Auto-generated captions are usually not accurate enough on their own to meet accessibility requirements for prerecorded lecture video. Automatic speech recognition (ASR) typically loses accuracy on technical terms, proper nouns, accents, overlapping speakers, and poor audio, and it commonly omits speaker labels and non-speech sounds altogether. The workable standard in higher education is automatic captioning plus a human correction pass, not automatic captioning alone. 

A student who relies on captions to follow a recorded lecture doesn’t get a second chance at the sentence an auto-caption garbled. If that sentence contained the one term an exam question hinges on, the caption error isn’t a formatting flaw—it’s a gap in what the student was actually taught. Multiply that across a semester of recorded lectures, and the gap compounds in a way a single spot-check rarely catches. 

Most comparisons of auto captions and professional subtitling stop at “auto is cheaper, human is more accurate” without saying what an accurate caption has to include or how to measure it. Defining that standard and checking captions against it is what turns this from a vague trade-off into something a media team can actually plan around. 

The gap between the two isn’t uniform. It widens sharply as soon as content stops being a single clear voice reading from a script, which describes almost no real lecture recording. 

The Web Content Accessibility Guidelines (WCAG) don’t set a percentage. They define captions as a synchronized text alternative for both the speech and the non-speech audio needed to understand the content. In practice, that means the right words, a label when the speaker changes, punctuation that keeps the meaning clear, meaningful sounds such as laughter or an alarm, and timing that matches the video. 

The practical step is to measure accuracy instead of guessing at it. Take a five-minute clip; type a word-perfect reference transcript; count every wrong, missing, and extra word in the captions; and divide by the number of words in the reference. That error rate, checked against the target your team sets, shows whether a caption track is ready to publish. Uncorrected ASR output generally doesn’t meet the standard on its own, especially on lecture content with technical vocabulary. 

Captions are only one of five success criteria that apply to recorded and live media. Each one maps to a different deliverable. 

Audio description matters when the screen carries something the speaker never says, such as an equation written on a whiteboard or a chart the instructor points to. The simplest fix starts with the instructor: saying aloud what is written or shown often removes the need for separate description.

For video, captions meet success criterion 1.2.2, and a transcript is best practice rather than strictly required. For audio-only content—a podcast or recorded audio lecture with no video track—a transcript is the requirement, since there’s no caption track to attach it to. Transcripts also double as indexable, citable text, which captions alone don’t provide, and they’re what makes a recorded lecture searchable for a student trying to find one specific point later in the semester. 

Public colleges and universities need accurate captions on the prerecorded video they publish, outside a few narrow exceptions, such as archived content. The Department of Justice (DOJ) rule under Title II of the Americans with Disabilities Act (ADA) adopts WCAG 2.1 Level AA, and uncorrected automatic captions are unlikely to meet that standard. On April 20, 2026, the DOJ published an interim final rule setting two compliance dates. Public entities serving 50,000 or more people, a group that includes most public colleges and universities, must comply by April 26, 2027. Smaller entities have until April 26, 2028. 

The same standard includes captions for live content, so livestreamed lectures and events need real-time captioning, such as Communication Access Realtime Translation (CART), not captions added after the fact. 

1. Pull a sample of recent lecture recordings across a few different courses and check whether their captions were auto-generated only or reviewed by a person before publishing. 
2. Build a course-specific glossary of technical terms, names, and acronyms for reviewers to check ASR output against, since this is where automatic transcription fails most often. 
3. Set a minimum accuracy target in writing, a specific number rather than “good enough,” so vendors and in-house reviewers are held to the same bar and progress can actually be measured. 
4. Decide, per content type, whether a transcript is needed alongside captions, and budget for both where audio-only content like podcasts or recorded audio lectures is involved. 
5. Separate livestreamed classes and events from recordings, and plan real-time captioning for them, since captions corrected afterward don’t help the people watching live. 

A caption workflow built entirely on free auto-generated output is usually the fastest way to accumulate a backlog of technically “captioned” video that still fails a manual review. Captioning services that pair ASR with a human correction pass close that gap without discarding the speed advantage of ASR entirely. 

If the backlog is bigger than your media team can review, ContentA11Y’s accessibility services cover lecture recordings, with ASR output reviewed and corrected by people rather than published as is, plus transcripts and translated subtitles where you need them. We can review a sample of your captioned videos at no cost and measure how accurate they are before you budget for the rest.