The narrator tells employees to “use this link if that happens.” A box appears on the screen, and an arrow points to the employer’s alternate-format contact. The automatic captions reproduce the six words perfectly. A person listening without the picture still can’t learn what “this link” refers to; a person reading the caption on a noisy break room floor can’t tell whether “that” refers to the slide before or after the cut. Accessible training starts by locating the parts of a lesson that disappeared into gestures and graphics.
Use an employer-controlled example or an approved synthetic demo. Do not publish payment codes, live accounts, barcodes associated with a worker’s card, screenshot notifications or contact lists with personal information. A program name that appears in an employer’s example is context, not proof that a pictured vendor feature is still available or accessible.
Audit the script before making a caption file
Read the narration aloud without watching the screen. Mark any moment you hear “here,” “below,” “this page,” or “where the button turns green.” Replace unnecessary visual references with descriptive language: “Open the payroll notice, then choose the link labeled ‘Ask HR for an alternate format.’” Where the video must demonstrate movement, position and result, explain what changes. A listener should be able to map that statement back to the task without somebody describing the same slide again in person.
Make the recording about a task, not a gallery of menus. If there are separate paths for an employer notice and for an official provider channel, name that boundary in the narration. A sample record with fictitious names and dates makes it easier to pause over the correct step without exposing anything. Avoid dense superscripts and callouts that require the viewer to freeze each frame. When a visual fact matters, decide whether integrated narration will carry it, or whether a separate audio-description version is needed. W3C’s accessible-media guidance explains the different roles of narration, captions, description, transcripts and the player.
Captions are a timing and meaning check
Compare captions with speech across the whole video, including the short opening and close. Watch a second time with the sound off. Are speaker changes clear when they matter? Are meaningful sound cues included, such as an alert that signals the example has moved to the next step? Do captions stay on screen long enough to read and not cover the link or label that the speaker wants people to find? W3C’s caption guidance separates speech from other audio information and gives examples of errors that change meaning.
Verify specific strings manually. In a payroll training clip, an auto-caption error can turn “do not put that information in the email” into permission to do it. Check negation, dates, names, sequence words, and the difference between an example and a real instruction. Add punctuation according to the spoken meaning rather than wrapping lines automatically wherever the text is longest. If the source narration is corrected, update the captions at the same time, or replace the version together after review.
Make the transcript a route to the same content
A downloaded caption file can start a transcript, but remove timing clutter when it is not helpful and add descriptions where the video uses on-screen text or action to teach. Break it into meaningful sections people can skim. Describe visuals as procedures rather than as paint colors: “The page closes the expanded Help panel and focus returns to the Help button” is useful; “The right side changes” often isn’t. W3C’s transcript resource discusses descriptive transcripts and recommends placing them where users can find them easily.
Place the transcript link beside the player. Check it from a test email in a narrow browser and in a situation where playback is blocked. Can a worker still locate the employer’s next step? Give a person who prefers text a dated, comparable lesson rather than telling them to call a supervisor to hear a reading of the video. If a private training is required, make the alternate path fit its access restrictions without adding an unnecessary second system or wait.
Keep the lesson and transcript under the same version label so someone can refer to them in a classroom discussion without guessing which one changed.
Give the final player its own test
Check the actual published player’s play, pause and caption controls with a keyboard, and check what the controls announce. A correctly authored caption track does not fix a control that can’t be reached. Use a phone as well as a desktop. If live training is offered too, plan an accessible question route and verify the format that will be available afterwards; don’t assume a recording inherits the accessibility arranged for the live session.
For translation changes, refer to the translation review guide. Run a keyboard and screen-reader pass on the actual training page, not only an embedded player demo. Document escalation and alternatives in the rollout review.