ChatGPT Live Voice: The Voice That Sounds Almost Human and Already Searches in Real Time

Editor's note: This article has been expanded with historical context, concrete data from sources, impact analysis, and relevant comparisons, while maintaining the original structure.

What happened?

OpenAI has updated ChatGPT's live voice mode (Live Voice), integrating multimodal capabilities that allow the assistant to listen, speak, and perform internet searches simultaneously. According to ZDNet, which tested the feature exclusively, the experience is "almost human" (ZDNet, 2025). The update is rolling out gradually to ChatGPT Plus subscribers ($20/month) and requires manual activation in the app settings. This functionality represents a significant advance over the previous voice mode, which could only process one input at a time (audio or text) and did not access the internet in real time. Now, the underlying GPT-4o model combines low-latency speech recognition, natural speech synthesis, and web search into a single stream, allowing interruptions, pauses, and corrections without losing the thread of conversation.

Why is it important?

This capability marks a qualitative leap in human-machine interaction, comparable to the shift from one-way voice assistants (like early Siri or Alexa) to bidirectional, contextual conversation. Historically, voice assistants had to alternate between listening and processing: the user spoke, the system processed and responded, then listened again. This cycle introduced latency and limited fluency. With the new update, ChatGPT can maintain a continuous conversation while consulting up-to-date information, bringing interaction closer to natural human conversation, where we often pause, interrupt, or speak while thinking. According to ZDNet, testers noted that the assistant could verbally confirm a request while searching the web, something that previously required an explicit pause. This advance is part of the trend toward more intuitive, less screen-dependent user interfaces, with direct implications for accessibility for people with visual or motor impairments.

What consequences will it have?

For users: Productivity could increase significantly by being able to make complex queries without needing to rephrase questions or wait for sequential responses. For example, a student could ask: "Explain the theory of relativity while searching for the latest LHC experiments," and receive a contextualized explanation with updated data. However, the "almost human" feeling may create unrealistic expectations about AI capabilities, leading users to attribute intentions or emotions it does not possess. Additionally, real-time search introduces risks of misinformation if the model does not properly verify sources. OpenAI has implemented citation mechanisms and content filters, but accuracy depends on the quality of web results and the model's judgment in selecting them.

For businesses: This opens the door to applications in customer service (live support with access to knowledge bases), education (virtual tutors that research while explaining), and personalized assistance (travel planning or shopping with up-to-date information). However, infrastructure costs are high: each multimodal query consumes more computational resources than text-only interactions. This could limit mass adoption or translate into higher prices for users.

Comparison with previous events: This launch recalls the introduction of ChatGPT's voice mode in September 2023, which was an incremental advance. Now, the simultaneity of functions represents a leap similar to the integration of DALL-E into ChatGPT, which unified text and image generation. In the voice assistant market, competitors like Google Assistant and Alexa have incorporated web search, but not with the conversational fluency that OpenAI offers. The race for real-time multimodality is intensifying, with implications for privacy (audio is processed on servers) and regulation (the EU is already evaluating transparency requirements for conversational AI systems).

What should readers know?

• The feature is available only to ChatGPT Plus subscribers ($20/month) and is rolling out gradually. It is expected to reach free users in later months, though with limitations.
• It requires manual activation in the app settings (Settings > Beta features > Live Voice).
• Conversation quality depends on audio clarity (recommended: external microphone or quiet environment) and a stable internet connection (at least 5 Mbps).
• OpenAI recommends not sharing sensitive information during conversations, as audio is sent to its servers for processing and may be stored for model improvement purposes (according to the privacy policy).
• Internet search is performed via the Bing engine, so results may be biased or outdated if the model does not access verified sources.
• Users can disable web search at any time from settings, reverting to offline voice mode.

"It's the first time I feel like I'm talking to someone, not a program," commented a ZDNet tester. However, the same tester warned that occasionally the assistant made up data when the web search was inconclusive, underscoring the need to verify critical information.

In summary, the Live Voice update represents a milestone in the evolution of voice assistants, but responsible adoption will require users to understand its limitations and businesses to develop use cases that maximize benefits while minimizing risks of dependency and misinformation.