Search your interviews, voice notes and research chats in one place

Research that lives in five places

You know a source said something perfect about her grandmother's shop. Was it in the second interview or the fourth? The structure for chapter six came to you on a walk and went into a voice memo. The historical detail you checked last spring is somewhere in a chat with an AI assistant. Each piece sits in a different app, and none of them searches the others.

Quotes add a stricter need. Paraphrasing from memory is risky; you need the exact words, in context, from the recording.

Revision makes it worse. Months after the interviews, an editor asks where a claim came from, and you need to show the source said it, not that you remember them saying it.

How writers keep it together

What helps is one index that holds recordings and text together, so a single question covers all of it.

• Transcribing every interview. Makes quotes searchable, at the cost of time or a transcription bill, and cloud services mean uploading what sources told you.
• A notes system. Scrivener binders, index cards or a Markdown vault. Excellent for what made it into notes; recordings stay outside.
• Scrolling back through chat history. AI apps keep your conversations, but finding one idea among hundreds of threads is hard.

One private research library

See searching ChatGPT history, searching Claude conversation history and searching Obsidian notes with AI for the import details.

1. Add your interview recordings and voice notes. MediaFind transcribes them on your computer with word-level timestamps, and a names and terms list keeps people and places spelled right.
2. Bring in your notes: a folder of Markdown notes, such as an Obsidian vault, or plain text files, through Memory.
3. Import your AI conversations from a ChatGPT or Claude data export, which you request from that service's settings. Re-importing later skips what hasn't changed, and chats and notes don't count toward the free plan's 10 files.
4. Search by meaning across everything: “the grandmother's shop on the corner” returns interview moments, notes and chat messages together.
5. Ask a question such as “what did I find out about the 1952 flood?” The answer lists the messages or moments it used, or says your material doesn't cover it.

Getting quotes right

Treat a transcript as a pointer, not the record. Search takes you to the moment; play it and check the exact wording before it goes in print, especially with accents, crosstalk or names. Speaker labels separate you from the interviewee, and naming a source's voice once carries the name into their other interviews; unsure matches stay unnamed.

When you need the text, export the transcript as Word, PDF or plain text. Per-file summaries and chapters help you re-orient in a two-hour interview you haven't heard in a year.

Sources, consent and privacy

Interviewees should know they are being recorded and how the material will be used, and recording rules differ by place. Some sources speak on condition of anonymity; keeping their recordings off third-party servers is part of honoring that. MediaFind transcribes, searches and answers questions on your computer, and the on-device language model behind Ask sends no prompts anywhere.

Memory can also forget on request: delete a conversation or reset a whole library when a project ends. If you connect an AI assistant over MCP, the results it asks for go to that assistant, and on to its provider if it is a cloud AI, so leave that off for sensitive projects.

Frequently asked questions

Can I search my ChatGPT or Claude conversations next to my interviews?

Yes. Memory imports a ChatGPT or Claude data export, and chat messages turn up in the same search as your transcribed recordings and notes.

Can it read my Obsidian vault or other notes?

Yes. Memory imports a folder of Markdown notes such as an Obsidian vault, plus text files and pasted chats. Chats and notes don't count toward the free 10-file limit.

Are transcripts accurate enough to quote from?

Use them to find the moment, then check the wording against the audio before publishing. Accents, crosstalk and names are where transcripts most often slip.

Is my research uploaded anywhere?

No. Transcription, search and Ask run on your computer. Only an AI assistant you choose to connect over MCP receives the results it asks for.