ADAM ENGST 13 December 2024

Apple has integrated ChatGPT into Siri with the second set of Apple Intelligence features that debuted in this week’s operating system releases (see “OS X.2 Updates Boost Apple Intelligence and More,” 11 December 2024). Don’t get too excited.

I’ve been trying to use this feature in the betas and now in the release versions of macOS 15.2 and iOS 18.2, and if anything, it has increased my frustration when interacting with Siri. Worse, I fear that some deeper issues may argue against the integration of Siri and ChatGPT.

Functional Problems

A few of the functional problems I’ve encountered include:

• You must enable ChatGPT separately for each device, so the first time you issue an involved query to Siri on a new device, your query will fail, and you’ll be prompted to enable ChatGPT. It doesn’t feel welcoming.
• On my M1 MacBook Air, I am continually told that ChatGPT is unavailable and to try again later, even while it works fine on my iPhone 16 Pro. My MacBook Air is seemingly cursed because Apple Intelligence summarization never works on it, either. A call to Apple support is in my future.
• For privacy reasons, Siri asks if you want to use ChatGPT on each prompt that goes beyond what Siri can answer internally or with a simple search. That’s annoying, but you can eliminate the confirmation step in Settings > Apple Intelligence & Siri > ChatGPT > Confirm ChatGPT Requests.At least ChatGPT got the general area of Settings right.
• On my M1 MacBook Air, triggering Siri by clicking its Dock icon brings up Type to Siri, and I have to click the microphone button to be able to dictate to it. That’s a change: on my 27-inch iMac, which doesn’t support Apple Intelligence, clicking the Siri button causes Siri to listen to the microphone instantly.
• On the iPhone, you can invoke Siri by holding the side button or using “Hey Siri.” However, if you want to continue the conversation with ChatGPT, you may find that Siri doesn’t always listen while its splooshy animation jiggles around the edge of the screen. Holding the side button down was more reliable but more awkward. Similar animations appear at times when you can’t dictate, too, so you can’t assume fancy graphics mean Siri is listening.
• Depending on what words you use, Siri may give you a seemingly random response or provide Web search results rather than allowing you to engage with ChatGPT. Those random responses may even come during a discussion with ChatGPT. To ensure your prompts go to ChatGPT, say its name somewhere in your prompt.
  
• Although you can talk to ChatGPT using Siri, its responses always come back as text. That’s fine in many cases, but anyone accustomed to ChatGPT’s Advanced Voice Mode (where ChatGPT provides spoken responses) will be disappointed.
• The longer your prompt to ChatGPT, the more likely it is that Siri will stop listening at some point and send whatever it has up to that point. If you think it’s irritating when people interrupt you while you’re speaking, just wait until Siri indicates it’s bored with what you’re saying.
• There’s no way to review the ChatGPT transcript to refer to previous responses—you can see only the last response. However, if you sign into your ChatGPT account in Settings > Apple Intelligence & Siri > ChatGPT, you can view the full transcripts of all your chats. (Signing into my ChatGPT account also fails on the M1 MacBook Air. Cursed, I say!) At least you can delete abortive transcripts that you mistakenly triggered with Siri.

Deeper Concerns

As frustrating as these issues are, I have deeper concerns. Is Siri a good way to interact with ChatGPT? Will increasing the number of ways Siri can mess up reduce our desire to use it? We hoped Siri’s Apple Intelligence enhancements—and particularly the ChatGPT integration—would make Siri less frustrating. Might the reverse be true?

Apple markets Siri as a digital assistant, capable of carrying out simple commands and performing highly directed searches. In my experience, Siri works fairly well for playing music using artist names, controlling HomeKit devices, setting timers, and making reminders. Some searches, such as asking about tomorrow’s weather, also work reasonably well.

But using Siri to trigger Web searches is frustrating, particularly if you become accustomed to using Siri on a HomePod. Such prompts usually generate, “I’ve found some Web results. I can show them to you if you ask again from your iPhone.” rather than a useful response. Also, although Apple has made slight improvements in Siri’s ability to maintain context in a conversation of late, we have 13 years of experience in failure with anything but single, separate commands. If Siri heads off down an incorrect response path, our only recourse is to shut it up with, “Hey Siri, stop,” and then issue a differently worded request rather than redirecting the conversation.

In contrast, ChatGPT cannot carry out commands of any sort, and it’s not a search engine, although OpenAI recently gave paying ChatGPT Plus users access to such capabilities. Much has been written about how generative AI systems get facts wrong and make things up, and that’s not wrong—if you want to search the Web, use a search engine. ChatGPT is far more valuable for analyzing data, creating content, and exploring unfamiliar topics. It’s designed for conversation, with follow-up queries, comments, and additional information necessary for optimal results. Siri and ChatGPT simply don’t do the same sort of things.

Apple doesn’t want us to anthropomorphize Siri, but that’s nearly impossible when speaking to a digital assistant that responds with a natural-sounding voice. So when Siri responds randomly, sometimes stops listening to you before you’re done speaking, and is generally a lousy conversationalist, it’s impossible to avoid the feelings of pique that a person with similar conversational traits would trigger. You wouldn’t keep trying with such a person, and many of us won’t keep trying with Siri either.

What’s up with Siri’s seemingly random responses? In my conversation above, Siri suggested I call emergency services, created a reminder, did a Web search, thought I was asking for driving directions, and was just generally confused a few times. Yes, my prompts were an attempt to speak naturally, but isn’t Siri supposed to be able to handle that now?

If Apple Intelligence is going to improve Siri, it has to understand what’s being asked and do something sensible. In the past, Siri’s failure mode was mostly binary—it either did what you wanted or failed in a predictable way. With Apple Intelligence, Siri seems primed to fail in ever broader and more unpredictable ways, which could reduce our enthusiasm for using it for even simple tasks.

Apple will undoubtedly keep working on Siri, but I worry that it will be too little, too late. From experience, I know that I’m unlikely to retry a particular task with Siri after failing enough times, as has been the case with trying to add text to a note in Notes.