Skip to main content
Trackbase runs each of your prompts on every AI model you track and scores the answers per model. Models are set once for your whole workspace, not per prompt. You choose them under Settings.

​
Available models

  • ChatGPT
  • Gemini
  • Claude
  • Perplexity
  • Copilot
  • Google AI Overview
  • Google AI Mode
  • Mistral
  • DeepSeek
  • Grok
  • Llama
Which of these you can turn on depends on your plan.

​
How to choose

  • Start where your customers are. For most markets that means ChatGPT and the Google surfaces first.
  • Add models deliberately. Every model you add answers every prompt every day, and each of those answers counts in your averages.
  • Expect differences. The same prompt gets different answers on different models, because they use different sources. That is the reason to track more than one.

​
Read results per model

A blended score hides where you win and lose. Use the model filter at the top of the dashboard to look at one model at a time. On the AI Visibility page, the Heatmap tab shows visibility per model for you and your competitors. When a number moves, look per model first.
Adding a model where you are rarely mentioned lowers your blended averages from the day it starts. Nothing changed on the models you already tracked. See why did my score drop?.

​
Fan-out queries

For ChatGPT and Gemini, Trackbase also records fan-out queries: the searches the model ran while composing its answer. You find them on the AI Visibility page, in the Fan-outs tab. They show which search terms your content needs to answer.
Last modified on November 12, 2026