Typing a vague question into Claude or Gemini and getting a vague answer back is not a skill problem — it is a prompt problem. AI models are remarkably responsive to how you frame a request. Give them context, a role, a constraint, and a clear output format, and the quality of what comes back shifts noticeably. Here is how to do that without turning every query into a ten-minute ritual.
Set the context before you ask anything
AI models have no idea what course you are in, what your lecturer emphasized, or how much you already know. When you skip that setup, you get a generic answer written for a hypothetical person who may not resemble you at all.
The fix is a short context block at the start of your prompt. Think of it as briefing a smart tutor who has just walked into the room. A useful context block covers three things: your course and level, what you already understand, and what specifically is confusing you.
Compare these two prompts:
Weak: Explain opportunity cost.
Strong: I am a first-year economics student. I understand that opportunity cost means what you give up by choosing one option over another. What I cannot grasp is why economists include implicit costs — like your own time — but not sunk costs. Can you walk me through the logic with a concrete example?
The second prompt produces an explanation targeted at exactly the gap you have, rather than a textbook definition you could have found in thirty seconds.
Another lever here is asking the model to adopt a specific stance. Phrases like “explain this as if I have read the introductory chapter but not the advanced sections” or “push back on my reasoning” change the register of the response entirely. Anthropic’s usage documentation for Claude notes that specifying a persona or role consistently improves response relevance — and in practice, it does.
Use structural constraints to avoid walls of text
One of the most common complaints students have about AI answers is that they are long, exhausting, and hard to use when you are trying to revise or write. That is mostly a prompt design issue.
Structural constraints tell the model exactly what shape the output should take. Some of the most useful ones:
- Limit the length. “Answer in under 150 words” forces the model to prioritize.
- Request a specific format. “Give me a numbered list of the three strongest objections to this argument” is far more actionable than “what are the problems with this argument.”
- Ask for comparisons in a table. When you are trying to distinguish between two theories, two historical events, or two statistical methods, a side-by-side table is often clearer than paragraphs.
- Request a follow-up question. Ending your prompt with “then ask me one question to check my understanding” turns a passive read into active recall, which research on retrieval practice — covered thoroughly in Make It Stick by Brown, Roediger, and McDaniel — consistently links to better long-term retention.
- Separate explanation from application. “First explain the concept, then show me how it applies to this scenario” keeps the response organized and easier to reread.
These constraints are not about making the model work harder. They are about making its output immediately usable for what you actually need to do next — draft a paragraph, prepare for a seminar, or check whether your understanding holds up.
Iterate instead of accepting the first answer
The single biggest missed opportunity most students have with AI tools is treating the first response as final. That first answer is more like a first draft in a conversation. Pushing back, narrowing, or redirecting almost always gets you somewhere more useful.
A few iteration moves worth keeping in your toolkit:
Challenge the answer directly. “You said X — but my textbook argues Y. Which is more accurate for an undergraduate-level answer, and why do they differ?” This forces the model to reconcile sources rather than just assert.
Ask for the counterargument. If you are writing an essay and Claude has just helped you build your main argument, ask it to steelman the opposing position. Knowing the strongest version of the other side makes your own writing sharper.
Zoom in on one sentence. If an explanation mostly makes sense but one part does not, quote that sentence back and ask the model to unpack it specifically. “You wrote ‘the multiplier effect amplifies fiscal policy impacts’ — can you show me the mechanism step by step?” gets you further than rereading the paragraph four times.
Test your own understanding. Once you think you have grasped something, explain it back to the model in your own words and ask it to identify any errors or gaps. This is a form of the Feynman technique — articulating an idea as simply as possible to expose where your understanding is actually shaky.
One practical caution: AI models including Claude and Gemini can produce confident-sounding errors, particularly on niche topics, recent events, or highly technical material. Cross-check anything that will end up in assessed work against your course materials or library databases. Use the AI to build understanding; use your sources to verify facts.
What Selene does with this: every article here is written to sit alongside your actual coursework rather than replace the thinking you need to do yourself. If a prompt pattern in this piece works for you, adapt it — the best version is always the one shaped around your specific course and your specific gap.