# Gemini Screen Automation Not Working? Check Phones, Regions and Permissions

Google expanded Android task automation, but account eligibility and device-specific controls still determine what you can actually delegate.

By Paras Madan | 2026-11-11


Google expanded Gemini’s Android task automation to more countries on November 10, including India, Japan and the UK. If it does not appear on your phone, start with eligibility and rollout status. A more detailed prompt cannot enable a feature that your device or account does not support.

The [Gemini Intelligence announcement](https://blog.google/products-and-platforms/platforms/android/gemini-intelligence-november-features-devices/) bundles several capabilities together. Screen automation is the part that works through supported Android apps to carry out multi-step requests. It needs a different troubleshooting approach from a normal chat response or a missing calendar connection.

## Check the exact phone, account and region

Google’s [screen automation help page](https://support.google.com/gemini/answer/16940971) lists Pixel 10, 10 Pro and 10 Pro XL, plus Pixel 11, 11 Pro, 11 Pro XL and 11 Pro Fold. Supported Samsung devices include the Galaxy S26 series, Z Flip 8, Z Fold 8 and Tab S12 series.

The current country list is India, Japan, South Korea, the UK and the US, with Pixel 10 devices excluded in South Korea. Users must be adults using a personal Google account. Work, school and supervised accounts do not qualify. The beta supports selected apps in English, Japanese and Korean through Gemini on Android, with gradual availability.

Check the precise model name rather than treating every phone in a generation as interchangeable. Pixel 10 Pro Fold, for example, does not appear in that screen-automation list. A broader announcement about Gemini Intelligence on a device does not establish support for every individual feature.

If the account is managed by your employer, changing the wording of a request will not solve the restriction. Likewise, an eligible phone does not guarantee that the rollout has reached it yet. Google’s [November Pixel Drop](https://blog.google/products-and-platforms/devices/pixel/november-2026-pixel-drop/) says updates begin rolling out on November 10 and continue over the following weeks.

## Identify which kind of task Gemini is attempting

A useful diagnosis begins with the route, because similar requests can use different mechanisms. A Connected App may fulfil a request without Gemini navigating the Android interface. Screen automation is also separate from scheduled tasks.

Think of a failed grocery request as three questions. Did Gemini understand the intended order? Could it access the relevant app? Did the app reach a state where a purchase could be reviewed? A chat answer that lists groceries only addresses the first question.

This distinction prevents unnecessary changes to unrelated settings. Reconnecting a calendar service is unlikely to fix an unsupported shopping app, and repeatedly requesting a scheduled reminder does not demonstrate that screen automation is available. Keep the first task narrow enough that you can see which part failed.

## Give the task a clear stopping point

For an illustrative first request, use something like: “In my grocery app, prepare a cart with one loaf of wholemeal bread and one litre of milk. Show me the chosen items and total before any purchase. If either item is unavailable, ask me rather than choosing a substitute.”

This is a suggested prompt, not a tested result. Its useful feature is the boundary: a cart can be inspected before money changes hands. The same pattern works for other supported tasks by specifying the target, acceptable alternatives and the point where a human decision is needed.

Review any proposed plan before letting it proceed. If the assistant interprets “milk” as several cartons or chooses a different delivery address, correct that before execution rather than expecting the final summary to reveal every detail.

**A clear request reduces ambiguity; it does not guarantee correct execution.** Check the app’s actual state and confirmation details instead of treating the conversation as the only record of what happened.

## Separate access problems from control limitations

First-use permission prompts matter. When an app is blocked from screen automation, describe the problem as an access issue before treating it as a model failure. Review the relevant permission rather than broadly enabling unrelated access just to make the task work.

Also check what your particular device lets you do when intervention is needed. Google currently distinguishes stopping a task from taking control of it: Pixel 10 and Galaxy S26 devices support stopping, but not taking control. A missing takeover control can therefore be an expected limitation.

That matters when choosing a first workflow. Avoid starting with something that is likely to require complex manual intervention halfway through. A small, easily checked preparation task is a better way to understand the feature’s boundaries than an urgent booking with several dependent choices.

## Decide what success actually means

For a prepared order, success means the right items, quantities, destination and total are visible for review. For a booking, it means the relevant time, passenger or customer details and cancellation conditions have been checked. “The assistant finished” is too weak a completion criterion for either case.

Keep passwords and payment details out of the conversation, and handle sensitive steps through the service’s own interface. If you stop a task, inspect the app before starting again. Repeating a request without knowing the current state can create confusion about what is already prepared or committed.

The practical improvement here is less time spent moving between screens. It is useful when the request is bounded and the result is easy to inspect. Start there, and expand delegation only after you understand where your device, apps and own review process still need to do the work.

I’m Paras. I work at the intersection of AI, engineering and distribution.

I help technical teams turn complicated product behaviour into clear guides and launch stories people can actually use.

[Book a 30-minute call](https://cal.com/paras-madan/30min).