How to choose an AI model when your phone starts giving you options
- 2 days ago
- 3 min read
At some point soon, your phone is going to ask you a question it has never asked before.
“Which AI model do you want to use for this?”
Apple is planning to let people choose from multiple third‑party AI services across iOS 27, iPadOS 27, and macOS 27, for things like writing, editing, image generation, and even how Siri handles certain tasks (Reuters, 2026). TechCrunch says this will show up as a system called “Extensions,” where apps can plug their own models into Apple’s interface so you can pick them from Settings (TechCrunch, 2026).
In other words, the iPhone is becoming an AI model marketplace. And you are suddenly the procurement team.
You are not going to read technical papers. You are not going to benchmark models. You have other things to do. So what is a reasonable way to choose?
A good starting point is to treat each model like a mix of three things:
a capability profile,
a data‑handling arrangement,
and a default you may end up living with for years.
Capability is the flashy part. Different models are better or worse at writing, coding, reasoning, or image work. Early reports suggest Apple has been testing integrations with services from Google and Anthropic, alongside its own stack (Reuters, 2026; TechCrunch, 2026). If you already use one of these on the web for serious work, choosing the same one on your phone is a reasonable first filter. It keeps your mental model of “what this thing is good at” consistent.
Data handling is usually where people stop reading the fine print, but it is the part that actually touches your life. Some models run more work on‑device, which is better for privacy but may be less powerful. Others lean heavily on cloud compute, which can mean better performance but more data leaving the phone. If you care about this, look for two simple things in the settings or provider docs:
whether your prompts are retained and used for training,
and whether you can turn that off without losing the product entirely.
The third piece is the one that gets underestimated: default power. Once a model becomes the thing your phone calls silently in the background for suggestions, autocompletes, summaries, and “small” tasks, it stops feeling like a choice and starts feeling like part of the environment.
That is where some of the more interesting policy thinking is starting to show up. A recent paper on “assistant neutrality” argues that regulators should care less about individual answers and more about whether people can easily switch, understand what is being used, and see which service is behind a response (Berkeley Technology Law Journal, 2026). That sounds abstract until you realise it maps almost exactly onto what you, as a normal user, need from Settings:
a clear label of which model is active,
a simple way to change it,
and some basic visibility into what it is doing with your data.
So a practical, human checklist for iOS 27 looks something like this:
1. Pick the model you already trust for serious tasks on the web, if you have one.
2. Check whether the provider lets you opt out of training on your data without breaking the experience.
3. Favour setups where you can see and change the default quickly.
4. Be willing to change your mind. The first choice does not have to be the final one.
All of this sounds like more work than just tapping the first option suggested. It is. But this is the layer that will sit between you and most of your writing, searching, and small daily decisions. The question is not which model is “best” in the abstract, but which one you are comfortable letting into the background of your life.
For the first time, your phone is going to ask. It is worth having an answer.
Sources
TechCrunch. (2026). Apple plans to make iOS 27 a Choose Your Own Adventure of AI models.
Reuters. (2026). Apple to let users choose rival AI models across iOS 27 features, Bloomberg News reports.
Berkeley Technology Law Journal. (2026). Assistant neutrality in the age of generative AI.
Comments