AI Mode
Google's AI-powered search experience — essentially Gemini with search, calculator, time, location and Python tools — answering queries conversationally.
Under the hood, Google’s new AI-powered search experience, known as AI Mode, is essentially the Gemini language model equipped with a powerful toolkit. This toolkit includes Google Search, a calculator, a Python sandbox, and real-time location and time data. Instead of just giving you a list of links, it uses these tools to answer your questions conversationally.
Testing reveals exactly how this process works. When you ask a question, the system first injects your current date, time, and location to keep the context accurate. Then, it translates your query into multiple background search calls to gather information. It follows a strict pattern: prepare, retrieve, signal, and then serve the final answer. If you disable the search tool, the system still runs those background searches, but it refuses to package them into a final response.
Interestingly, AI Mode handles web pages through its own browsing and content-fetching paths. At launch, it lacked the ability to read live pages directly through a URL, a feature the standard Gemini app can easily handle. For anyone looking to understand how search is evolving, AI Mode is now the primary battleground for visibility.
AI Mode is Google's AI-powered search experience, and under the hood it is essentially Gemini equipped with a set of tools: Google Search, a calculator, time, location, and a Python sandbox. It answers queries conversationally, using search results as grounding rather than just returning a list of links.
Our testing exposed how it works: date, time and location are injected as real-time context to sharpen time- and place-sensitive answers, and an internal tool_code block shows the search queries it constructs — rewriting a question into several google_search.search(...) calls before synthesising a reply. When we disabled Google Search, it still ran the searches but Gemini declined to synthesise them into an answer.
Its retrieval follows a Prepare → Retrieve → Signal → Serve pattern, and it reads pages through distinct browsing and content-fetcher paths. Notably, at launch AI Mode could not read live pages via the URL Context tool the way the Gemini app could. It's a prime surface for AI visibility work.