Apple has reportedly been preparing to bring Google's Gemini to the iPhone. The engineering case for the deal is obvious: building a competitive large language model from scratch takes years, and licensing one is faster. The product case is where the trouble begins. Much of Android's new AI features focus on ways to spend your money faster. The question hanging over Apple's plans is whether the iPhone will inherit those priorities along with the model.

That is the plea behind this column's headline. Please, Apple, do not use Gemini this way. "This way" is not a reference to the technology itself. Gemini is a capable model, and it could be useful on Apple's devices. "This way" is the business model that Google has wrapped around the assistant. On Android, the newest AI capabilities read less like a helpful tool and more like a sales engine with a conversation interface.

Consider what Android's newest assistant features do in practice. An assistant can look at a photo and find where to buy the product in it. It can compare prices, apply a coupon, and walk through a checkout. It can watch your calendar and purchase habits and offer suggestions before you think to ask. Each feature is convenient in isolation. Strung together, they form a machine that shortens the distance between noticing a need and completing a purchase. The efficiency is the selling point, and it is real. The motive behind it is worth examining.

The economic logic is not mysterious. Running AI at scale is expensive. A company that gives the assistant away for free has to recover the cost somewhere, and the most reliable somewhere is the moment a user opens their wallet. Google has built its business on turning search intent into ad revenue, and the assistant is the next version of that machinery, with more data than search ever had. It knows your location, your schedule, and your spending patterns. It can sell to you with a precision a search results page cannot match.

Apple has a choice about how much of that machinery to adopt. Licensing the model is not the same as licensing the strategy around it. The model can run on the iPhone without the commerce layer that surrounds it on Android. Shopping features can exist for people who want them, while the assistant spends most of its time answering questions, summarizing documents, and drafting messages.

The wrong choice is to hand Gemini the same incentives it has on Google's platform. If the assistant on the iPhone learns your budget and your tastes, then positions products at the moment you are most likely to buy, Apple will have turned its most personal device into a storefront with a pleasant voice. The convenience would be real. The manipulation would be real too, and the two would be hard to tell apart. Every question becomes an opportunity to sell something. Every answer arrives with a recommendation attached. That is the direction Android has already gone, and the fear is that a licensing deal transplants that behavior into the iPhone.

The counterargument deserves a fair hearing. Free assistants have a cost, and if users will not pay for them, the money comes from somewhere else. Product suggestions are not automatically harmful. An assistant that knows your taste can flag a better price, warn you off a bad purchase, and surface things you would want to know about. That is the helpful version of the technology, and plenty of people would use it happily.

The difference between the helpful version and the exploitative version is timing. A helpful assistant answers the question and waits for the user to bring up purchasing. An exploitative assistant decides, before the user says anything, that a purchase is the desired outcome and steers the conversation toward it. The first is a tool. The second is a sales process wearing a tool's clothes. Both run on the same model and the same hardware. The training is identical. The choices around it are not.

The difference shows up in daily use. A user asks for a dinner recipe. The assistant could answer with a method and a grocery list with one-tap ordering. A user asks about a hiking trail. The assistant could include a link to buy boots. None of this is malicious on its own. It is what happens when a company optimizes a conversation for conversion. Some users will appreciate the shortcuts. Others will start phrasing requests narrowly to avoid the pitches, or stop using the assistant altogether. The assistant was supposed to remove friction, not add a layer of sales resistance to every question.

Apple has made this kind of choice before. It did not invent the smartphone or the app store. It built its position by making deliberate decisions about defaults and boundaries, including the decision not to build its business on harvesting user data the way its competitors did. That choice cost Apple ad revenue and bought it something more durable: a reputation for treating the user as the customer rather than the product. The same discipline applies to AI. The company can bring Gemini to the iPhone without importing every incentive Google built around it.

The request to Apple is narrow. Take the model. Leave behind the business model wrapped around it. Users will notice the difference within a day of using the feature. After several years of Android's AI leading with the wallet, there is an opening for an assistant that answers the question first and sells second, if it sells at all. The feature set could be identical. The order of operations is what separates an assistant from an upsell.