Veteran tech journalist David Pogue spent the summer beta-testing Siri AI, and once iOS 27 shipped this week, he ran the finished version through 125 real prompts to see how often it actually gets things right. His verdict, after months of use, is about as enthusiastic as tech reviews get.
Writing on his Substack, Pogue called the new Siri a phone-changer
, one that replaces most Google searches and AI chatbot detours with a single spoken question, because it can now see directly into a user’s own Mail, Messages, Notes, Files, Calendar, Contacts, Reminders, Photos, and Music.
Siri AI Is Best at Answering Questions About Your Own Life
The most impressive category in Pogue’s test wasn’t asking Siri to do something new, it was asking Siri to remember things a person had already told their phone without realizing it. Siri pulled a driver’s license number that existed only as a photo in Notes, never typed anywhere as text. It calculated exactly when to leave for a wedding by combining a calendar entry, a starting location, and live drive-time estimates, all from one spoken question with no other details supplied. It even correctly named where Pogue had been on a random date months earlier using nothing but the location data buried in a single photo he had taken that day.
Out of 38 tests in this personal-data category, only two failed, and both traced back to the same gap. The Health app hasn’t been updated to work with Siri AI yet, so anything involving step counts or medication reminders still gets punted back to opening the app manually.
Some Old Siri Tricks Don’t Work Anymore, and the Reason Is Structural
Phone and app control was the least reliable of Pogue’s five categories, with roughly seven outright failures and a couple of partial ones out of 47 attempts. Siri AI couldn’t unsubscribe from an email newsletter, delete an app, skip to the next chapter in an audiobook, or add an item to a checklist-style note as an actual checkbox instead of plain text.
he old Siri could call an Uber or send a Venmo payment directly. The new one can’t, not because the feature regressed, but because Siri AI requires each third-party app to build its own support for the new system, and Uber and Venmo haven’t done that integration yet. It’s the same reason Gmail isn’t Siri-compatible at all, which Pogue suggests may be less about technical limits and more about Google not wanting people to stop opening the Gmail app and seeing its ads.
Visual and Screen Intelligence Are the Categories Worth Learning First
Siri AI’s camera and screen-reading abilities came through almost flawlessly, and they cover tasks that used to require several separate steps. Pointing the camera at a restaurant bill and saying Split Bill divides it by who ordered what and settles up through Apple Cash. Looking at a recipe in Safari, Siri can double every quantity, swap an ingredient, drop items already sitting in the pantry, and add exactly what’s left to a grocery list in one spoken request. It can identify the show playing in a random video clip, fact-check a news article, and read the fine print of a phone’s Settings menu out loud.
Combined, Pogue’s Visual Intelligence and Screen Intelligence categories, close to 30 tests in total, produced only a single partial failure, reading text aloud worked on iPhone but not on Mac.
Across all 125 prompts, Siri AI answered roughly 110 of them cleanly, stumbled on a handful more, and failed outright about ten times, a strong showing for a company Pogue describes as the last major tech firm to fully embrace AI.
Our Take
The most useful part of Siri AI has nothing to do with flashy demos and everything to do with not having to scroll back through months of texts to find one detail somebody mentioned once. That’s a genuinely hard problem to solve well, and a 110-out-of-125 hit rate on a review this thorough is a real result. The bigger long-term risk for Siri AI has less to do with Apple’s own execution and more to do with whether enough third-party apps bother building support for a system that quietly reduces how often anyone opens their app in the first place.
