ChatGPT: New Dictation Feature -- Significantly Better Voice Input Now Available for All Users

You are sitting in a cafe, wanting to quickly dictate a question to ChatGPT -- and you get word salad. Background noise, chatter, the espresso machine: Until now, voice input failed in these situations. And when you spoke quietly because colleagues were sitting next to you? Even worse results.

What changed?

On June 26, 2026, OpenAI rolled out a completely overhauled speech-to-text model for the dictation feature in ChatGPT. The update was applied automatically -- you do not need to change any settings. And the best part: It applies to all plans, including the free tier.

The key improvements in detail:

1. Reliable recognition in noisy environments

Voice input now works significantly better in cafes, open-plan offices, public transport, and anywhere background noise used to be a problem. The model filters out ambient sounds much more effectively and focuses on your voice.

In practice, this means you can now use ChatGPT productively via voice on the go -- while commuting, during lunch at a busy restaurant, or in a waiting room.

2. Quiet and whispered speech is recognized

You are at the office and do not want everyone to hear what you are asking ChatGPT? Previously, you had to choose between privacy and functionality. The new model now correctly recognizes quietly spoken and whispered input.

This finally makes voice input practical in situations where you cannot or do not want to speak loudly.

3. Better recognition of letter-number combinations

Addresses, serial numbers, order numbers, IBAN numbers -- anything that mixes letters and numbers was often transcribed incorrectly before. The new model recognizes these combinations much more accurately.

Example: 'Order number A-7-3-2-1-B' or 'zip code 80331' are now reliably recognized instead of being interpreted as individual spelled-out characters.

4. At least 10% lower error rate

OpenAI states that the Word Error Rate has dropped by at least 10% for the most-used languages. This sounds small but makes a noticeable difference in longer dictations: less post-correction, fewer misunderstandings, smoother workflow.

5. Improvements across many languages

Besides English, Japanese, Korean, Chinese, Vietnamese, Spanish, French, Italian, and Portuguese have been improved. Particularly relevant for German-speaking users: recognition of accented English has also been optimized -- so when you dictate in English with a German accent, ChatGPT now understands you better.

Practical use cases:

- Commuting: Quickly dictate a question on the train or bus instead of typing on a small screen
- Cooking: Hands full, but you need a recipe or want to convert a measurement -- just ask
- Brainstorming: Capture ideas by voice while walking. ChatGPT will structure them for you afterward
- Dictating emails: 'Write a professional reply to this email' and summarize the content by voice
- Notes: Enter meeting thoughts directly into ChatGPT by voice right after the conversation and have them organized
- Accessibility: For people who have difficulty typing, voice input becomes a real alternative

How to use voice input:

1. Open ChatGPT (app or browser)
2. Tap the microphone icon next to the input field
3. Speak your prompt -- you will see the transcription in real time
4. Tap 'Send' or wait for automatic recognition to end

Tip: You can also use the Voice Chat mode (headphone icon) if you want to have a real conversation with ChatGPT -- the improvements apply there as well.

Privacy note:

Important to know: Your voice data is transmitted to OpenAI servers and processed there.

- Free plan: Your conversations may be used by default for training OpenAI models. You can disable this in Settings under 'Data Controls'
- Plus/Team plan: Same default setting, same opt-out option
- Enterprise/Edu plan: Data is not used for training and is subject to stricter privacy policies
- GDPR: Processing takes place on US servers, which is relevant for sensitive or personal content. Do not dictate confidential information such as passwords, health data, or customer data
- Recording duration: The voice recording is not permanently stored as an audio file after transcription -- only the transcribed text remains in the chat history

Comparison with alternatives:

- Claude: Currently does not offer voice input directly in the app. However, you can use your operating system's dictation feature (iOS, Android, Windows) and paste the text
- Gemini: Also has voice input, but recognition in noisy environments is less robust than ChatGPT's new model
- Apple Dictation / Google Voice: These system-wide dictation features work with any app but are not optimized for AI prompts and struggle with longer, complex inputs

Sources: help.openai.com/en/articles/6825453-chatgpt-release-notes, developers.openai.com/blog/updates-audio-models