Executive summary
Every finished lecture video is checked for something. Audio levels, colour, caption timing, the
logo in the right corner. Almost none are checked for the thing that matters most to a student: whether what is
shown on screen is academically correct.
For most of the history of lecture video this gap did not matter much, because the only visuals were the
professor's own slides. That has changed. Editors, agencies and — increasingly — AI tools now add diagrams,
illustrations, footage and on-screen text that the professor did not create. Each one is a new claim made in the
institution's name. This paper sets out why accuracy is missing from current workflows, what it costs when it
fails, and how to make it a standard part of production rather than an afterthought.
1. Accuracy has no owner. Editors are not subject experts, faculty review under time pressure, quality
teams check technical specifications, vendors contract for delivery, and AI tools have no knowledge of the
discipline. The one check that matters most belongs to nobody.
2. AI makes the problem larger and harder to see. Generative tools are fluent, not accurate. They
produce visuals that look authoritative, which is exactly what makes a subtle error dangerous.
3. Errors are cheapest to prevent and most expensive to fix late. A wrong diagram caught in a
production plan costs a sentence. The same diagram caught after captioning, upload and student use costs a
re-render, re-captioning, re-upload, an errata notice and credibility.
4. Full-length review does not scale. If accuracy depends on a professor watching every finished
minute, a 400-hour programme needs 400 hours of expert review — time that does not exist.
5. Accuracy can be designed into production. Faculty as source of truth, discipline-aware visual
policies, automated gates on real people and documents, a readable production plan and a named approval step
together turn accuracy from a hope into a process.
The principle. AI assists production. Faculty remain the source of truth.
What is inside
15 pages, 9 exhibits, an About Lecture Studio page, sources and assumptions listed.
- The missing check
- Why it matters more now
- A taxonomy of visual errors
- Why current workflows miss errors
- The cost of an error
- Faculty as source of truth
- A discipline-aware visual policy
- An Academic QA standard
- A ten-point policy for institutions
- How Lecture Studio implements this