The anatomy of a study prompt

Four things the model cannot guess

A weak prompt is not usually too short. It is missing information the model has no way to infer and will therefore invent: who you are, what you already know, what the answer is for, and what shape it should take.

Compare.

Explain enzyme inhibition.

against

I am in the second year of a biology degree. I understand Michaelis-Menten kinetics and can read a Lineweaver-Burk plot. Explain the difference between competitive and non-competitive inhibition in terms of what happens to Vmax and Km, and why. Around 250 words. Do not define enzyme or substrate. End with one question that would catch me if I had only memorised the result rather than understood it.

The first produces a textbook page you have already read. The second produces something you can use. Nothing in it is clever. It states four things:

1. Level and prior knowledge: so the answer starts where you are, not at the beginning.
2. The specific task, not the topic, the job. Explain the difference in terms of Vmax and Km.
3. Constraints: length, what to omit, what to assume.
4. The form: prose, list, table, worked example, question.

Say what you already know, especially the wrong bits

The highest-value sentence in most study prompts is a statement of your current, possibly mistaken, understanding.

My current understanding is that a competitive inhibitor lowers Vmax because it blocks the active site. Tell me where that is wrong and why, and do not just give me the correct version.

This works because it converts a general explanation task into a diagnosis task. The model now has something specific to act on, and you get the correction aimed exactly at your error rather than a tour of the topic that happens to contain the answer somewhere.

It also gives you a much better test of whether you have learned anything, because you wrote down a claim before reading the answer. Generating an answer before being told is one of the reliable memory effects in the literature; writing your wrong version into the prompt gets it for free.

Ask for the shape you actually need

Schoolwork has shapes. Name them.

• Give me a worked example with the reasoning shown at each step, then a second, similar problem with the answer hidden.
• Give me the three strongest objections to this argument, strongest first, in one sentence each.
• Give me a five-line summary and then the part I would be most likely to misunderstand.
• Rewrite this paragraph in my own register without adding any new claims, and list what you changed.

Each of these is a different tool. A prompt that does not name a shape gets the default shape, which is a mid-length balanced essay, which is almost never what a study session needs.

The negative constraints do most of the work

Models over-explain, hedge, and pad with caveats. This is a consequence of how they were tuned, not a fault you can argue them out of, but you can instruct around it.

Useful lines to keep:

• No introduction, no summary, start with the substance.
• Do not tell me it depends. Give the standard case, then name the exception in one line.
• Assume I have read the chapter.
• Do not apologise or restate my question.
• If you are not confident about a specific figure, say so instead of giving one.

That last one does not make the model reliable (it has no honest confidence signal in the sense of the previous module), but it does shift the style towards hedging on specifics, which makes the risky parts easier to spot.

Iterate rather than restarting

When an answer misses, the fix is usually one sentence, not a rewritten prompt. Too general: do it for the specific case of a non-competitive inhibitor at saturating substrate. Too long: cut it to the third of that which I could not have got from a textbook. You defined terms I said I knew: try again without those.

The habit worth building is noticing which of the four things was missing. Answer too basic: you omitted level. Answer wandered: you omitted the task. Answer too long: you omitted constraints. Answer in the wrong format: you omitted the form. Almost every disappointing answer maps onto one of those four, and diagnosing it takes five seconds.

The prompt is not the skill

Providers have spent enormous effort making these systems tolerant of vague prompts, and they are much better at it than they were. So do not turn prompt writing into a craft with rituals. Four items and a willingness to say no, like this covers essentially everything. The skill worth having is knowing what you actually want, which is a thinking problem, not a typing one.