Claude and ChatGPT know a great deal and can search the web. What they do not have is your material: the lecture series you are working through, the interviews behind your piece, the talks you are drawing on for a book. Ask about those and the model guesses, and a guessed quote looks exactly like a real one.
For anyone who studies, researches or writes, the best setup is both: the model's knowledge and the web, plus your own sources, organised and cited, for the parts that matter. This article explains what changes when the assistant can read your sources, how to combine them with everything else or keep it to them, and how to set that up with INDXR.
What the assistant cannot see
A language model knows what it learned in training and what it can find on the web. For famous material that is often enough. A specific lecture, a niche podcast or an interview you recorded yourself is usually in neither, so the model has little to go on, and it fills the gap with something that sounds right.
Quotes are where this hurts most. A paraphrase that is slightly off is a nuisance. A quotation that was never said, attributed to a real person, is a mistake you can publish or hand in. The model does not mark the difference for you.
The fix: add your sources
The fix is old advice: when a claim matters, go to the source. For an assistant, that means it can open the transcript and point to a place in it, while still using everything else it brings: background, context, the wider literature, a web search.
You decide the mix per question. “Using my Early Middle Ages collection and what you know about the period…” combines the two. “Answer only from my Kagan lectures” keeps it to your sources.
Pasting a transcript into the chat works for one video. It stops working when the material is a course of twenty lectures or a year of interviews, and it loses where each passage came from. What you want is a library the assistant can search, read in pieces, and cite.
How INDXR's connector works
INDXR keeps every transcript you make in a library: YouTube videos and playlists, and your own uploaded recordings. Each transcript keeps its source and a ready APA citation, and a playlist is filed as its own collection. The connector lets Claude, Claude Code or ChatGPT read that library through the Model Context Protocol, an open standard for connecting assistants to outside tools. The assistant can:
- List your collections, with how many transcripts each holds.
- Search the library: titles, channels and the transcript text itself. Each hit comes with a short excerpt, its timestamp and a link to that moment in the video.
- Read a transcript in blocks of minutes, with timestamps and speaker names. If you corrected a transcript, it reads your corrected version.
- Look up the source: channel, link, length, language, publication date and an APA 7 citation.
- Read a summary you already made in INDXR.
That is all it can do. It cannot add, edit or delete anything, spend credits, or see your account. It sees only what it asks for to answer you.
Setting it up
In Claude, you add INDXR as a custom connector once, sign in to INDXR and allow read access, then turn it on in a chat. Claude Code takes one command and a sign-in. ChatGPT needs developer mode, on a paid plan, on the web. Each takes a few minutes; the exact steps and the connector URL are in the Claude and ChatGPT guide.
Three prompts that use it well
The pattern is the same in each: name the collection or the source, and ask for quotes with timestamps. That makes the assistant search and read your sources for the specifics, on top of what it already knows.
Study
“Turn lectures 6 to 10 of my Early Middle Ages collection into study notes. One section per lecture, key arguments as bullets, and the APA reference plus timestamps under each section.”
The assistant lists the collection, reads each lecture in blocks, and writes the notes. Because each section carries timestamps, you can jump to the moment a point was made when you revise. There is more on this in study notes from a lecture playlist.
Research
“Which of my sources discuss consent? For each, quote the passage with its timestamp and give the APA reference.”
Here the assistant searches across the whole library, not one video, and comes back with a list of sources and passages. It is a starting point for a literature review, not the review itself: read the passages it found in context before you rely on them.
Writing
“Draft three paragraphs on why the Moon programme was framed as a challenge, using only my Kennedy transcript. Quote him with the timestamp for each quote.”
Say you transcribed the 1962 Rice University address, which is in the public domain. An answer built from the transcript might include a line like this, illustrative only:
Kennedy framed the goal as a test rather than a convenience: “We choose to go to the moon in this decade and do the other things, not because they are easy, but because they are hard.” (Rice University address, at the timestamp the transcript gives, with a link to that moment.)
The point is not that the quote is famous. It is that the quote now comes with a place in the recording you can open, so checking it takes one click instead of a search through the whole speech.
Check the answer anyway
Reading from your sources makes an assistant far more trustworthy, not infallible. It can still pick a passage that does not quite support the point, merge two speakers, or summarise loosely. The timestamps are there so you can catch that quickly. A good habit: before a quote goes into anything you publish or submit, open it at the timestamp and hear it once.
For the reference itself, the citation the connector returns follows APA 7. Two details, title capitalisation and italics, need a quick touch by hand; see how to cite a YouTube video in APA.
What it costs
The connector is free on every INDXR account. What costs credits is making the transcripts: YouTube captions are free for a single video, and AI transcription, for videos without captions and for your own recordings, is 1 credit per minute. A free account includes 10 credits. There is no subscription, and credits never expire.
Start with one source
Make a free account, transcribe one lecture or interview you know well, connect Claude or ChatGPT, and ask it something you already know the answer to. If it finds the right passage and the right minute, you will know what it can do with the rest of your sources.