ChatGPT, like many chatbots, is presented as an extremely competent personal assistant. But of the many things that confuse him, one is particularly confusing: he can’t tell the time.
When I ask ChatGPT what time it is, I’m never quite sure what I get. Sometimes he tells me it can’t be done. “I don’t have access to the real-time clock on your device or your location, so I can’t give you the exact local time,” they wrote to me about a week ago at 4:15 p.m. Eastern Standard Time. “But me Down I know today’s date according to my system: 2025-11-20” (I assume it’s bolded by ChatGPT to make sure I haven’t missed what it’s doing right.) Sometimes it asks me to specify a city or time zone, but reveals that it can’t reliably check the time that way either – “It’s 12:42 in New York (EST, assuming the system clock is correct)” – ChatGPT wrote to me at 11:08. And sometimes that does give me the exact correct time until I ask a few minutes later and it gets it wrong again.
We’re not the first to bring this up. Time is often an issue on Reddit AND ChatGPT Forums. One user he insisted OpenAI to “bring attention to this” because it gives a “bad name” to an AI model “with cognitive abilities far superior to mine.” Features such as Internet Search offer some workarounds. Yet years after its launch, standard ChatGPT remains blissfully indifferent to the ticking of the clock – and while the situation may seem absurd, there is a straightforward reason for this.
Telling the time is easy on any computer or phone, thanks to the tiny chips that tick inside them. However, generative AI systems such as ChatGPT-enabled large language and visual models, Google’s Gemini, Anthropic’s Claude, and others are built for a very different purpose. By default, they accept user queries and predict responses based solely on training data. This does not include constant real-time updates such as the time, unless they specifically search the Internet for this information.
“The language model operates in its own space of language and words. It only refers to things that have entered this space,” said Yervant Kulbashian, an AI robotics expert who he wrote on the concept of time as understood by AI in 2024, until Edge. It’s like being shipwrecked on an island in the middle of the ocean, equipped with a huge collection of books, but no watch.
Why can’t OpenAI just build a bridge to this island and give ChatGPT access to the system time clock? The short answer is: maybe. When I spoke to Pasquale Minervini, who researches natural language processing at the School of Computer Science at the University of Edinburgh in Scotland, his desktop application ChatGPT immediately gave him the correct time in Milan, Italy—where he was during our interview. “It can tell the time if you give it access to a clock. Otherwise, it’s something that was kind of born in that moment,” he said. According to him, the current information was likely “embedded in the context” of the application. He had previously enabled the “Search” feature in the ChatGPT application, which means ChatGPT has permission to use the timing tools built into his computer, in addition to the Internet, to check the time.
OpenAI told us this. “ChatGPT-enabled models do not have built-in access to the current time, so to get the current facts, ChatGPT must sometimes call the search function to get the latest information,” spokeswoman Taya Christianson wrote to us. Edge.
Kulbashian said there are tradeoffs in keeping LLM informed about timing. ChatGPT has a finite space in the so-called context window, i.e. the part of the information “remembered” at a given moment. Every time ChatGPT checks the system clock, it adds information to this context window – to use another metaphor, imagine someone stacking a stopped clock on their desk every second. “If you start adding more stuff to your desk, eventually you’re going to have to start pushing it away,” Kulbashian said.
Updated often enough, clock clutter can simply mean noise for the AI system. “This could end up confusing the robot,” Kulbashian said. “If we were talking and then every once in a while someone would come over and say, ‘It’s 5:45.’ ‘It’s 5:46 now.’ In contrast, something like a date is relatively easy to include in a system prompt at the beginning of a chat – which visible leak from the ChatGPT system seems to show.
ChatGPT users can determine the time without much fuss by asking the chatbot to find it. (Some other chatbots, such as Google Gemini, will automatically find the time.) You can also use context protocol of the open source model to connect the AI application with your data. That said, sending AI models to crawl the Internet or allowing them to access personal data carries risks, such as injecting the bot with malicious prompts scattered across the Internet, Minervini said.
Minervini, whose research finds weaknesses in consumer AI technology, says there’s actually a whole list of time-related tasks he hasn’t mastered. He challenged leading AI models with photos of analog clocks and found that the models had difficulty reading the position of the clock’s two arms. The calendars, he told me, are “also weird.”
Perhaps a bigger problem for the average user is that ChatGPT cannot reliably explain what its limitations are. A human assistant who simply doesn’t know the time might be understandable; one that regularly lies knowing he would probably be fired. But of course big language models don’t lie – they just predict what you want to hear, as usual.
OpenAI’s Christianson said, “we continue to improve how consistently it knows when to do this.”