Android Contextual Suggestions Rolling Out — On-Device AI
One day after Google unveiled Gemini Intelligence at The Android Show and redefined Android as an intelligence system, a quieter and arguably more privacy-forward AI feature began appearing on Pixel 10 devices. Google is using on-device AI to analyze how you use your Android phone and provide “Contextual suggestions.” As of today, we’re seeing this setting rolling out on the Pixel 10 series, including the Pixel 10a, running Android 16 with stable version 26.18 of Google Play services.
The timing is not accidental. Contextual Suggestions is the on-device behavioural learning layer that feeds the broader Gemini Intelligence vision — but it operates in a fundamentally different way from the cloud-assisted features Google announced yesterday. Understanding the distinction matters both for users deciding whether to enable it and for developers evaluating how it integrates with their apps.
What Contextual Suggestions Actually Does
A recently published Android support article explains how the feature works: “Before you use contextual suggestions for the first time, you’ll find a notice about the feature that provides a link to your settings.”
The feature analyses your usage patterns on-device — what apps you open and when, what actions you perform routinely, how your daily schedule maps to specific interactions — and uses those patterns to surface timely suggestions before you need to actively seek them out. The practical examples include suggesting your morning playlist before you think to open your music app, prompting your exercise tracker when you typically start a workout, or surfacing a navigation shortcut when your phone detects you are about to leave for a regular destination.
This is distinct from triggered suggestions — where you ask the assistant and it responds — and from Gemini Intelligence’s cross-app automation, where Gemini executes multi-step tasks on your behalf. Contextual Suggestions is ambient and predictive: the phone learns your patterns and surfaces relevant options at the right moment, without you asking.
The Privacy Architecture — Why “On-Device” Actually Means Something Here
The privacy model of Contextual Suggestions is its defining characteristic, and it is worth reading carefully because it represents a genuinely different approach from most AI features.
Unless you give permission to share your data for some other purpose, the data used for Contextual Suggestions is never shared with apps or Google, and never leaves your device. Additionally, apps and services “can’t see this data, but can use the predictions to offer you timely suggestions.” You have the option to disable “device location” and “Delete all stored data” at any time.
That architecture is worth unpacking. The raw behavioural data — what you opened, when, where — stays entirely on the device. Apps cannot access it directly, and Google’s servers never receive it. What apps can receive are the predictions derived from that data — not “this user opened Maps at 8:47 AM on Tuesday from home” but “this user is likely to want a navigation suggestion right now.” The prediction is a signal; the underlying data is sealed.
This is a meaningful privacy distinction from cloud-based personalisation systems, where behavioural data leaves the device to train or query a remote model. Contextual Suggestions runs its inference entirely on the Pixel’s Tensor chip. The trade-off is that predictions are bounded by what on-device compute can support — but for pattern recognition over daily routines, that constraint is not meaningfully limiting.
The ability to delete all stored data at any time — not just disable the feature going forward, but wipe the accumulated pattern data retrospectively — is a user control that most AI personalisation systems do not offer. It positions Contextual Suggestions as a feature designed for users who want AI assistance but have been reluctant to accept the data exchange that most AI features require.
Contextual Suggestions vs Magic Cue vs Gemini Intelligence — Clearing Up the Confusion
Three AI assistance features are now operating across the Pixel ecosystem simultaneously, and they are easy to conflate. Here is the clean distinction:
Magic Cue is the proactive notification surface that prompts users with contextual actions based on on-device signals — suggesting a ride booking when you have a meeting starting soon, offering to start your workout playlist at your usual gym time. It is available on Pixel 10 and Pixel 10 Pro but not the Pixel 10a due to hardware constraints.
Contextual Suggestions is the underlying on-device learning layer that analyses usage patterns to feed timely suggestions across the system. It is not currently appearing on older Pixel phones or Android 17 Beta releases. It is also available on the Pixel 10a — which does not support Magic Cue — making it the more broadly accessible of the two on-device AI personalisation features. Think of it as the pattern recognition engine that Magic Cue draws from, now available to a wider device tier.
Gemini Intelligence is the cloud-assisted, cross-app task automation suite announced yesterday at The Android Show. It requires Gemini on the server side to understand intent, navigate apps, and execute multi-step tasks. It is richer and more capable than either of the above — but it sends data to Google’s servers to do so, which is why its privacy model is opt-in with granular controls rather than the sealed on-device model of Contextual Suggestions.
The three features are complementary layers of the same vision: on-device pattern learning (Contextual Suggestions) feeds ambient suggestions, which combine with Gemini’s cloud intelligence (Gemini Intelligence) to handle complex execution when the user wants more than a suggestion. For users who are uncomfortable with cloud AI data practices, Contextual Suggestions offers a meaningful subset of that intelligence without the trade-off.
Where to Find It and How to Enable It
Once available, scroll down to Settings → [your name] → All services to find the Contextual Suggestions toggle.
The feature is rolling out via Google Play Services 26.18 rather than through an OS update, which means it does not require a full system update to arrive — it propagates through the standard Play Services background update mechanism. If you have a Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, Pixel 10 Pro Fold, or Pixel 10a running Android 16, check your Settings path now. If the option is not yet visible, it will arrive as Play Services updates in the coming days.
The initial notice before first use provides a summary of what the feature does and links directly to your settings page — Google’s stated commitment to informed opt-in before any data collection begins. You do not have to enable the feature to keep your device; the notice is informational and the settings toggle gives you clear control.
Two specific controls to note once you find the feature: the device location toggle, which controls whether your physical location is part of the pattern data used to generate suggestions, and the Delete All Stored Data option, which permanently removes all accumulated pattern data if you decide to stop using the feature or want to reset the learned patterns.
What This Means for Developers
Contextual Suggestions’ current rollout is limited to the Pixel 10 series, which limits its immediate developer impact. But the feature’s architecture — on-device pattern inference feeding predictions to apps without exposing raw data — is a template for how Android is likely to expose AI personalisation signals to apps at scale.
Apps and services cannot see the underlying data, but they can use the predictions to offer timely suggestions. That API contract — receive a prediction signal, not raw data — is a privacy-preserving model that addresses one of the most contentious aspects of in-app AI personalisation. Apps can offer personalised experiences without building their own user tracking infrastructure, and users can benefit from those personalised experiences without handing their raw usage data to every app they install.
For developers building apps in categories where timing and routine matter — fitness, navigation, music, productivity, reminders, food ordering — Contextual Suggestions is worth watching closely as it expands to more devices. Being ready to receive and act on prediction signals when the API becomes broadly available positions your app to deliver a genuinely better user experience at the moment personalisation arrives at scale on Android.
Given that the feature is rolling out via Play Services rather than through Android OS version gating, broader device expansion beyond Pixel 10 is plausible within the current Android 16 ecosystem. The Play Services delivery model is specifically designed for exactly this kind of capability rollout — ship to one device tier first, expand progressively based on stability and compute availability, without waiting for an OS update cycle.
The full developer API surface for Contextual Suggestions, if and when Google formalises one, is likely to be detailed at Google I/O on May 19 — now six days away. Given the timing of this rollout relative to the developer conference, it would be unusual for Google to ship a feature on devices without following up with developer documentation the following week.
Related on Android News Wire: