Google has opened a developer preview for third-party apps to work inside AI Mode in Search and the Gemini app. Developers can connect through a remote MCP server or, for Android Gemini users, Android’s AppFunctions route, while Google reviews and tests each app before it goes live.
Watch Desk analysis
What happened
The preview gives developers a formal route to let Google’s AI products hand tasks to connected services. Remote MCP servers can reach Search and Gemini users on the web, Android and iOS; AppFunctions is limited to Gemini on Android. Developers register through Google Connected Console, choose an app category and identify at least three capabilities for users to consider when connecting it.
Google’s policy bars misleading or overly broad capability descriptions intended to steer the model towards one app. Apps may be rejected, suspended or removed for breaking the rules, and developers may lose their accounts. Search Engine Journal reports that connected apps are currently available in AI Mode in the US, in English, and that Google says more than 50 apps have launched across Search and Gemini.
Why it matters
This is a bid to make AI assistants useful beyond answering questions: they can route a task to a booking, shopping or other service. That makes the app-selection rules consequential. A service’s description is not just marketing copy if it can influence which tool the model chooses.
The preview also gives developers a concrete integration path, but access is not a free-for-all. Google requires review and testing, including test accounts and sample data for apps that require sign-in. If Google cannot fully test an app, it says it may reject it.
Our read
The useful promise is an assistant that can take action across services; the useful safeguard is a rule against gaming the model with inflated claims. Whether that balance holds will depend on how consistently Google reviews competing apps, and how much control people retain over which service handles a request. “The AI picked it” is not much comfort if the user cannot tell why.
What to watch
- Which developers and categories Google admits as the preview expands.
- Whether users can clearly choose or change the app handling a task.
- How Google applies its rules when app descriptions compete for the model’s attention.
- Whether the planned expansion to more markets and Google products arrives over the coming months.
Discussion spark: Should Google decide which connected app handles a request, or should users have to choose one themselves?
Sources and evidence
- Source update (31 October 2026, 07:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.