Google brings Gemini to Maps and Earth

Ask any chatbot for a quiet dinner spot nearby, and there is a decent chance it sends you to a bistro that went out of business two years ago. For spatial and location tech, hallucinations are not just annoying; they are a non-starter.

Google just announced a series of generative AI updates for Google Maps Platform and Google Earth, designed to tether large language models directly to real-world geospatial data.

Grounding LLMs with 250 million places

Over 10 million websites and apps already rely on Google Maps Platform, from food delivery services to real estate platforms. But as developers try building AI agents and chat assistants into their apps, keeping place info fresh has been a major pain point.

To solve this, Google introduced "grounding with Google Maps." The feature lets developers tie their LLM-powered experiences directly to Google’s live database of over 250 million places. The goal is to make AI responses significantly more factual, cutting down on hallucinated addresses or outdated opening hours.

We are already seeing real-world adoption on the road. Following the preview of Gemini features in the Places API at Google I/O earlier this year, EV maker Rivian will start rolling out AI-generated summaries of restaurants, shops, and supermarkets directly onto its vehicle infotainment screens next month.

Google Earth gets multi-step spatial reasoning

Google is also rolling out Gemini capabilities to Google Earth, specifically targeting urban planners and researchers who deal with massive city-level datasets. Google claims it can shrink data analysis workflows from days to minutes.

In one scenario shared by Google, a transportation planner trying to install electric vehicle chargers can simply type a natural-language prompt: "Can you map the five zip codes with the fewest EV chargers relative to their geographic area size?"

Google Earth then uses multi-step reasoning to filter the areas, run the spatial calculations, and generate a custom visual map. The user can follow up by asking which nearby hotels or shopping centers lack chargers within a certain radius. Built alongside Google Research and Alphabet's moonshot factory X, the capability will begin pilot testing with trusted users next month.

Why it matters

We think this is one of Google's smartest enterprise AI moves this year: turning its most unassailable real-world moat into essential plumbing for the AI era.

Raw LLM reasoning is becoming commoditized, but real-time physical-world data is nearly impossible to duplicate. By allowing developers to ground models with 250 million continuously updated places, Google ensures that Maps remains the default backbone for AI-driven local commerce, travel, and logistics.

For Google Earth, introducing natural-language spatial reasoning fundamentally lowers the barrier for geographic information systems (GIS). Complex queries that previously required specialized querying and data wrangling can now be done in plain English. However, we think the stakes are high: unlike a casual chatbot slip-up, errors in infrastructure planning can be costly, meaning Google’s multi-step reasoning will have to prove practically bulletproof during next month's pilot.