YuzuYuzu
← All writing
Kyle Nelson7 min read

Track Brand Mentions in ChatGPT and Perplexity (with Evidence)

Track brand mentions in ChatGPT and Perplexity by whether your brand gets named for recommendation questions, the exact sentence each reading came from, and whether the answer cites your site, not by vibes. Yuzu is built for exactly that, and it also pairs the AI reads with Google Search Console data.

What to track when buyers ask for recommendations

You’re trying to show that when someone asks an answer engine for a recommendation, the engine names your brand and, often missed, cites your site.

Yuzu measures four things together for each reading:

  1. Whether buyer questions in your category get your brand named in AI answers.
  2. The ranked field, meaning which brands the engines recommend instead.
  3. Citations, meaning when an answer links your site, not just names it.
  4. Search visibility alongside, from Google Search Console.

That combination gives you something you can inspect. You can check the sentence and see whether the answer points to your site. Yuzu’s measurement is designed so every reading can be checked directly.

If you’re deciding what to measure, use the posts that match the exact outputs you need:

How to track AI brand mentions without fooling yourself

The failure mode is usually the same. A team sees a number, or one screenshot, assumes it reflects reality, and starts acting on it. Then nothing moves because the thing being measured never matched the questions buyers actually ask.

The workflow that prevents that is simple in principle: sentence-level checks.

Yuzu derives the questions your buyers ask from your own homepage, puts them to the answer engines, and records who was named: you or the ranked field of competitors named instead. Every number links to the answer text it was read from, so you can verify the exact sentence the engine produced.

That detail matters because AI recommendations can change phrase by phrase. If a tool says you were mentioned, you still need to know how. A line that says you are an alternative is different from a line that recommends you outright, and different again from a passing mention in a non-recommendation answer.

If you want a practical way to build this discipline yourself, start with:

ChatGPT vs Perplexity: track what each one names

A lot of teams assume ChatGPT and Perplexity will behave similarly for brand recommendations. Sometimes they do. Often they don’t, and the sensible way to handle that is to track what each one actually names.

Yuzu explicitly measures both ChatGPT and Perplexity naming outcomes when buyers ask for recommendations. It quotes the sentence each reading came from and records the ranked field, meaning which brands the engines recommend instead. It also tracks citations, meaning whether the answer links your site.

Here’s the comparison I’d use for decision-making, because it forces you to confirm what your team is actually optimizing:

| What you want to know | What to check in ChatGPT and Perplexity | What Yuzu records | |---|---|---| | Are we named for recommendation questions? | The exact reading for a recommendation request | Whether your brand gets named | | Are we top relative to competitors? | The ranked field of what the engine recommends instead | Which brands get recommended instead | | Does the answer cite us, not just mention us? | Whether the answer links your site | Citation presence and the reading sentence | | Is this movement reflected in web demand? | Google Search Console trends like impressions and clicks | Google Search Console visibility alongside |

If you need examples of what “actually named” looks like, use:

Google Search Console pairing: connect AI reads with search data

If your AI naming doesn’t move, don’t immediately blame the models. First check whether the underlying signals you control are shifting.

Yuzu puts Search visibility alongside, from Google Search Console. That means you can compare AI naming and citation outcomes with what’s happening in impressions and clicks. If you’ve ever looked at a drop in AI mentions on Tuesday and then opened GSC to see whether branded demand or page visibility also changed, this is the pairing you need.

If you’re looking at GSC day-to-day and don’t know which signals to trust, start with:

The point is that GSC gives you a reality check. Pairing it with AI outputs keeps you from guessing why an answer changed.

Where llms.txt fits (and where it doesn’t)

Teams love a single technical lever because it’s clean. But even if a file helps, you still need to verify what the answer engine actually does.

The Yuzu guidance is explicit: llms.txt can help signal sources for “where to look,” but it doesn’t replace sentence-level checking. Yuzu measures whether ChatGPT, Perplexity, and Google actually name your brand and quote the sentence they used.

So when you track brand mentions, treat llms.txt as an input signal, not your measurement. Your measurement is the sentence-level output. In practice, that means you might publish the file on Monday and still need to read the answer text later in the week to see what changed.

For that specific topic, read:

A practical checklist for tracking brand mentions in chatgpt and perplexity

Here’s how I’d set this up so you end up with evidence you can show and explain.

  1. Pick recommendation questions derived from your category demand. Yuzu derives buyer questions from your own homepage.
  2. Run a sentence-level check for each answer engine output. Yuzu quotes the sentence each reading came from.
  3. Record who was named. Yuzu records whether it was you or the ranked field of competitors named instead.
  4. Record whether there was a citation. Yuzu tracks citations, meaning when an answer links your site.
  5. Compare the AI outcomes with Google Search Console trends. Yuzu shows Search visibility alongside from GSC.
  6. If ChatGPT doesn’t mention you, verify what it outputs before you change anything else. Start by checking what it says sentence by sentence and then publish what the answers are missing.

If you want a more structured workflow for “what to do when it fails,” use:

What evidence looks like when it’s actually measurable

The evidence you want has three properties:

  • It names your brand when asked for a recommendation.
  • It includes the exact sentence the engine used.
  • It shows whether the answer cited your site.

Yuzu’s output model is built around those properties, with daily measurement and sentence-level quotes. That means you can show the reading itself, not only a summary on a dashboard. If someone on your team asks why a result was counted, you can open the source sentence and inspect it.

If you can’t do that, your team is stuck in the same loop: you see names without verification, citations without context, or metrics without the reading text. Yuzu avoids that loop, because every number links to the answer text it was read from, so a reading can be checked directly.

Sources