AI assistants cannot borrow your trust to sell you things

When an AI assistant says it is working from your conversations, calendar, photos, or purchase history, people form a pretty concrete picture in their heads. They assume it saw something. They assume that is why the answer showed up. They may also assume the recommendation is trying to help them rather than help somebody else sell something.

Those assumptions are not a footnote. They shape what people share, whether they speak freely around a device, and how much weight they give a suggestion. The FTC’s endorsement guidance makes the same common-sense point for product recommendations: a relationship that would change how someone evaluates a recommendation should be disclosed clearly and conspicuously.

For AI assistants, that should mean the explanation appears before the useful-looking answer turns into a tap, a call, or a purchase. Not in a policy link after the fact.

A good privacy and recommendation screen does not need to make people read a legal memo. It does need to answer a few ordinary questions in plain language: Did this feature use my microphone, or did it not? Did it use my past activity, a broad search, or nothing personal? Is this a paid placement or a partner result? Can I turn off that use without losing the rest of the assistant?

The awkward sentence is usually the one worth putting on the screen. “We did not use audio for this.” “This result is sponsored.” “We only checked your calendar for this answer.” “This setting is off.” If a company is reluctant to say one of those sentences clearly, the problem is not the label design.

There is a real upside here. Clear limits let people use an assistant for the small, boring things without wondering whether every helpful answer came with a hidden audience or a sales target attached.

The test is not whether a company can explain its data practices to a conference room. It is whether someone can understand a recommendation while standing in the kitchen, late for pickup, with one hand on the door. If they cannot tell what fed the answer or who paid to appear in it, they are not making an informed choice. They are just being asked to trust the tone.

AI assistants will keep getting closer to the errands and decisions that fill a week. That can be genuinely useful. But as they become more familiar, their claims need to get less slippery. A helpful system should never make a person guess whether it heard them, used them, or sold them something.