Google Opens Connected Apps Preview for AI Mode and Gemini

Google has opened the Connected Apps Platform preview to developers. Integrations that pass review will be able to work in Search’s AI Mode and the Gemini app.

· 4 min read · 4 comments

Google has opened a public preview of the Connected Apps Platform: developers can apply to make their apps available in Search’s AI Mode and the Gemini app. The platform was announced on October 28; Google Search Vice President Robby Stein announced the preview’s availability on X. As Search Engine Journal reports, Google reviews every integration before it goes live.

How to Connect an App

There are two ways to integrate. A remote MCP server—using the protocol that connects AI models to external tools—can reach Gemini and Search users on the web, Android, and iOS. Google calls the second option, AppFunctions, Android MCP; it is designed for Gemini users on Android. The technology a developer chooses determines where the app can run.

Developers apply through the Google Connected Console. During registration, they choose an app category, which sets the minimum requirements for what the app can do. The category cannot be changed later. They must also specify at least three capabilities to show users when they decide whether to connect the app.

In the console, developers select the countries, languages, and Google products where the app should be available. Google then reviews and tests the integration. If the app requires users to sign in, the developer must provide test accounts with sample data. An app that cannot be fully tested may be rejected.

How an App Is Chosen for a Task

Connecting to the platform does not mean the model will send every relevant request to the app. According to Google’s guidance, server instructions help the model understand which tasks to assign to it. Capabilities therefore need to be described clearly enough for a request to reach the right tool.

The rules prohibit overly broad or false descriptions intended to influence the model’s choice. An app must not set rules for using other tools or refer to them. Integrations cannot contain advertising or promotional material for other companies, though information about the app’s own brand is allowed. For a violation, Google may reject the app, suspend access to it, remove it from the platform, or suspend or close the developer’s account.

Users have a choice, too. In AI Mode, the system may default to an app the person used for a previous task. Users can name another app in their request or select one from a list after a general search query.

Connected apps currently work in Search’s AI Mode in the US and in English. The apps listed in Google’s documentation include Instacart, Canva, and YouTube Music. More than 50 such apps have launched across Search and Gemini. After publication, developers will be able to see traffic, latency, and error rates.

What This Means for Web Service Teams

Developers now have a documented application process and review rules. A team can choose an integration method, describe what its product can actually do, and prepare for testing. The difference between channels matters: an MCP server covers more platforms than AppFunctions, while connected apps in Search’s AI Mode are currently limited to the US and English.

For marketers, a description of capabilities is not an advertising claim without consequences. It helps the model decide whether an app fits the task and must not mislead it. Product teams would do well to check three things against each other: what the server instructions claim, what people see when connecting the app, and what the service can actually do.

Access will expand gradually. Google plans to add developers, markets, and products over the coming months. In the coming weeks, partner apps already available in Gemini are expected to appear in Search; a Calm integration for both products has also been announced.

My View: Start with Clear Capabilities

I wouldn’t treat the platform as a place where finding the right words is enough to make the model choose a service more often. First, it’s worth defining which task the app can perform reliably and how people will understand what they’re being asked to connect. I see the ban on misleading descriptions not as an obstacle to promotion, but as a useful test of product clarity.

What happens after someone chooses an app matters to me, too. If people can switch to another app and the system can suggest one they’ve used before, the team needs to think through both transitions: what action is expected, what the app can access, and what the outcome will be for the user. A technical connection alone does not remove the responsibility to make the interface and the service’s behavior clear.

Sources

Where the news comes from. The text is a retelling in the author’s own words; the facts come from the source, the opinion is the author’s.

  1. Google Opens Connected Apps Preview For AI Mode & Gemini via @sejournal, @MattGSouthern www.searchenginejournal.com

About the author

Katrin Laas

UX and Design Systems Expert

I help product teams build interfaces that can evolve without endless exceptions. When training designers, I usually start with accessibility and component behavior, not with choosing a trendy tool.

All posts by the author

Leave a comment

Not published. We’ll send a link to confirm your comment.

No links or ads. Your first comment is published after review. By sending a comment you agree to the privacy notice.

  1. When we’ve designed service flows that depend on external tools, the handoff is where users most often lose confidence: they need to know what the app can do and what happens next. For a Connected Apps integration, I’d make those capabilities explicit in both the app’s onboarding and its interface, then test the flow on a real phone—especially for Android AppFunctions.

  2. What criteria will Google use to decide which reviewed app gets selected for a task? Passing integration review is one thing; users need to understand why Gemini chose a particular service and what it can actually do.

    1. Katrin Laas Post author

      That distinction matters: the post describes Google reviewing integrations, but it doesn’t specify how Gemini will rank eligible apps for a particular task. I’d look for clear capability descriptions, task fit, and an explanation users can inspect when an app is selected; review alone doesn’t tell users why that choice makes sense.

    2. On a small frontend project, I found it easier to explain an integration by listing the exact actions it supports; that would also help users see why Gemini picked it for a task.