I Tried ChatGPT in CarPlay, and It Immediately Hallucinated

Back in February, we learned that Apple had plans to integrate AI assistants into CarPlay. The news came along with the first beta for iOS 26.4, though Apple didn't actually include a way for testers to try out the integration in the software. Instead, the company confirmed the news in its CarPlay Developers Guide, stating that "CarPlay voice-based conversational app" support would require iOS 26.4 at a minimum.

Well, Apple launched iOS 26.4 last week, and while the update delivered a number of new features, none of them were AI chatbot integrations with CarPlay. The company even dropped the first beta for iOS 26.5 without including a mention of the upcoming feature. I'll admit, the idea of using ChatGPT, Claude, or Gemini with CarPlay totally fell off my radar, instead replaced by thoughts of when Apple's big AI Siri upgrade would actually launch. But OpenAI didn't forget about it.

ChatGPT is now available in CarPlay

In a surprise move, OpenAI is now rolling out ChatGPT support in CarPlay. The company announced the news in a Thursday post on X, with the following tagline: "The voice mode you know, now available on-the-go." OpenAI confirmed in the post that the feature works with iPhones running the latest version of iOS (iOS 26.4)—and in cars that support CarPlay, of course.

I happen to have a car with CarPlay, so I had to give the feature a try. My iPhone is already running iOS 26.4, so that wasn't an issue, but I did need to update the ChatGPT app to its latest version. Before I did, I couldn't see the app in CarPlay. Following an update, ChatGPT appeared on the last app page in my CarPlay window.

I tried it out, and it works just about as you'd expect: When you launch ChatGPT in CarPlay, it opens directly in voice mode. (You are driving, after all.) There are only two buttons on the screen: "End," which ends the session, or "Mute," which mutes the session, presumably so you can talk to other people in the car (or swear at other drivers?) without ChatGPT listening in.

Having launched the app, you can simply talk to ChatGPT, and it can respond. It really is a driving-friendly version of familiar voice mode.

CarPlay doesn't stop ChatGPT from hallucinating

I don't really have many more takeaways here, other than this: In my very short time testing the feature, the AI began hallucinating almost immediately. I asked if it knew what I was doing, and it said it didn't. When I pressed that I thought it'd be able to guess given the context, it admitted it did know I was using CarPlay, so I was probably driving. I asked how long this had been a feature, and it mistakenly thought I was asking about CarPlay, so it said 2014. When I corrected it, and asked how long ChatGPT integration had been a feature in CarPlay, it said late 2023.

"Late 2023?" I asked. "Yeah, exactly!" responded ChatGPT. "So it’s still pretty recent, but it’s been expanding a lot since then, especially with updates from OpenAI and Apple. So, yeah, you’re kind of right at the beginning of that wave!" I asked if it knew what year it was right now. ChatGPT conceded that it was 2026, which meant it had been "a few years since that initial rollout," and thanked me for pointing that out.

"Okay," I said. "But I was under the impression that this feature is brand new, like it literally just came out today." This marked the first time in this conversation that ChatGPT had to "think," signaled by a subtle tune that played in the background. When it got back to me, it confirmed the feature launched with iOS 26.4. "That’s the first time third-party chat apps like ChatGPT became available in CarPlay, so it’s super new!" When I asked why it said 2023 originally, it said that was a mistake. When I pushed it to tell me what source it was pulling that information from, it said it "must have been recalling an earlier timeline by accident."

This is not a new problem. All generative AI hallucinates, not just ChatGPT, and it's something to expect when using the technology. But I can't help but think that CarPlay only makes the problem worse: You can't easily fact-check the AI yourself when you're out on the road; all you can do is ask it to fact-check itself. Sure, it might correct the mistake, like it did when I challenged it, but it may just hallucinate the correction, too. If you rely on this tool to answer your questions while on the road, you might just get fed some misinformation that you file away as truth.

Disclosure: Lifehacker’s parent company, Ziff Davis, filed a lawsuit against OpenAI in April, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.

View the full article