On-device AI means language and image models running directly on a phone or laptop rather than in a remote data centre. Apple Intelligence on recent iPhones, iPads and Macs, and Google’s Gemini Nano on supported Android devices and in Chrome, are the main examples. For marketers the change is practical: these models summarise emails and notifications, rewrite text and sort messages before a person reads them, so your message is often filtered through an AI layer first.
Apple introduced Apple Intelligence in 2024 with on-device models and a server option called Private Cloud Compute for heavier requests. Google’s Gemini Nano is the smallest model in the Gemini family, built for devices, and powers features on Pixel and other Android phones as well as experimental built-in AI APIs in Chrome. Feature sets differ by device, region and language, and both companies keep adding capabilities.
What does on-device AI actually do on a user’s phone?
The features that matter most for marketing are the ones that stand between a brand and its audience.
- Notification summaries that condense several app or message alerts into one line
- Email summaries and priority sorting in Apple Mail
- Writing tools that proofread, rewrite or summarise text in many apps
- Smart replies and message summaries in some Android messaging apps
- Scam and spam call detection on some Android devices
- Browser-level summarising and writing help in Chrome built on Gemini Nano
Why should marketers care about notification and email summaries?
If a phone summarises your promotional email before the subscriber opens it, the summary may be the only thing they read. Apple Mail can show a short summary in the inbox list instead of the first lines of the body, and it can sort promotional messages into a separate category.
Preheader tricks and teaser subject lines lose some power when an AI reads the whole message and reports what it actually says. A clear, honest offer in the first paragraph has a better chance of being summarised well. Burying the main point under three images and a generic greeting does the opposite.
Practical changes for email
- State the offer and the deadline in plain text near the top
- Use real HTML text, not text inside images, so models can read it
- Keep one main message per email; summaries struggle with five
- Check how campaigns appear on a recent iPhone with summaries enabled
- Watch open rate less and clicks and conversions more, since privacy features already blur opens
How does on-device AI affect privacy and data?
Processing on the device means less personal data leaves the phone. That is a selling point for both Apple and Google, and it fits the direction set by GDPR, Turkey’s KVKK and app tracking rules.
For marketers, it means fewer signals reach your analytics. You will not see what the device’s model concluded about your email or what it summarised. Measurement shifts further towards first-party data you collect with consent: logged-in behaviour, purchases and direct responses.
Can brands build on on-device models directly?
Increasingly, yes. Apple has opened its on-device foundation model to developers through a framework for apps, and Google offers Gemini Nano access on supported Android devices and experimental built-in AI APIs in Chrome for tasks like summarising or translating. These let an app or website run AI features without sending user text to a server.
Consider a retail app that summarises product reviews on the device, or a travel site that translates user questions locally. The benefit is lower server cost and better privacy; the limit is that small models are less capable than large cloud models, and only some devices support them.
What are the limits of on-device AI?
- Small models make more mistakes on complex reasoning than large cloud models
- Support depends on recent, higher-end hardware
- Language coverage expands gradually, so not every market gets every feature
- Summaries can misread tone, sarcasm or promotions
- Developers cannot assume every visitor has the feature available
Apple paused some notification summaries for news apps in early 2025 after inaccurate headlines drew complaints, which shows how a small model can distort a message. Brands should expect occasional misreadings and keep messages simple enough to survive them.
How Alien Road applies on-device AI insights
In a typical engagement we start with an audit of how the client’s emails, push notifications and SMS look when filtered through current device features. We test campaigns on recent iPhones and Android phones with summaries and categorisation enabled, and capture screenshots of what the subscriber actually sees.
Setup work covers rewriting templates so the key message sits in readable text at the top, cleaning up sender names and subject lines, and adjusting push notification copy so a summary keeps the point. Measurement moves weight from opens to clicks, conversions and revenue per send, and we compare segments by device where the email platform provides it. Reporting includes the screenshot set, so the team can see changes in presentation, not just numbers.
Where do these features go next?
Both companies are moving towards assistants that act inside apps on the user’s behalf, booking, searching and filling forms. When that happens, clear structured information on your site and in your app will matter even more, because an assistant has to understand it quickly to use it.
What should app and website teams test?
Product teams should check how their apps behave when system writing tools and summaries are active. Form fields, chat windows and review sections can all be rewritten or summarised by the operating system.
- Push notification copy read as a summary on a locked screen
- Long product descriptions summarised in the browser
- Text fields where users may paste rewritten content
- Customer support chats where replies are suggested by the device
Where to start this week
Send your last three marketing emails to a test inbox on a recent iPhone with Apple Intelligence enabled and to an Android phone with Gemini features. Look at the inbox summary, the category and the notification. If the summary misses your offer, rewrite the first paragraph so the offer is the first thing a model reads, and send the next campaign that way.
Keep reading: AI Optimization · AI Optimization