New Siri AI improves—but tests expose accuracy gaps

Apple’s new Siri AI is available via a waitlist for supported devices running OS 27 developer beta, but a set of Mac-based tests shows it can still miss facts, struggle with back-and-forth conversations, and get details wrong—especially when identifying images
On my Mac, I finally got to test Apple’s new Siri AI the way I test other chatbots: with direct questions, real-world tasks, and a willingness to push when it gets things wrong.
The promise is clear. At WWDC 2026 earlier this month, Apple touted a Siri that’s more conversational, more responsive, and less error-prone. But to try it, you have to jump through a few gates first—starting with a waitlist.
To access the new Siri AI, an iPhone, iPad, or Mac needs to support version 27 of its respective OS and also support Apple Intelligence. After that, you install the developer beta. Since those betas can be unstable, I used a spare iPhone 15 Pro and a spare MacBook Air M1 rather than my main devices.
Then comes the wait. On a Mac that meets all the criteria, I was able to get access without waiting too long. On my iPhone, I joined the waitlist last week and I’m still waiting.
Getting it running on a Mac takes another step: open Settings (System Settings on a Mac), select the Siri setting, and tap the button for Turn Siri On. A message says you’ll be notified when the new Siri is available.
Once it’s there, Siri AI can be reached in a few ways. If voice activation is enabled. you can say “Hey Siri” or “Siri.” You can also click the Siri AI app icon on the Dock. where Siri AI lets you type or speak your request. Pressing the Command key twice brings up a small text window for typing a question. and Command plus the spacebar launches the familiar Spotlight search window—now with searching through asking Siri. Right-click a window or other item and the pop-up menu includes an option to Ask Siri.
The way I tested it was simple: ask, follow up, and try to see whether it can stay in conversation the way systems like ChatGPT and Gemini often do.
The new Siri does work like chatbots in many ways—but it’s less chatty and more direct. When I started by asking what’s new, Siri didn’t linger in casual back-and-forth. Instead, it delivered a rundown of the latest news stories.
When I asked a history question—Why did the Roman Empire fall?—it gave a short explanation and then followed with bullet-point causes. The length was about the same as I get from other AI systems, and it also cited sources with links I could open to check.
But when I switched to a more personal recommendation, the limits showed up. I told Siri I had $2. 000 to spend on a laptop. and that I value keyboard quality and battery life more than performance. I asked what I should buy. Instead of offering its own judgement immediately. Siri linked me to a few articles and social media posts about laptops. but didn’t give its own opinion or summarize what it found. When I asked it to summarize the information and give me its own opinion. it did—though the first answer left me disappointed.
Then I tested whether it could find specific information inside my own library.
I asked Siri to find all photos of the statue of Abraham Lincoln in my Photos library. It found only three photos, even though my library actually contained six matching photos. I wasn’t able to tell what criteria it used—or why three were missed. The same kind of problem showed up when I asked it to find other photos using specific details: it found some but not all.
Image understanding, too, wasn’t reliable.
Like most AIs, Siri can analyze files you upload. I uploaded a photo of a painting by Toulouse-Lautrec and asked it to identify the name and artist and give background. Siri got it wrong. It provided the wrong name for the painting and the artist. I tried again with a different painting. This time, Siri got the correct artist but misidentified the name. On a third attempt—this time with a popular Van Gogh painting—it finally got it right.
When I switched from art to everyday life, the advice improved, but the interaction still felt awkward.
I told Siri my cat, Mr. Giggles, sometimes won’t eat his usual food and I’m not sure what else to try. I asked for suggestions. Siri offered advice that was clear and helpful, and it asked whether my cat typically eats wet or dry food. After I answered, Siri provided more information.
The content was solid, but the conversation mode didn’t flow. Instead of keeping the communication open, Siri seemed to stop listening after each response. Each time I wanted to speak again, I had to click the microphone icon. The back-and-forth fluidity I’ve experienced with other AIs wasn’t there.
I saw a similar pattern with on-screen content.
When I right-clicked a ZDNET story. chose the Ask Siri option. and asked it to summarize the story on the screen. my phrasing mattered. Initially, when I asked it to summarize the story, it didn’t seem to understand. When I asked it to summarize what it saw on the screen. it summarized only the text visible on the screen. But when I finally told it to summarize the story on the screen. Siri AI provided a concise but helpful summary of the key points across the entire story.
Siri AI also keeps track of conversations and syncs them across Apple devices. To test that. I right-clicked a previous chat. which brought up a pop-up menu with options to rename the conversation. pin it. open it in a new window. or delete it. I also tried resuming a past chat. I told Siri it was wrong when it gave me the name and artist for a Toulouse-Lautrec painting. and it tried again to identify it—but was still mistaken. Only after I told it the painting was by Toulouse-Lautrec could it identify the name and provide background.
Put simply, the new Siri AI is an improvement over the old Siri. But it’s not consistent enough yet to feel dependable. Incorrect or inaccurate answers are still part of the experience, and the conversation back-and-forth feels clumsier than it should.
This is the first developer beta of Siri AI, though—and Apple has several months to fine-tune it before its expected public release in September.
For now, I’m left with a mixed verdict: promising, but not ready. Not the way I’d want it to be for the kind of tasks where accuracy and smooth conversation matter.
Apple Siri AI OS 27 Apple Intelligence WWDC 2026 developer beta chatbots MacOS iPhone accuracy conversation flow Photos image recognition