To transcribe a long recording and ask questions about it inside ChatGPT or Claude, install Audioscrape in the assistant, send it the recording, and ask. In ChatGPT you attach the file to the chat; in Claude you paste a Google Drive, Dropbox or OneDrive share link. Audioscrape transcribes it with the speakers separated and every passage timestamped, and the assistant answers from the transcript with the speaker and a link to the moment. The recording stays in your private Audioscrape workspace, so next week you can ask across every recording you have sent, not just the one in front of you.
Why connect a transcription service instead of pasting text
An assistant on its own works with what is in the current conversation. A two-hour interview is a lot of text to paste, a folder of them is more than any chat holds, and pasted text has no speaker names or timestamps to check a quote against. With Audioscrape connected the transcripts live outside the chat, the assistant searches them when you ask, and each answer points back to the second in the audio.
How to do it
1. Install Audioscrape.
- ChatGPT: open Plugins in the sidebar, search for Audioscrape and click Install. No paid ChatGPT plan is needed. Setup guide.
- Claude: add Audioscrape from the connector directory. Setup guide.
Then sign in to Audioscrape when the assistant asks (Google or Microsoft is one click). The free plan is enough to start.
2. Send the recording.
- In ChatGPT, attach the audio or video file to your message and say “Transcribe this with Audioscrape.”
- In Claude, Claude cannot hand an attached file to a connector. Share the recording from Google Drive, Dropbox or OneDrive with “anyone with the link” and paste the link: “Transcribe this recording: (link).”
- Either way, a direct link to an audio or video file works too, and so does uploading on audioscrape.com: it shows up in both assistants. Links to web pages (a YouTube or TikTok page) are not files and are refused.
Accepted formats: mp3, wav, m4a, ogg, flac, aac, opus, mp4, mov, webm and mkv. Sending the same file twice does not transcribe it or count it twice.
3. Wait a few minutes. The assistant checks the status for you; most recordings are done within minutes, and long ones take longer. You get an email when a long one finishes.
4. Ask. Some questions that work well:
- “Summarize what each speaker argued, with timestamps.”
- “Where did they talk about the budget? Quote it.”
- “List every action item and who took it on.”
- Later, in a new chat: “Across all my interviews this month, who mentioned supply problems?”
Every answer cites the passages it used, with the speaker and a link that opens the audio at that moment.
5. Name the voices once (optional). On audioscrape.com, open a transcript and name the people who recur. Later recordings come back with those names, so “what did Maria say” works across everything. How speaker names carry over.
What you get
- Long recordings handled: paid plans have no per-file limit, and the free plan takes files up to 3 hours.
- Answers grounded in the transcript, with speaker labels, timestamps and a link to the audio for each passage.
- An archive that grows with every recording you send, searchable from ChatGPT, Claude, the website and the API alike.
- The same assistant can search public podcasts, council meetings and court hearings in the same question when you want that.
- Your recordings stay private to your workspace and are never used to train models without your permission.
Limits and honest caveats
- Free includes 5 hours of transcription a month. Starter ($35 a month) includes 20 hours, Pro ($129 a month) 100. See pricing.
- The assistant decides when to use a connector. If it answers without Audioscrape, start your message with “@Audioscrape” in ChatGPT or “use Audioscrape” in Claude.
- Speaker names are machine labels and can be wrong, and audio quality limits accuracy. Check the linked moment before you quote someone.
- Recording people can require their consent. Send only audio you have the right to process.
Related: How do I search hours of my own recordings by what was said and who said it?