Eng Students · For engineering students

Let the AI quiz you on your lab results. Answer first.

Ask an AI chatbot to quiz you on your own lab readings and answer before it replies. What the research says, the steps, and what to check.

30 Oct 2026 · 5 min read · sources read up to 30 Oct 2026

Picture the silence

Example: your lab viva is next week. The report is handed in and you feel fine. Then the examiner points at the one reading that came out lower than the textbook value and asks why, and your stomach does a small flip.

You wrote that number down yourself. Whether you can say why it came out that way is a different question, and it is much nicer to find out at your desk than in front of the examiner.

We can't tell you what your examiner will ask. We looked, and found no source that says what lab vivas ask, and they will differ by college and by teacher. A "why" question is easy to imagine, though, and it is the kind you can practise.

The easy way to use a chatbot

The tempting move is to type your readings into a chatbot and ask, "What will the examiner ask, and what are the answers?" You would expect a tidy list back. It feels like progress.

But reading a list of answers is a lot like rereading your notes, and the research rates rereading low. In a well-known 2006 experiment, 120 undergraduates read short passages. Some then restudied the passage, and others took a recall test, writing down everything they could remember. Here is how much each group remembered ^[1]:

Rereading only came out ahead at the five-minute mark. By two days the quizzed group had pulled ahead, and it stayed ahead.

A second experiment, with 180 students, showed the same pattern after a week, and the rereaders felt surer they would remember it (4.8 against 4.0 on a seven-point scale) ^[1]. That gap between feeling ready and being ready is the part that stayed with us.

A big 2013 review of ten study techniques rated practice testing high and rereading low ^[2][3]. A practice test is a quiz you take to learn, not for marks.

Now the fair warning. Those were university students in the US reading prose passages, and nobody gave them feedback on their answers ^[1]. None of it involved lab results, "why" questions or an AI's quiz, and none of it measured a viva. Reading still helped a little: after a week the rereaders remembered 42% ^[1].

The version we would try

So we would flip the order. Instead of the AI answering first, it asks and you answer. Here is a prompt we wrote for that, with your own readings typed in:

Here are my lab readings: [type your own readings].
Ask me 5 viva-style questions about why my results
came out this way, one at a time. Wait for my answer
before you say anything about it. After each answer,
tell me what was right and what was missing, and
what to check in my lab manual.

We haven't run this prompt on any chatbot yet, so we can't say how a particular one behaves. Treat it as a starting point and change the wording until it asks the way you like.

A few habits matter more than the wording:

• Type in only your own readings. No names, no roll numbers, nobody else's data.
• Ask for "why" questions, one at a time.
• Answer before the AI says anything, even if your answer is shaky. Typing a rough answer is the whole point.
• Let it tell you what was right and what was missing, then check that against your lab manual.

The last habit has a reason behind it. The 2006 authors note that their quizzes gave no feedback, and say that checking your answers afterwards may help even more ^[1]. They say "may", so we hold that loosely.

When the AI marks you wrong

It can. The developer docs from Anthropic, the company that makes Claude, say even the most advanced models can write text that is factually wrong, that its techniques reduce this but do not remove it, and that you should validate anything critical ^[7]. Google's Gemini privacy page puts it in plainer words: these tools "can hallucinate and present inaccurate information as factual" ^[8]. To hallucinate here means to make something up and state it as fact.

So treat the AI's marking as a second opinion. Your lab manual and your textbook outrank it. If they disagree, trust the manual and ask your teacher.

On privacy, the same Google page says human reviewers may look at some chats, and asks you not to enter confidential information you would not want a reviewer to see ^[8]. That is one tool's page and we haven't read the others, so check yours. Your own readings are fine to type. Names, roll numbers and classmates' data are not.

What this isn't for

This is practice. The AI asks, you answer. It is not a way to get a lab report, a lab record or viva answers written for you, and we would not use it that way. Anthropic's usage policy lists submitting AI-assisted work without proper permission or saying who wrote it among the things not to do ^[6]. That is one company's policy, not your college's, so follow your college's rules on AI for anything you hand in.

Check your chatbot's age rules before you start. Claude's consumer terms require users to be at least 18 ^[9]. Gemini with a personal Google Account asks for 13, or the age set where you live ^[10]. We haven't re-read any other tool's terms.

Some makers have built question-first modes of their own. Anthropic described a Learning mode in Claude for Education, aimed at universities, that asks things like "How would you approach this problem?" ^[4]. Google described Guided Learning in Gemini, which uses probing questions and quizzes ^[5]. Both are launch announcements. We haven't tried either and haven't checked who can use them.

How we checked

We read both experiments in the first paper, the publisher's summary of the 2013 review and the vendor pages in the list below, all on 30 October 2026. The full text of the 2013 review would not open for us, so we only use its high and low ratings for practice testing and rereading ^[2][3]. OpenAI's own pages would not open either, so nothing here is about ChatGPT.

We haven't run the prompt ourselves yet. Everything above comes from the papers and pages, and the prompt is our own wording. Treat it as a sensible bet. No source we read measured viva results, marks or placements.

Researched and drafted with AI assistance; checked by a human. zenlite is independent of Anthropic, Google and OpenAI.