“Answered from memory, not from your book”: the case for a visible warning
Every reply in the regular teacher app carries a one-line warning. Why it is there, what it does and does not tell you, and why hiding it behind an icon would have been a mistake.
6 October 2026 · 5 min read · Product information as of 6 October 2026
The situation
A teacher asks the app to explain the water cycle for Class 6. The reply arrives quickly, reads well and sounds right. Thirty seconds later it is on the board.
Nothing in that sequence tells the teacher where the explanation came from. If it is correct, fine. If it contains a slip, the slip now has the teacher’s authority behind it.
The tension
AI replies are fluent whether or not they are right. The usual way products handle this is a small information icon, a terms page, or a line in the onboarding that nobody reads twice. The warning exists legally and disappears practically.
The supplied website content for VIDYA Copilot is direct about this: show the answer warning and the voice limits clearly, and readers should not have to tap a small icon to find them.
The decision
In the regular app, every reply carries the line:
Answered from memory, not from your book — worth checking.
It is a statement of mechanism, not a disclaimer. It says three things in eleven words:
- Where the reply came from. The AI’s learned knowledge, not a search of your textbook.
- What it did not do. It did not open your book. The feature that adds textbook facts was tested and switched off because it did not improve replies enough.
- What to do next. Check it.
Why “memory” and not “AI”
The word choice matters for teachers who are not interested in how models work. “From memory” is something every teacher understands: a confident recall that may be slightly off. It sets the right expectation without jargon. “Worth checking” is an instruction a teacher can act on in the moment, by comparing the important facts with the textbook or their notes.
What the warning does not say
It does not say how often the reply is wrong. The supplied reports give one data point: in one reported phone test of the newer version, about one in ten replies was hard to make sense of, with worse results on unfamiliar topics and in Marathi. That is a reason to check every reply, not a rate you can rely on in every situation.
It also does not cover the phone’s voice tools. Those sit outside the app, may use the internet and have their own privacy settings. The website explains that separately rather than overloading the warning.
The same principle, applied to marking
The warning is one expression of a wider rule: do not present a feature as reliable before it is. The tool that suggests marks has been built for testing and is switched off in the regular app because it is not reliable enough yet. When it is ready, the plan is for it to show the student’s own words that support each suggested mark, with the teacher deciding. Until then, it stays off rather than shipping with a warning attached.
Takeaway
A visible warning is only useful if it changes behaviour. The test for this one is simple: after reading it, do you compare the reply with your book before using it in class? If yes, it is doing its job. Read more on the trust and teacher control page.
Facts in this article come from the supplied product documentation dated 6 October 2026. No customer quotes, adoption figures or results beyond the reported tests are claimed. Leadership views are presented as company perspective, not as direct quotations.
