The measurement
On 26 September 2026 we ran a first check for AutoChase, a tool that automatically chases unpaid invoices for freelancers and small businesses. We asked five questions a buyer might type, such as "best invoice reminder software for freelancers" and "Chaser alternatives for small business", across ChatGPT, Gemini, Claude and Perplexity through their APIs.
Fifteen answers came back. AutoChase was named in none of them. Chaser, an established competitor in the same category, was named in five. Upflow appeared once.
That is a small sample from one day, and we would not publish a percentage from it. But the pattern — a newer product invisible, an established rival named repeatedly — is the most common situation we see, and the reasons behind it are consistent.
Reason 1: the model has never heard of you
Asked directly about AutoChase, without web search, the model said it did not recognise the company. Models answer from what was in their training data. A product launched recently, or discussed rarely, simply isn't there. No amount of on-page optimisation changes this quickly: memory updates when models are retrained, which happens on the model vendor's schedule, not yours.
This is why "why doesn't ChatGPT mention us?" is often the wrong first question. The right one is: which of the sources that models learn from know about us yet?
Reason 2: the competitor exists in more places
Our AI Awareness audit for AutoChase that day found it on LinkedIn and in a Reddit discussion, and Perplexity could find its website. What it did not find: listings on independent review sites, a Wikidata entry, or Organization schema on the homepage connecting the brand to its profiles.
An established competitor usually has years of this: review-site listings, comparison articles, forum threads, directory entries. When a model or a search-backed assistant assembles a "best tools" answer, those independent sources are what it leans on. A vendor's own homepage is one voice; ten independent pages describing a competitor are a chorus.
Reason 3: the questions are category questions
Buyers rarely ask about you by name. They ask "what's the best way to chase unpaid invoices", and the answer is built from whichever brands the sources associate with that problem. If your site describes you with different words than your buyers use — "accounts receivable automation" when they say "invoice reminders" — you can be absent even from search-backed answers.
What actually moves it
In order of return on effort: make sure AI crawlers can read your site (/blog/robots-txt-front-door); add Organization schema with sameAs links to every official profile (/blog/organization-schema-sameas-for-ai); get listed and genuinely reviewed on two or more independent sites in your category; publish direct answers to the questions buyers ask, including honest comparison pages; and take part — openly, as yourself — in the discussions where buyers ask for recommendations.
Search-backed engines such as Perplexity and Google's AI Overviews can reflect these changes within days or weeks. Models answering from memory will lag until they are retrained.
Measure the change, not the mood
Because answers vary run to run, a single re-check after publishing proves nothing either way. Track the same questions over time and compare before and after on the prompt you targeted — that is the only way to know whether a fix worked (/blog/is-your-geo-working). Anovox does this automatically: mark a fix published and it compares that prompt's answers before and after, and tells you when an engine cites your new page.
See this on your own brand
Run the free check — same scoring the dashboard uses, no account needed.
Run free check