We track responses from most major AI models. We are the only GEO platform that can isolate between base model knowledge and search enhanced knowledge by tracking responses through the model's API and its consumer app.
AI models try first to answer prompts using its internal knowledge (base model). If the AI model does not inherently know how to respond to your prompt, it will retrieve responses from live search.
Here's how we name the different AI models in our platform.
Company | Foundational Knowledge Name | Training Cutoff Date | Foundational Knowledge Model Version | Release Date | ConsumerApp Name |
Gemini | January 2025 | Gemini 2.5 | June 2025 | Gemini Search | |
OpenAI | ChatGPT | August 2025 | ChatGPT-5.4-mini | March 2026 | ChatGPT Search |
Anthropic | Claude | January 2026* | Claude Sonnet 4.6 | February 2026 |
|
Meta | Llama | August 2024 | Llama 4 | April 2025 |
|
Perplexity | N/A |
| N/A |
| Perplexity |
DeepSeek | DeepSeek | Not published officially | DeepSeek v3.2 | December 2025 |
|
Microsoft | N/A |
| N/A |
| CoPilot |
*According to Anthropic, Claude's training data cutoff is January 2026, but the reliable knowledge cutoff is August 2025.
Definitions
Base Model/Foundational Knowledge: This is the isolated foundational knowledge, accessible only via direct API integration with the LLM provider. This represents the information that the model was trained on.
Consumer App: What you see when you go to chatgpt.com or the ChatGPT mobile app. When users type in a prompt, the AI model either generates the answer from its base model knowledge (if the model has a base model) or it retrieves the answer from live search. This process is called RAG (retrieval-augmented generation). Learn more about RAG here.
What about Google AI Overviews and AI Mode?
AI Mode and AI Overviews are not stand alone LLMs since they live within Google Search. These LLMs combine traditional Google search infrastructure with Gemini capabilities. We track insights from both Google AI Overviews and Google AI Mode.
Our Methodology
Most competitors collect data from the consumer-facing app as their only methodology. So much like SEO tools of old, they pull insights from the consumer UI.
We do this too, but we also plug into the model maker API's as:
Clean, Reliable Data Collection: We pay the model providers for access to connect directly through their API. This provides us scaled access ensuring clean, reliable, repeatable data collection.
Statistical Significance: Because we tap into the API, we can analyze prompt responses at scale, giving us statistically significant, reliable and robust metrics rather than small sample sizes.
Base Model Insights: The APIs are the only way to isolate the base model’s knowledge. There isn't a way to force this in the consumer app. This allows us to understand what AI's preferences are at its core. This is especially important as the consumer app and all agent experiences are built on top of the base models. So these agents inherit brand preferences from the base model.
Brand Building: API is a more stable measure of long-term AI brand building. It does not have access to live searches, so is less reactive to changes in search indexes. It shifts when there has been a true model update.
Learn more about the different AI models and model versions in our AI Model Release Tracker!
FAQs
What is AI Overview Trigger Rate?
AI Overview Trigger rate measures the number of prompts which return with AI Overviews being triggered and included in responses. AI Overview Response Rate is calculated by responses with AI Overviews / responses successfully returned.
Why don't you track the consumer app for Claude, Llama or DeepSeek?
Tracking Claude, Meta AI, and DeepSeek through their consumer apps would require going behind a login, which would violate the terms of service of all three platforms and puts user privacy at risk. Instead, we pay to access access the foundational knowledge via direct API access or open source models.
Why don't you track the foundational knowledge for Perplexity or CoPilot?
Microsoft Copilot is built on OpenAI's proprietary models
Perplexity's default model, Sonar, is Perplexity's own fine-tuned model built on Meta's open-source Llama. Pro and Max subscribers can manually select third-party models instead, with options varying by search mode.