When you run a Visibility Audit (formerly known as a report), some of the AI models search the live web before they answer and others answer from what they learned during training. Both matter for understanding your visibility, and Gumshoe tells you which is which before you run anything.
Does a model decide for itself whether to search?
Not inside a Gumshoe audit. Every model option in Gumshoe is set to one mode or the other, and that setting is fixed. A model option with live web search on searches before answering every prompt in your audit. A model option with it off never searches, on any prompt.
That is different from using an AI model yourself in a chat window, where the model may search for one question and not the next. Fixing the mode is deliberate: it means each model option measures one thing consistently, so your results stay comparable from run to run.
How do I check which mode a model uses?
Open your project settings and find Model families under Audit settings. Each model has a help icon next to it. The tooltip gives you an overview of the model, who typically uses it, and a Live Web Search line that reads Yes or No.
That line is the answer, and it is the one place worth checking, because the model list changes as providers release new versions and as Gumshoe adds support for them.
You may also see the same model listed twice, once on its own and once labeled (Web Search). Those are two separate options you can select independently, which lets you compare what one model says with and without a live lookup. Which options you can select depends on your subscription.
What does live search mode show me?
- Fresh, real-world web content, including your newest pages and press mentions.
- How AI models are citing live domains and categories.
- Which domains are driving influence right now.
Live search answers also return the source URLs the model used, so they are what populate your Citations.
What does foundational training mode show me?
- How AI models recommend brands by default, without checking the live web.
- The underlying narratives and associations baked into each model's training.
- How your brand appears in default responses, which still power a large share of what people see.
These answers do not return source URLs, so they contribute to your visibility results but not to your Citations.
Why should I track both?
- Looking only at live search results shows you recency, but misses how people experience AI search when a model answers from training instead.
- Looking only at training-based results shows default reasoning, but misses how fresh content and citations can shift an answer.
- Running both gives you a balanced picture: static brand perception from training, and dynamic influence from live search.
It also means you are not caught off guard when a model's training is refreshed, because you already know what that model says in both modes.
Pro tip: Foundational training changes slowly, while live search results can shift day to day. If you have just published something new, the live search models are where you will see it first.