Using Video Transcripts for Research and Citation

Transcripts as primary and secondary sources

In research methodology, the original video is the primary source. A transcript derived from that video is a secondary source — it represents the primary source through a layer of processing. For most research purposes, this distinction matters: if you quote from a transcript, the authority behind the quote is the original spoken content, not the transcript file.

This has practical implications. If the transcript contains an error — a misheard word, a disfluency removed by the AI model, a name misspelled — and you quote from it without verifying against the original audio, you are attributing words to the speaker that they did not say. For research that makes specific claims based on quoted language, every direct quote should be verified against the original source.

When AI-generated transcripts are sufficient for research

Most research use of video transcripts does not involve direct quotation. Transcripts are used for:

1. Content identification. Scanning a large body of video content to identify which videos are relevant to a research question. For this purpose, 90-95% accuracy is entirely functional.
2. Coding and thematic analysis. Qualitative research that identifies themes and patterns across transcripts. Minor transcription errors do not materially affect thematic coding unless they cluster around key terms.
3. Background research. Understanding the content of a video interview or presentation before conducting follow-up research. Working-document accuracy is sufficient.

For these purposes, AI-generated transcripts are standard research tools. The methodological note to include in research documentation: transcripts were generated using AI speech recognition; direct quotes were verified against original audio.

Citation formats for video content

Standard citation styles handle video sources. The transcript is generally not cited as a separate source — the original video is cited, with timestamp notation in the in-text reference.

APA 7th edition:

In-text: (Speaker Last Name, Year, 00:14:32)

Reference list:
Speaker, A. (Year, Month Day). Title of video [Video]. Platform.
https://url-of-original-video

Chicago 17th edition (notes-bibliography):

Footnote: Speaker Name, "Title of Video," Platform, Month Day, Year,
video, 14:32, https://url-of-original-video.

If you are citing your own transcript of a video rather than the original, note this in your methodology section: transcripts were produced by the researcher using [tool name] and verified against the original audio. Do not cite the transcript file itself as an independent primary source. The APA Style guide for audiovisual sources covers the exact format for citing videos with timestamp notation.

Storing transcripts for research integrity

Research transcripts should be stored alongside the original source material or with a permanent, verifiable link to the source. A transcript separated from its source video is unverifiable — other researchers cannot confirm its accuracy or access the primary source.

Best practices for research transcript storage:

1. Name files with the source identifier: youtube_abc123_2026-03-14_ai-generated.txt
2. Include a metadata note at the top of each transcript: source URL, date accessed, tool used, whether reviewed
3. Keep original video files or stable archival links alongside transcripts
4. Mark edited transcripts as reviewed versions distinct from the raw AI output

For large-scale research involving many transcripts, the transcript archive guide covers searchable storage and organisation for ongoing research use.

Qualitative research: coding transcripts from video interviews

Qualitative researchers working with video interview data use transcripts as the primary coding medium. The standard workflow: transcribe all interviews, import transcripts into a qualitative analysis tool (NVivo, ATLAS.ti, or a spreadsheet approach), and code passages by theme. Transcripts enable thematic analysis at scale without rewatching every interview multiple times.

Accuracy requirements for qualitative research depend on what you're analysing. Discourse analysis needs verbatim accuracy including disfluencies (um, uh, false starts). An AI transcript that produces clean readable text has removed exactly the features a discourse analyst needs. For thematic content analysis, where the question is what ideas were expressed rather than exactly how, 90-95% accuracy is sufficient for the initial coding pass with spot-verification of quoted passages.

The key methodological disclosure is stating in the methods section which tool generated the transcripts and whether they were human-reviewed. Reviewers need to know the accuracy level of the primary data to assess the research. A methods section that omits the AI origin of transcripts is not standard practice.

Citing social media video content in academic work

Social media videos as academic sources require careful handling because they can be deleted or modified after you access them. Standard citation practice for online sources recommends including the access date. For video specifically, including a timestamp for any direct quote helps readers locate the relevant passage even if they access the same video later.

If a video is likely to be deleted — political statements, controversy-related content, content from accounts at risk of being removed — archive it through the Internet Archive's Wayback Machine or download a local copy at the time of access. Citing both the original URL and an archived version is best practice for research that will be reviewed over time. The transcript file itself, with a metadata header showing access date and source URL, is a form of archival record.

One practical note on APA 7th edition: it cites the original video, not a transcript file. In-text citations include the timestamp for the quoted passage — (Author, Year, 1:23:45). If you're quoting from a transcript you generated, note in the text that the quote was derived from an AI transcript you reviewed, and verify the exact wording against the original audio before the paper goes to submission.

The recommended practice when using video transcripts as sources in formal writing: keep both a raw AI output version and a reviewed version as separate files, named clearly to indicate which is which. This matters for audit purposes — a reviewer or editor who asks whether a quoted passage was verified needs to see a clear record of which version was relied upon at time of writing. A single file labelled only with the source URL does not provide that clarity. In my own research practice, I append _raw and _reviewed to every transcript filename — a simple convention that has saved significant confusion on more than one occasion when returning to a project after a gap of several months.