Research Guides
Test the Captions on an Embedded Research Video
A caption file in the media folder does not tell you what the reader sees. Check the published embed with the sound off, using the evidence the article asks readers to verify. The task is to make that evidence understandable at the moment it appears.
1. Start with the released player, not the editor
Open the article in a signed-out browser when the page is meant for the public. Start on the exact embed and viewport readers get. Turn on the intended caption language if the player has a toggle; if the words are permanently visible in the video, check that version as-is. Record the page URL, the video revision or public identifier, the language and the date checked.
W3C’s explanation of WCAG 2.2 Captions (Prerecorded) describes the Level A requirement for prerecorded audio in synchronized media, with an exception for clearly identified media alternatives to existing text. Captions need to convey more than dialogue where the soundtrack contributes other information. A separate transcript can be useful, but do not count it as synchronized captions for a video that adds information.
2. Review the evidence with the sound off
Make a list of the segments the article cites, plus any moment where an audio cue changes how to interpret the screen. For example, imagine a fictional competitor demo that reads a monthly price aloud, then uses a warning chime to flag a rejected selection. The monthly qualifier and the warning are both part of the evidence. Check that the words appear over the right action, that a speaker change is clear when it matters, and that the warning is represented without requiring the reader to hear it.
Then replay with the original audio and compare what the captions show with what the soundtrack actually says. Flag any changed number, name, date, negation or condition. If the captions cover a needed label or footnote, verify placement on the narrow player too. A correct caption can still make a comparison harder to check if it hides the evidence underneath it.
3. Record defects someone can reproduce
Use one row per finding: time range | observed audio or action | caption shown | consequence | owner and retest date. For example: 00:42–00:48 | presenter says $19 per seat, billed annually | caption omits “billed annually” | readers cannot verify the term | video editor, then retest the published embed.
After replacing a track or editing a video, check the same segments in the article again. Do not mark a defect closed just because the caption file opens. For a separate worksheet on whether a linked video still supports a written claim, see the video citation check. This test covers caption usability for the reviewed embed; it does not certify overall conformance or prove a claim true.