Getting reliable local answers from an AI has been tricky. You had to stitch a model to a maps API, manage rate limits, and still watch for errors. Gemini API Grounding with Google Maps makes that simpler. It lets Gemini use fresh Maps data on more than 250 million places to answer location-aware prompts, with citations and an optional widget for context. Below, we recap what’s new, pricing, supported models, how to enable it, and a few concrete use cases.
What “Grounding with Google Maps” is
Grounding connects a model’s output to verifiable data. In this case, Gemini can call Google Maps as a tool when a query has geographic context, then return an answer based on Maps data like places, addresses, hours, and reviews. This improves factual accuracy for prompts such as “best cafés near me,” “open pharmacies within 2 km,” or “kid-friendly museums in Boston.”
Why grounding matters for factual answers
Grounding helps reduce made-up details by tethering responses to a trusted source. It also adds citations so users see where information came from. That’s especially useful for time-sensitive local details like opening times and ratings.
What data Gemini can pull from Maps
The service taps into the Google Maps ecosystem, which includes structured information on 250M+ places worldwide. That coverage is useful across travel, retail, food delivery, and local discovery.
What changed this week
Google announced general availability for Grounding with Google Maps in the Gemini API on October 17, 2025. The developer guide and examples went live, and AI Studio references were updated to highlight the feature.
AI Studio and docs now live
The Gemini API docs show the exact parameters, supported models, and example code for enabling the googleMaps tool, including how to pass user location and how to render the optional contextual widget token.
Pricing, quotas, and availability
Grounding with Google Maps uses a per-grounded-prompt charge. The docs list $25 per 1,000 grounded prompts, with free request-per-day buckets that vary by model and billing setup. In published examples for Flash and Flash-Lite, free tier shows 500 RPD, while paid projects show 1,500 RPD free before the $25/1K rate applies. Always check the live pricing table for your model.
How quotas vary by model
The pricing page breaks out rates by model family. Flash and Flash-Lite entries explicitly list Maps grounding RPD buckets alongside the $25/1K grounded prompts rate. Numbers can differ across models and may change, so confirm for your chosen model before launch.
Where it’s supported and any territory limits
Docs note prohibited territories you must not target. Outside those, coverage is broadly available. If you serve global users, add territory checks to your app to stay compliant.
Supported models and current limits
Supported models include Gemini 2.5 Pro, 2.5 Flash, 2.5 Flash-Lite, and 2.0 Flash. Grounding is off by default, and the current scope focuses on text plus an optional Google Maps widget context token that you can render in your UI.
Known limitations
The docs call out that you must enable the tool per request, that you should pass location when known, and that you must display Maps sources when grounding is used.
How to enable it in your app
At a high level, you add a googleMaps object in the tools parameter of your generateContent call. Optionally provide latitude and longitude for better relevance. The response can include groundingMetadata with Maps sources and, if enabled, a google_maps_widget_context_token you can render.
Passing user lat-long
You can send a lat-long in the retrieval config, which helps the model retrieve hyper-local results, for example “within a 15-minute walk.”
Showing sources and attribution
You must display the Maps sources returned in groundingMetadata and follow Google’s text attribution rules when you present them. The docs include UI guidance and even CSS.
Fast examples
Trip planner for New York
Prompt: “Plan a Saturday art crawl in Chelsea with 3 galleries within 1 km, open after 11 am, and a late lunch nearby.” Grounding returns places with hours and links so you can map the route and show citations in-app.
Food delivery radius for Mumbai
Prompt: “List top-rated vegetarian restaurants delivering to Lower Parel within 30 minutes.” Use location context and display the widget token to let users browse sources and reviews.
Performance and cost tips
Only turn Maps grounding on for geo-intents. Keep it off for generic questions to save latency and cost. Monitor your P95 and confirm RPD usage in your billing dashboard.
Compliance checklist
- Show Maps sources with links as returned by
groundingChunks. - Keep the “Google Maps” text attribution intact and styled as documented.
- Respect the prohibited territories list.
- You may cache specific IDs and the widget context token as allowed by the docs.
Why this matters for dev teams
Teams no longer need to wire a generic model to a separate maps stack, wrangle multiple quotas, and hand-roll citations. For many local tasks, you can stay inside the Gemini API and ship faster with fewer moving parts.
Next steps
- Skim the announcement, then open the docs and try a prompt in AI Studio.
- Start with a single geo-intent, measure CTR on source links, and expand.
Frequently Asked Questions (FAQs)
How do I turn on Google Maps grounding in the Gemini API?
Add {"googleMaps": {}} to the tools array in your generateContent call. Optionally pass latitude and longitude for context.
What’s included in the response besides text?
You can receive groundingMetadata with Maps sources and, if enabled, a google_maps_widget_context_token to render a contextual widget.
How accurate are opening hours and ratings?
They come from Google Maps’ live data, which updates regularly and covers 250M+ places worldwide. Always present sources so users can verify.
What are the current prices and free tiers?
Docs show $25 per 1,000 grounded prompts. Free RPD buckets vary by model, with examples showing 500 RPD free on some models, and paid projects showing 1,500 RPD free. Check the live table.
Which models support it?
Gemini 2.5 Pro, 2.5 Flash, 2.5 Flash-Lite, and 2.0 Flash are listed as supported.
Is the feature global?
Docs list prohibited territories you must not target. Outside those, it’s broadly available.
Do I need AI Studio for this?
No, but AI Studio is a convenient place to test prompts before wiring code. Studio was updated alongside the announcement.
What about Search grounding vs Maps grounding?
Search grounding helps with web facts and citations, while Maps grounding is best for geo-specific answers. You can use both when needed.
Optional comparison table
| Use case | Maps Grounding | Search Grounding |
|---|---|---|
| Local POI answers | Best choice, returns places and widget token | Useful but less specific for local POIs |
| Citations | Maps links from groundingMetadata | Web URLs with citations |
| Pricing unit | Per grounded prompt, $25/1K | Per grounded prompt, $35/1K |
| Ideal prompts | Nearby, open hours, routes, venues | Facts, news, reference lookups |
