Glossary
Transcripts and LLM context windows
Updated · By the WithTranscript team · How we checked
Transcripts and LLM context windows: A context window is how much text a language model can read at once, measured in tokens; a transcript has to fit in it (with room for your question and the answer) for ChatGPT or Claude to work with the whole video.
A token is a piece of a word. OpenAI gives the rule of thumb that one token is about four characters, or three quarters of an English word. People speak roughly 130 to 160 words a minute, so an hour of talk is about 8,000 to 10,000 words, or roughly 11,000 to 13,000 tokens.
That fits comfortably in the context windows of current ChatGPT and Claude models, so for a single video the limit is rarely the window itself. It becomes one with several long videos at once, a whole lecture series, or a free plan that caps the length of what you paste.
What helps more than raw length: chapter headings (so the model can find sections), timestamps (so it can cite a moment and you can check it), and a header with the title and source link (so the answer can name where it came from). Long blocks of uncased, unpunctuated caption text are the hardest for a model to quote accurately.
How WithTranscript handles it
WithTranscript's Markdown copy and transcript.md URL include the title, source link, chapter list and paragraph timestamps for this reason. See the YouTube to ChatGPT guide.