Some keyboard word suggestions work even in airplane mode
Your keyboard may suggest the next word before you finish typing. Some devices can do this offline using a model installed on the device. Whether your feature works this way depends on the keyboard, device and settings. This guide shows you how to test one harmless example and check what the result does and does not tell you about privacy.
Why this matters
A local model can suit limited text tasks when the app and device support it. In this guide, you will see how to spot this on your own phone.
First try: test one feature that runs on your device
- 1
Open your phone’s keyboard, dictation or AI settings
Find the section in settings for the keyboard, dictation or AI assistant. On many phones, at least some of these functions are described there as on-device or offline features. The wording varies by manufacturer.
- 2
Choose a feature already on your phone
Choose word suggestions or dictation that your phone’s help describes as supporting offline use. Availability varies by device, language and settings. Dictation may require a language package, including for Finnish, to have been downloaded first. Use an existing feature; if it is not ready for offline use, note that limitation rather than assuming it will work.
- 3
Switch on airplane mode and try it
Switch on airplane mode, then check that both Wi-Fi and mobile data are off; Wi-Fi can remain enabled on some phones. Try harmless, made-up text without names, health information or other personal details.
- 4
Reconnect and compare
Turn the connection back on and try the same feature again. Compare its speed and result: did it work just as well in airplane mode? Finally, check what the settings or the app’s help pages say about data processing.
Try this
Type this in airplane mode and look at the suggestions
I will remember to buy milk and call Mum tomorrow.
Compatible with: Your phone’s local keyboard (or dictation, if you already use it offline)
Check and improve
Explain what just happened
Explain in everyday language why keyboard word suggestions or turning dictation into text can continue on a phone without an internet connection, while creating an AI image, for example, usually stops as soon as the connection drops. Briefly explain what this difference generally means for privacy, too.
If the answer claimed to know your phone
Tell me exactly what you cannot know about my particular phone in this conversation, and where in the settings I need to check it myself.
This test shows whether your chosen feature can complete that task without a connection. It does not prove that all processing is local in every mode, or that nothing will be sent after you reconnect. Check the app’s help and privacy information as well.
Where the difference shows up in everyday life
Dictating at a cottage without a connection
Example prompt:
You are at a summer cottage with no network connection at all, dictating a note using your phone’s dictation feature. What difference does it make whether the phone processes your speech on the device or sends it to a server first?
Result:
If supported on-device dictation is enabled and ready, it can turn speech into text without a connection. That processing step does not need a server. Server-based dictation needs a connection for that step; accuracy depends on the system and task. Check the app’s settings and privacy notice for storage, later uploads and other features before using personal information.
Translating a sign abroad
Example prompt:
You are abroad without mobile data and try to translate a sign using your phone. Why do some translation apps still work while others only show an error message?
Result:
Some translation apps download the language to the device in advance as a small model. It works offline, more slowly and with more limitations than online, but it still works. Apps that send the image or text to a server for translation need a working connection every time; without one, they show an error. You cannot tell the difference from the app’s appearance: what matters is whether the language has been downloaded to the device in advance.
Your gallery finds a photo by a word even when the internet is down
Example prompt:
Your phone’s photo gallery finds an old picture when you search for “dog” or “beach”, even though you have never labelled the photos with those words yourself. The photos are already on your phone. Does that mean the search also works offline?
Result:
Not always. Some gallery apps recognise the contents of images directly on the device. In that case, word searches work offline too, because both the photos and their recognition are local. Other apps send image analysis to a cloud service, so search stops as soon as the connection drops, even though you can still see the photos in the gallery. Having material on your device does not, by itself, tell you whether the AI processing happens there too.
A smaller model makes different mistakes
Keep requests short and check the result
Suitable tasks are short and clearly limited, and you should always check the result yourself. A small on-device model is typically more lightweight than a large cloud model: it understands less context, makes mistakes more easily with long or complex requests, and does not handle the nuances of your language or facts as accurately as a large online model.
Turn a claim about local processing into something you can check
A common request:
Does this AI run safely on my phone?
Try asking this instead:
My keyboard’s word suggestions just worked without a network connection. Make a short checklist: what happens directly on the device, what information the feature might send online, what permissions it asks for, and where I can find this information in the phone’s or app’s own settings.
A checklist that separates the device, information sent online, permissions and the app’s own information gives you a concrete answer. A simple yes-or-no question about safety does not tell you where the answer should come from or what you can check yourself.
Once you have the basics
If you want to go further, you can later try installing a separate app that runs a small language model entirely on your device. If you would first like to understand the general differences between models, continue to the guide on Mistral or open models.
Free practice on this website
Separate on-device processing from online processing
Read the practice material once, hide it and answer in your own words first. The hint and checking criteria only become available after you have tried or chosen “I don’t know yet”.
The answers you write stay only in this browser tab. They are not sent to AI Licenses’ server or to analytics. Use a fictional example and do not enter the name of your actual app or your own personal information here.
Question 1/2
Read the material before your first attempt
Once you have read the material, hide it using the button on the left and answer from memory.
Use the practice activity in your own AI service (optional)
On the device or online? See how to tell.
An on-device model is more lightweight than a cloud model, which means it stumbles in different places. See how to tell which method is being used right now.
Experiment · about 5 min
Where shall we start?
How the request is built
- Task
- Separate features that run on the device from those that run online
- Background
- In general, rather than on my particular phone
- Format
- One identifying sign for each feature
Tell me which AI features on a phone can typically work on the device without a network connection and which require a connection. For each one, explain how I can tell which applies. Do not claim to know the features of my phone.
Ready to reflect on what you saw?
This opens a regular cloud-based assistant to help you describe the difference. It is different from the local feature on your phone that you just tested. You can use any conversational assistant.
Compare the answer with the original text: Both tasks use general example material. A lighter model is particularly likely to drop conditions and exceptions, so always compare its answer with the original before relying on it as it stands.
Skills and concepts in this guide
Publication details and sources
- Published:
- July 2, 2026
- Sources checked:
- September 4, 2026