What changes with Live Dictation?
OpenAI describes a transcript that updates while you speak. You can toggle whether it is visible by tapping or clicking the waveform. OpenAI also says it removed the previous ten-minute recording limit. Start on the device you actually use and check whether text can be shown during a short recording. Note the account, platform and test date. A worldwide rollout can reach individual accounts at different times. If you only see the earlier dictation experience, repeat the same test note once live text appears. A short recording is enough for a workflow with clear review points.
Where should you review the transcript?
Dictation through the microphone in the message composer produces editable text for your message. Read the full transcript after stopping and edit it before sending. In a test, include two similar project names, a quantity and a negation, such as a fictional delivery that has yet to be confirmed. Write down the expected wording beforehand. Check that the draft input keeps the meaning you intended. For regular use, keep a short list of the fields someone should check in every draft, such as date, owner and status.
What does OpenAI document about the audio?
OpenAI's dictation FAQ says the recording is sent to its models for transcription. The FAQ says the audio is generally retained while the associated chat remains in history. If the chat is deleted, OpenAI lists a window of up to 30 days for the audio and describes exceptions in the FAQ. The same page distinguishes audio sharing for consumer users who opt in from the default for business users. Determine which account the workflow will use and what your own rules allow people to say on a recording. Fictional names and work items are sufficient to try it.
How can you measure the benefit at work?
Choose a recurring task, such as a short internal project handover. Prepare three fictional notes with different challenges. Include a number and unit in one, similar-sounding names in another, and an issue that must remain open in the third. For each note, record the time from starting to the reviewed text and count the required corrections. Repeat with your usual text input and the same notes. Apply the same acceptance checklist to both runs. A gain in speed is useful when both methods carry the important details through with comparable reliability.
What should the team sign off on?
An agreed checklist makes handover results easier to compare across the team. Start with one approved use case and identify who reviews the content. Agree which tasks will still be written directly and how someone should report a transcript error with a possible downstream consequence. If relevant, test the usual working conditions, including background noise, and record how results differ from the quiet test. Write down the devices and accounts covered by the approval. If you change devices later, repeat the acceptance checks with the same notes. That keeps the pilot limited to the setup your team has actually evaluated.
Keep it verifiable
Primary sources
- OpenAI – ChatGPT Release Notes vom 18. November 2026Source checked:
- OpenAI – Voice Dictation FAQSource checked:



