See what AI says about your brand

When your buyers ask ChatGPT, Perplexity or Google for a recommendation, some brand gets named. Yuzu checks whether it is yours, every day, and shows you the sentence it came from.

14-day trial. Card required, cancel in one click.

The dashboard

One screen: what changed, and what to do about it

Search traffic and answer-engine visibility on the same page, because they are the same question asked twice. Nothing on it is a vanity metric; every number links to the run that produced it.

A live view of Yuzu watching three sites. The numbers are real measurements, not a mock.

Every score quotes its source

Other tools give you a sentiment number. Yuzu gives you the sentence: which model said it, on which day, answering which question. If a judgement cannot quote text that is actually in the answer, it is discarded rather than shown.

It tells you when it does not know

Each reading is judged several times with the options presented in a different order. Agreement is reported, not hidden. A split panel shows as a split, because a borderline answer is a real finding and rounding it up would be a lie.

It finds the questions nobody owns

Across every question your buyers ask, Yuzu tracks which brands the engines name. The valuable output is the gaps: questions where no brand is named at all, and questions held by a single incumbent. That is the cheapest visibility on the board.

Search and answer engines, one loop

Search Console for what Google shows, live probes for what the answer engines say. Impressions and clicks on their own axes, because two clicks a day against two hundred impressions would otherwise lie flat on the floor of a shared chart.

It checks itself

A daily self-check runs every lesson this engine has learned as a rule: pages that are tracked but do not exist, canonicals pointing elsewhere, publishing that quietly stopped, a probe running without a key. It tells you before you notice.

Bring your own model

Supply an OpenAI, Anthropic or Perplexity key and the drafting and probing run on your account at cost. Or use ours and pay the flat price. Keys are stored encrypted and never leave the server.

How it measures

Built by someone who kept getting it wrong

Yuzu exists because its own dashboard kept lying to me. It reported a page as waiting for Google that did not exist. It asked me to add an API key I had already added. Its brand matcher was hardcoded to one site, so every other site read zero. Each of those is now a rule that runs daily.

  • A tracked page that serves a not-found body is a critical alert, not a page that is still indexing.
  • A credential is judged by what the loop actually did with it, never by what the dashboard's own machine can see.
  • Evidence carries a date. A reading from a run three days old says so instead of pretending to be current.
Pricing

Priced to try, not to negotiate

No sales call, no annual lock-in, no seat maths. The trial needs a card so you never get an invoice you did not expect, and cancelling takes one click.

Starter

€29per month

One site, 25 buyer questions, two answer engines, measured weekly.

  • Search Console + answer engines
  • Auditable sentiment with quoted evidence
  • Daily self-check
  • Email when something breaks
Start free

Pro

Most useful
€79per month

Three sites, 100 questions, every engine, measured daily.

  • Everything in Starter
  • Daily measurement and the prompt-to-brand index
  • Competitor tracking and share of answers
  • Content briefs from the gaps it finds
  • Bring your own model keys
Start free

Bring your own OpenAI, Anthropic or Perplexity key and take €10 off Starter or €30 off Pro. You pay your provider directly for what you use.

FAQ

The questions people actually ask

Is this SEO or AI visibility?

Both, and they are one job. Yuzu reads Google Search Console for impressions, clicks, positions and index status, and separately asks the answer engines your buyers' actual questions. Tools that do only the second half cannot tell you whether the page that got cited is also ranking.

How can sentiment be accurate?

It cannot, and anyone claiming otherwise is selling you a number. Published agreement between language models and human coders sits around 0.58 on Cohen's kappa, and hosted models are not deterministic. So Yuzu does not sell accuracy. It sells a reading you can check: the quoted sentence, the model, the date, and how many passes agreed.

Do I need my own API keys?

No. The flat price includes the measurement. If you would rather run it on your own account, add a key and take the discount. A Claude, ChatGPT or Perplexity consumer subscription cannot be used for this — those plans are licensed for their own apps, not for third-party tools, so it has to be an API key.

What happens when the trial ends?

The card on file is charged for the plan you picked, and you keep everything measured so far. Cancel before the fourteenth day and you are charged nothing. There is no notice period and no annual commitment.

Where is the data?

Hosted in the EU. Yuzu reads your Search Console through Google's own OAuth, which you can revoke at any time from your Google account, and it never asks for write access to your site.

Who is behind it?

One engineer in Germany who built it for his own products first and still runs it against them daily. If something is broken you will get a reply from the person who can fix it.

Find out what the engines say about you

The first measurement takes a few minutes. If you are invisible for every question your buyers ask, you will know today rather than next quarter.

Start free

14 days. Card required. One click to cancel.