompts favour whoever appears in listicles and roundups. Price prompts favour whoever publishes real pricing rather than "contact sales", because a model cannot quote a number it was never given. Review prompts favour whoever has G2, Capterra, or Trustpilot presence. Seeing which of the eight categories you lose tells you what to build, and it is usually not more blog posts.
A mention can hurt you. ChatGPT naming your product as the one people find hard to set up, or as the expensive option, is a mention that costs you the sale. Sentiment carries 20% of our score, equal with position, because "mentioned" is a genuinely misleading measure on its own. When a check shows negative sentiment, the source is usually traceable: a widely-cited review, a Reddit thread, or a comparison page written by a competitor that the model has read and you have not.
These models are non-deterministic. Run the same prompt twice and you can get two different lists, which means a single answer is an anecdote and a set of answers across categories is a measurement. It also means you should be sceptical of any tool reporting a precise score off one question, and of your own instinct to re-roll a prompt until you appear. The number that matters is the one across the full prompt set, tracked over time.
In the checks we run, absence from ChatGPT is rarely a website problem. The site is usually fine. What is missing is corroboration: the model has one source saying what you do, and it needs several, ideally ones it already trusts. That means third-party coverage, review platforms, comparison pages that name you, and a presence in the communities models cite heavily. It is slower than an SEO fix and considerably more durable.
No, and this is the part people underestimate. An answer typically names three to five brands. Everything else in the category is absent, not ranked lower, which is why "we are on page one of Google" does not carry over.
We send the prompt and record the response. You can verify any individual result by pasting the same prompt into ChatGPT yourself. The one caveat is that models are non-deterministic, so your answer may differ in wording while showing the same pattern.
Yes, considerably. With browsing enabled, ChatGPT can find and cite pages published yesterday. Without it, you are looking at what the model absorbed during training. A brand can be visible in one mode and invisible in the other, and the fixes are different.
Not by asking it to. There is no submission form and no paid placement. What changes the answer is changing what the model reads: more consistent, more authoritative, more corroborated content about you across sources it already trusts.
Monthly. Retrieval results shift week to week and training data shifts with model releases, so anything more frequent mostly measures the model, not you.