OpenAI Support Bot Fails Its Own Product Test
Artificial intelligence is often touted as the future of customer support, promising efficient and instant help. However, if OpenAI's own support for its flagship product, ChatGPT, is any indication, that future might be more frustrating than helpful.
Recently, an experience with the ChatGPT app's support function revealed a surprising lack of knowledge about the application's own features, leading to a confusing and unhelpful interaction for a paying customer.
ChatGPT's support bot hallucinates a feature that doesn't exist in the app. Photo: Tiernan Ray/ZDNET
A Simple Bug Leads to a Support Nightmare
The issue began with a recurring bug in the iPad Pro version of the official ChatGPT app. For a user of the paid ChatGPT Plus subscription, which costs around $21.78 a month, the app would freeze completely when its window was resized to a full-screen or near-full-screen view. The app would become entirely unresponsive, with interface elements stretching and failing to refresh. This problem was easily reproducible on multiple iPad Pro devices.
Seeking a solution, the user turned to the in-app Help Center. After standard troubleshooting articles proved useless, the next step was the chat function, which is powered by an automated ChatGPT bot.
The Support Bot's Hallucination
Initially, the support bot suggested generic fixes like force-quitting and reinstalling the app, none of which worked. Then, the conversation took a strange turn. The bot confidently suggested, "Send Feedback in the App," and elaborated, "You can report this bug directly from within the ChatGPT app (usually under Account or Support > Report a Problem/Send Feedback)."
The problem? This feature does not exist. There is no "Report a Problem" or "Send Feedback" option within the app's support section. The only reporting function available is for flagging harmful or illegal content within a specific chat, not for reporting software bugs.
When confronted with this fact, the bot politely conceded the error, stating, "Thank you for pointing that out -- you're correct. The ChatGPT iPad app currently only allows reporting individual messages or chats."
Automated Emails Repeat the Error
The frustration continued when the user escalated the issue to email support via an address found on the OpenAI developer forum. An automated response repeated the exact same incorrect advice, once again directing the user to a non-existent bug reporting feature within the app. Just like the chatbot, the automated email system acknowledged its mistake only after being corrected.
OpenAI Confirms The Missing Feature
Contact with a human spokesperson at OpenAI finally brought clarity. They confirmed via email that there is no dedicated bug reporting feature for consumers in the ChatGPT app.
"We don't currently offer a dedicated consumer-facing bug reporting portal for the ChatGPT app. People can find troubleshooting steps and ways to share feedback through our official Help Center at help.openai.com."
The spokesperson mentioned a formal bug bounty program for security issues and noted that user feedback is gathered through thumbs-down ratings on responses. They also acknowledged, "We're working to make the automated support replies more accurate and effective."
The absence of a standard bug reporting feature is surprising for major software, but the fact that OpenAI's own AI, presumably trained on its internal documentation, was unaware of this is particularly ironic. It highlights a significant challenge: if an AI can't get the facts right about its own product, its utility in a customer support role is severely undermined.
A Glimmer of Hope for AI Support
Interestingly, in follow-up tests, the support bot's behavior changed. It no longer hallucinates the bug reporting feature. Instead, it now correctly states that no such option exists in the app and offers to forward the problem description to the development team. This suggests that the feedback was received and the bot was updated.
While the initial experience was a showcase of AI support gone wrong, this update shows that these systems can learn and improve. At least that's progress.
