When a buyer asks ChatGPT or Perplexity which tool to use, the answer names a few brands. Yuzu asks the same questions every day, records whether your brand is one of them, and keeps the sentence each answer gave.
Free during early access, without a card.
“best tools to track brand mentions in AI answers”
The whole answer is kept, not just the verdict, with the sources it grounded on.
“It is the more thorough option, though it costs more than the lightweight trackers.”
Two passes said positive, one said negative.
Reported as a split. A borderline answer is a finding in itself, so the panel is never rounded up to a verdict.
A question with no incumbent needs one good page, not a campaign.
Swipe for the rest.
Search Console traffic and answer-engine visibility sit on the same page, because your buyers use both. Every number on it links to the run that produced it.
Impressions, clicks and index status from Search Console, next to what ChatGPT and Perplexity said today. Clicks and impressions get their own scales, because two clicks a day drawn against two hundred impressions would lie flat on a shared chart.
A product demo. The figures illustrate an account a few months in; no customer's account looks exactly like this.
You write no prompt list and import no keyword research. Yuzu reads your public homepage, works out what you sell, and derives eight questions your buyers ask. A question that names your brand is dropped before it is asked, because it would guarantee a mention and measure nothing.
A competitor list typed in by hand misses whoever actually wins the answers. Yuzu extracts the field from the answers themselves, so the brands on your chart are the ones the engines named. The engines and their crawlers are excluded from it.
Where other tools give you a sentiment number, Yuzu gives you the sentence: which model said it, on which day, answering which question. A judgement that cannot quote text present in the answer is discarded.
Each reading is judged several times with the options in a different order, and the agreement is reported. A split panel shows as a split, because a borderline answer is a finding in itself.
Across your buyers' questions, Yuzu tracks which brands the engines name. The useful output is the gaps: questions where no brand is named, and questions held by a single incumbent.
Search Console shows what Google shows; live probes show what the answer engines say. Both are on one page, with impressions and clicks on their own axes so a small site's clicks stay visible.
A daily self-check runs every lesson this engine has learned as a rule: tracked pages that no longer exist, canonicals pointing elsewhere, publishing that stopped, a probe running without a key. It tells you before you notice.
Add an OpenAI or Perplexity key and the probes run on your account. Add an Anthropic, Google or xAI key and Claude, Gemini and Grok are measured too. Keys are stored encrypted and never leave the server.
Anyone can query a model and print a percentage. The question is whether the number survives someone asking where it came from. Every figure Yuzu shows traces back to the answer that produced it, on the date it was given, by the engine that gave it.
Yuzu is free during early access, with no card on file. These are the plans for paid use after it, and nothing converts to a paid plan unless you choose one.
One site, eight buyer questions, ChatGPT and Perplexity, measured every day.
Up to three sites, and more engines with your own keys.
Your own keys are stored encrypted and never leave the server. You pay your provider directly for what the probes use.
We would like to measure page views and product usage with Google Analytics and Amplitude, which set cookies. Neither is needed for the site to work, so they only run if you say yes.
What we collect