As of 13 August 2026, AI can transcribe an interview for your freelance client.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Confirm with your client that you have permission to upload and process the recording, and agree whether the delivered transcript should be verbatim, lightly edited or timestamped.
   2. Open a chatbot that accepts audio files, or a speech-focused AI tool, and check that the file format and recording length are supported.
   3. Gather the original recording, the correct spelling of names and any agreed speaker list, then remove unrelated files before uploading.
   4. Paste the prompt above, replace the speaker slots with known names where appropriate, and attach the recording.
   5. Save the generated transcript and its uncertainty list as a separate working copy rather than sending the first output to your client.
   6. Play the recording while comparing every quote, name, number, speaker label and marked unclear passage, correcting the transcript and adding timestamps where the model missed them.
   7. Format the checked transcript to the client's agreed style, remove private material that was not part of the commission, and send the final file through the agreed secure channel.

   Prompt

   Transcribe the attached interview recording for my freelance client.
   
   Requirements:
   - Transcribe the spoken words faithfully and do not summarise, rewrite or improve them.
   - Use clear speaker labels, using these names only if they can be identified from the recording: [SPEAKER 1], [SPEAKER 2], [SPEAKER 3].
   - Add timestamps at each speaker change and wherever a passage is unclear.
   - Mark unintelligible audio as [inaudible 00:00:00] rather than guessing.
   - Preserve important hesitations, repetitions and corrections, but omit obvious filler only if it does not change the meaning.
   - Mark non-speech sounds that affect meaning as [laughs], [long pause] or similar.
   - Do not invent names, words or context.
   - After the transcript, provide a separate list of every uncertain word, speaker attribution and timestamp for me to check against the audio.
   - Do not include an executive summary unless I ask for one.
   
   Before processing the file, state whether you can access the whole recording. If any section cannot be accessed or understood, identify the timestamp. Treat the recording as confidential and do not reuse its contents for any other purpose.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.