Citing a Reel in APA and MLA: What to Include

What the transcript hands you

Our endpoint returns six fields: success, transcript, language, engine, fileSize and fileName. For citation purposes that means you get a block of words and nothing else. There are no timestamps, no speaker labels, no confidence scores and no line breaks. The 152-character result we got today held 0 newline characters, so you cannot point a reader at a line the way you would with a page number.

The name is the weak field

We built a test sentence with the four things a citation needs - an author, a date, a number and a title - and had it spoken by a text-to-speech voice:

"Doctor Amara Okafor published the report on March fourteenth twenty twenty four. It lists three hundred forty two cases across twelve clinics in Ohio. The study is called Water Quality in Rural Towns."

Here is what came back from the 36,080-byte file: "Dr. Anaro Kaffir published the report on March 14, 2024. It lists 342 cases across 12 clinics in Ohio. The study is called Water Quality in Rural Towns."

• Author: spoken "Amara Okafor", returned "Anaro Kaffir" - wrong
• Date: spoken "March fourteenth twenty twenty four", returned "March 14, 2024" - right
• Count: spoken "three hundred forty two", returned "342" - right
• Count: spoken "twelve", returned "12" - right
• Title: "Water Quality in Rural Towns" - right, capitalisation and all

We ran the same file a second time and got the identical text, wrong name included. Retrying will not fix a name. Uploading the same audio as a 192,466-byte 44.1 kHz stereo .m4a gave the identical text too, so the error is not in the encoding.

Where it gets worse

The same audio resampled to 8 kHz and saved as a 12,284-byte file came back like this: "Dr. Anar, those preferable three-port on March 14, 2024, it was 342 cases across the albunis in Ohio. The study is called what it's got all of you in rural towns."

The date and the 342 survived even there. Everything around them did not. If your clip is thin, muffled or heavily compressed, expect the same shape: numbers hold, names and connecting words fall apart.

What to take from the reel, not the transcript

APA and MLA both want an author, a date, a title and a source. Only the date is safe to lift from the transcript, and even that should be checked, because the model writes dates in its own format.

• Author or account name - copy it from the reel, never from the transcript
• Post date - read it off the post; the transcript gives you a date only if someone said one out loud
• Title or first line of the caption - from the post
• URL - from the post
• The quotation itself - this is the one thing the transcript is for

No timecode means no timestamp in the citation

APA asks for a timestamp when you quote a specific moment in a video. Our response carries no timing data at all, so you cannot produce one from the transcript. If your style guide wants "1:12", you have to find that position yourself by playing the clip, or cite the reel as a whole without one.

Check the quote before you paste it

Read the quotation against the audio once. Our test sentence returned a clean, grammatical, completely wrong name, and nothing in the response marked it as uncertain. A transcript that reads smoothly is not a transcript that is right.

If the reel is longer than a minute, run it through twice in two different encodings and diff the two texts. The words that change between runs are the words the model was unsure about, and those are the ones worth checking by ear before they end up inside quotation marks.

One limit worth stating plainly: we ran one synthetic voice reading one sentence. We do not have a labelled set of real reels to measure, so we are not putting a percentage on how often names come back wrong. What we can say is that it happened on both runs of a clean, full-quality file.