It’s taken a long time to reach this point. Those with long memories will recall that Apple promised an updated version of its Siri voice assistant at WWDC 2024. This would bring deep contextual awareness, we were told, and the ability to search emails, messages, and other personal data to improve accuracy and enable rich, natural conversations. There was even a series of ads showing off the feature and explicitly tying it to the iPhone 16 Pro.
Unfortunately the project stumbled right from the start and carried on stumbling for the following two years. An initial plan to launch New Siri as part of iOS 18 was quickly shelved in favor of iOS 18.4, then iOS 26, then iOS 26.4. Anyone who bought an iPhone 16 Pro with the feature in mind was headed for disappointment (and some compensation, on the bright side).
Now the new Siri has been announced again at another WWDC, this time rebranded as Siri AI and tied to the launch of the OS 27 software updates. Assuming nothing else goes wrong, it should finally hit Apple devices in the fall. The problem is that the world has moved on, and even at its own launch event, new Siri didn’t feel very new.
The part that struck me first was how slow it seems to be. These were stage-managed demos, presumably run in absolutely optimal conditions and with the option of a rerun if anything falls over or hits a snag. (We’ll be generous and assume they were genuine demos and not mocked-up simulations like in those Bella Ramsey commercials.) But even in these favourable conditions, there was a noticeable delay before Siri AI answered each command. The presenters would often have time to make an additional comment while waiting for Siri to think, shown by a spinning circle in the Dynamic Island.
I went through Mike Rockwell’s initial demo with a stopwatch, measuring from when Siri started showing a loading icon to when it returned the result. (You’ll also need to factor in the time to speak or type the command, as well as for Siri to register this and switch to thinking mode, but I was again feeling generous.) The quickest was 3.71 seconds, and the slowest 8.31 seconds. When Rockwell’s colleague Justin Titi started showing off more complex tasks the delays got longer still, stretching in one case to 10.43 seconds.
That may not sound like long, but bear in mind that this is supposed to be an action you run repeatedly, and it’s scarcely the conversational mode we were promised. (Imagine talking to a friend and having to sit through a 10-second pause after each of your comments.) It also doesn’t compare well to rival products. In my experiments, ChatGPT was repeatedly able to answer complex analytical questions in less than 2 seconds, peaking at 2.6 seconds when I added multiple additional parameters. And remember that this is comparing ChatGPT in the real world to Siri AI in lab conditions.
On functionality, too, Siri AI hardly feels new. Basically, everything we saw from it has been done before. Access to broad world knowledge? That’s standard. Integrated across the platform? That’s what Google announced last month. Ability to revisit past conversations? Gemini does that. As well as search your emails and messages, tell you where a photo was taken, and agentically organise a party.
Which makes sense, given that Siri AI was built on Google’s foundation after Apple couldn’t finish the job itself. If you’ve used Gemini, it’s unlikely you’ll find anything novel here. But the same applies in most cases to ChatGPT or any other modern LLM. About the only thing you might feel is unique to Apple’s offering is its privacy promises, and there’s still a bit of a question mark over that because of the server situation.
Apple
The positives
What we will say, however, is that New Siri feels like a big improvement on Old Siri, and that’s something. The following features might not be new across the industry, but they are new for Apple.
Better accuracy: This will need to be tested, and it was grating to hear Apple use the phrase “even higher accuracy,” as if Old Siri was remotely close to acceptable in this regard. But a specific reference to improved accuracy is exciting for those of us who’ve been tearing our hair out trying to get Siri to do what it’s told.
A dedicated app: For the first time you’ll find a Siri app on your iOS 27 iPhone. Open this app to view past conversations and the information Siri surfaced in response to those queries. A particularly nice part of this is that it’s device-agnostic: the same conversation history will be viewable on your iPhone, iPad, Mac, and so on. (Apple assures us that the syncing is done privately with iCloud.)
Contextual awareness: This is the big one. In the demos Siri AI knew what was on the screen, what it had been asked before, personal contextual information, and (when prompted) the contents of the user’s emails and messages. This means you can use commands like “Where was this taken?” while looking at a photo, or ask Siri what recipe a particular relative sent you recently. This is a huge step forward in power.
Writing tools in more places: Apple says its customers can use Siri AI to compose or edit text “virtually anywhere they type.” Describe what you’re looking for and Siri will deliver a draft for you to check out. Slightly creepily, the technology can mimic your writing style (on a per-recipient basis) when composing in Mail and Messages.
Customisable, more expressive Siri voices: If you have access to the top-tier version of Siri AI (you’ll need an iPhone Air, iPhone 17 Pro/17 Pro Max, iPad M4 or later with 12GB of memory, Mac M3 or later with 12GB, or Apple Vision Pro M5) then you’ll be able to customise Siri AI’s voice using two sliders. One will adjust speed, the other expressiveness. Quite a fun idea.
