Don't use ChatGPT voice mode if you want accuracy - here's why (www.zdnet.com)

Users and reporters found that ChatGPT’s Voice Mode—both Standard and Advanced—produces noticeably less accurate and more stubborn answers than the web/chat interface. In an extended test, Voice Mode repeatedly misidentified iPhone 16 Pro Max hardware (inventing an in-display fingerprint sensor, missing a real Camera Control button) and insisted on wrong answers to a date/riddle puzzle, even after corrections. Reddit threads and the reporter’s experiments suggest Voice Mode intentionally “rushes” responses to keep conversational flow, and Advanced Voice Mode behaves differently (reportedly not honoring Custom Instructions) and can be worse than Standard. The author notes you can disable Advanced Voice Mode in settings and that they’ve contacted OpenAI for comment. Significance: this highlights a fundamental UX–safety tradeoff in voice agents—lower latency and more natural turn-taking can come at the cost of less internal deliberation and higher hallucination risk. For developers and researchers, it’s a reminder to treat voice interfaces differently from text models: consider explicit think-time, confirmation prompts, stricter grounding to knowledge sources, or slower response pipelines for high‑stakes queries. For end users, the practical takeaway is to prefer the web/chat interface when accuracy matters and to be wary of treating voice responses as authoritative.