AI Models Gumshoe Tracks
Brand visibility varies across AI models because each one uses different training data, retrieval methods and ways of reading a question. Gumshoe tracks 12 model families across eight providers and reports visibility for each, so you can see exactly where your brand appears and where it does not.
Published Nov 2026
Gumshoe tracks 12 AI model families across eight providers
Gumshoe measures brand visibility across 12 model families, including ChatGPT and ChatGPT mini from OpenAI, Claude Sonnet and Claude Opus from Anthropic, Gemini Flash and Gemini Pro from Google, Google Search AI Overviews, Perplexity Sonar, Grok from xAI, and DeepSeek Chat and DeepSeek Reasoner. The FAQ lists every family. Each Visibility Audit on a paid plan runs six of them across 8 buyer personas and 480+ prompts.
Google appears twice on purpose. Google AI Overviews is a separate engine from Gemini and answers differently, so treating them as one source would hide real differences in how each surfaces brand content.
This page names model families rather than versions, because the version behind each family changes over time as providers release new ones.
| Where the count applies | Models |
|---|---|
| Model families you can choose from | 12, from 8 providers |
| Models in each Visibility Audit on a paid plan | 6, chosen by you |
| Models in the free sample audit | 4 |
| Models in each Pulse Audit | 4, across 80 conversations |
Some models look things up, and some answer from memory
A key distinction in the tracked set is whether Gumshoe runs a model with live web search. A model that searches, such as Perplexity Sonar, retrieves live pages before answering and returns the sources it used. A model without web search answers from its training data and may reflect content from weeks or months earlier. That difference affects which brands appear and why.
ChatGPT, Gemini, Google AI Overviews and Perplexity Sonar run with live web search. Claude, Grok and DeepSeek answer from training data alone, and the Playbook adds opt-in web search versions of Claude Sonnet and Claude Opus. The model coverage table on the Visibility Audit page shows each family, and the tooltip for each family in your project settings shows the same.
Per-model reporting shows where brands win and where they do not
Gumshoe reports visibility for each model rather than blending results into one average. The Model Visibility section of the Visibility Audit dashboard shows how often your brand is mentioned by each model. Those numbers differ for real reasons: training data, source preferences, how a model reads the question, and how fresh its content is.
A blended score would hide that variation. Gumshoe's study of more than 600,000 AI answers and 1,145 brands found that average brand visibility is 24.8%, with a standard deviation of 26.8 points, so a single figure can describe brands with very different footprints. See What the Average Hides.
The audit's CSV export includes a column for each model with your brand's rank in that answer, so analysts can see row by row where a brand placed in a given model's answer to a specific persona and prompt. The Model × Competitor heatmap takes this further and compares model-level performance across the competitors you track, up to 30 per project.
AI answers are not deterministic, and persona-level slices need more data before they drive a decision. How Gumshoe Works explains how the audit handles both.
You choose which models run in each audit
All 12 families are available on either paid plan, and you choose which six run, without setting up a new audit. On the project's settings page, check or uncheck families under Model families and save. The next scheduled Visibility Audit picks up the change.
Not every brand needs every model. A B2B software company may put the models used in research workflows first, while a consumer brand may weight the models tied to purchase questions. Choosing the right set keeps the audit focused on the models that reach your buyers.
Between full audits, the Pulse Audit runs weekly on Build your own and daily on the Playbook. It uses 4 models across 80 conversations and is a check-in between the monthly Visibility Audits.
Model choice and persona choice are not interchangeable. In Gumshoe's research across 7,475 buyer personas and 13 tested models, every model shifted its answers by 21 to 40 points depending on who was asking, per What the Average Hides. Outside research points the same way: Persona Conditioning of Brand Recommendations in Retrieval-Augmented Commercial Chat found that buyer persona characteristics change which brands AI systems recommend.
What this means for your audit
Tracking a single model, or one average score, leaves gaps. Because Gumshoe offers 12 model families, reports results for each model, and lets you choose the set for each project, you get a precise picture of where your brand appears across the systems your buyers use.
- Gumshoe tracks 12 model families across 8 providers, including ChatGPT, Claude, Gemini, Google AI Overviews, Perplexity Sonar, Grok and DeepSeek.
- Six families run in each Visibility Audit, and you choose which ones under Model families in project settings.
- Each family's tooltip in project settings shows whether it runs with live web search or answers from training data, which helps explain different visibility outcomes.
- Visibility is reported per model, and the Model × Competitor heatmap shows which models favor which brands.
- Gumshoe adds new model families as they become relevant to how buyers search.
See how each AI model talks about your brand
Run a free sample audit across 4 AI models, no credit card needed.
Free to start · No credit card required