On-device AI runs language and image models directly on a phone, tablet or laptop instead of in a cloud data centre. Apple Intelligence on recent iPhones, iPads and Macs, and Google’s Gemini Nano on Pixel and some other Android devices, handle tasks like summarising notifications, rewriting text, transcribing recordings and editing photos locally. Workflows built on them save time on quick, private tasks, and they change how audiences receive marketing messages.
The appeal is speed and privacy. A model that runs locally does not need to send text to a server, can work offline and responds quickly. The trade-off is size: on-device models are much smaller than cloud models, so they handle short, focused tasks rather than long analysis.
Apple pairs its on-device models with Private Cloud Compute for heavier requests and offers optional ChatGPT integration. Google uses Gemini Nano on the device and the larger Gemini models in the cloud, and Chrome has started exposing built-in AI features through experimental browser APIs. The exact features differ by device, region and language, and availability has expanded gradually, so check what your own devices support.
What can on-device AI do today?
The features most relevant to marketing and agency work fall into a few groups.
- Writing tools: rewrite, proofread, change tone and summarise selected text in apps such as Mail, Notes and many third-party apps on Apple devices.
- Notification and message summaries: long threads and grouped notifications condensed into a line or two.
- Recording and transcription: voice memos, calls and meetings transcribed and summarised, depending on the device and app.
- Image editing: removing objects from photos and generating simple images or emoji-style graphics.
- Smart replies and suggestions: short draft responses in messaging apps.
Several of these also exist in cloud tools. The difference is that they are built into the operating system and available to everyone with a supported device, without signing up for anything.
Which workflows save time?
Fast proofreading on the move
Account managers writing client emails on a phone can select a draft and use the proofread or rewrite tool before sending. It catches typos and awkward phrasing in seconds without copying text into another app.
Meeting and call notes
Recording a client call or a planning session and getting a local transcript and summary means notes are ready before the next meeting. For sensitive discussions, on-device processing keeps the content on the device, which is easier to justify to privacy-conscious clients. Always tell participants a recording is being made, as the law and courtesy require.
Content capture at events
At a trade fair or shoot, staff can record short interviews, transcribe them on the spot and send summaries back to the content team the same day. Photo clean-up tools can remove a stray object from a product shot for quick social use, while final assets still go through proper editing.
A simple workflow template
- Capture: record or write on the device.
- Process: use the built-in summarise, transcribe or rewrite tool.
- Check: read the result and correct names, figures and product terms.
- Share: send to the shared folder or project tool with a note that it is AI-assisted.
- Archive: store the original recording or draft for reference.
What does on-device AI change for marketers?
The bigger impact is on the audience side. When a phone summarises notifications and emails, the summary may be the only part a person reads.
Email subject lines and the first lines of a message now feed summary features. Clear, specific opening sentences make it more likely that the summary carries the right message. A clever subject line that hides the point may be summarised into something plain or confusing.
Push notifications are grouped and condensed. Brands that send many notifications may find them reduced to a single summary line, so fewer, clearer notifications work better than frequent ones.
Apple’s Mail app on supported devices can also sort messages into categories and highlight priority messages. Promotional emails may land in a promotions view, similar to Gmail tabs. Engagement and sender reputation matter more than tricks.
What are the limits?
- Small models make more mistakes on complex or long text than cloud models.
- Language support varies; some features arrive in English first.
- Summaries can drop important qualifiers, like a discount’s end date or conditions.
- Features depend on hardware; older devices do not get them.
- Admins in organisations may restrict features through device management.
How Alien Road applies on-device AI
In a typical engagement on-device AI matters most in email and app messaging. Discovery includes reviewing a client’s recent email campaigns and push notifications as they appear on current iPhones and Android devices, including how summary features condense them. We look at whether the key offer and deadline survive in the summary.
Setup changes are usually simple: rewriting the first sentence of each email to state the main point, putting dates and conditions early, reducing notification frequency and making sender names consistent. We also check that preheader text is set deliberately rather than pulling in navigation links.
Measurement uses the client’s email platform and app analytics. Open rates have been unreliable since Apple’s Mail Privacy Protection, so reports focus on clicks, conversions and unsubscribes, compared over several sends before and after the changes.
Should brands build their own on-device features?
For most brands, no. Developers can call Apple’s Foundation Models framework in their own iOS apps and use Gemini Nano through Android’s AI features on supported devices, which allows features like local summarising or smart suggestions inside a brand’s app. That makes sense for apps with heavy daily use and private content, such as notes, finance or health apps.
For a typical retail or service app, the effort is rarely worth it. The better investment is making sure your content reads well when the operating system’s own features summarise it.
Where to start this week
Send your last three marketing emails to a test inbox on a device that supports message summaries, and look at how they are summarised. Note whether the offer, deadline and call to action survive.
Then rewrite the first sentence of your next email so it states the point plainly, and compare clicks with your recent average. On your own team, pick one task, such as client call notes, and try the on-device transcribe-and-summarise workflow for two weeks.
Keep reading: AI Optimization · AI Optimization