Digging out an ancient text message is a real job, and it arrives when it arrives. Wired's tester says the rebuilt Siri made that easier than ever, thanks to personal context drawn from indexing the contents of the phone [3]. The capability is bounded by how often the need shows up. Better retrieval does not create more things to retrieve, so the ceiling on usage sits in the user's archive rather than in the model.
A step change in quality does not guarantee a step change in usage. The tester agrees about the quality, calling the new version quite the step change [4], and still reports going back to typing and swiping manually, on Instagram and on Yelp [6], with Anthropic's Claude already installed and used fairly often on the same handset [7]. The competition here was existing habit plus a chatbot that had already been chosen, not another assistant's answer quality.
The word that matters most here is neutral. The tester says they were not souring on Siri AI but growing more neutral, and did not mind it becoming essentially the search bar for the smartphone [5]. Neutral registers on a dashboard as a live install with some retrieval traffic and nothing obvious to fix, which is exactly the state a team watching those numbers would call healthy.
Against that, Avi Greengart of Techsponential reports the opposite direction from the same beta: using Siri more, and for more things, with buried information found faster than by any other method he has, better world knowledge and better dictation [10]. Wired allows that its reviewer may be an outlier and that many other beta testers are still using the assistant [11]. So the named first-person accounts in this story split one to one [15], and the piece rests on those two accounts rather than on request counts or retention figures [14]. Either self-report is one phone.
Dipanjan Chatterjee of Forrester told Wired that Siri will take criticism for not matching its peers and that it will not make a difference to the vast installed base or send iPhone users elsewhere [8]. Ramon Llamas of IDC expects users deep in the Apple ecosystem who had qualms about other AI tools to be open to trying this one [9]. Trial is the cheap part; the tester tried it with enthusiasm and still ended up at the search bar. Apple can absorb that outcome because the assistant is preinstalled and paid for by the phone, a cushion an app that has to be opened on purpose does not have.
The grid worth drawing before your own launch has two axes: whether the job recurs on the user's daily schedule or on its own irregular one, and whether the surface where it happens is already owned by a habit. Occasional need on an owned surface is where this tester landed, and that cell describes a feature, measured by whether a second and third retrieval happen in month two, not by sessions. Daily need on an unowned surface is the only cell where an assistant earns its own icon. There is also a slice of the base that never enters the grid at all: Wired notes iPhone owners who will avoid the new Siri because of the AI in its name, and who will not be moved by feature improvements [12].
If your assistant's best job is an occasional one, put it inside something the user already opens and accept that you will have no growth curve to show for it. Promising an everything tool [1] instead buys a strong first week and a neutral user by month two, and neutral is the hardest result to argue with in a review.
