ChatGPT prompts for answer writing.

The short answer

ChatGPT is good at three jobs in answer writing: planning an answer before you write, listing the dimensions a question has, and spotting a missing directive. It is weak at the job most aspirants give it: marking. It has no bank of answers that actually scored, its marks drift from one run to the next, and it will invent a report or a judgment with confidence. Use the prompts below for the first three jobs, and check the marking against real topper answers.

Prompt 1: check an answer you have written

Type your answer in (or paste the text of a photo), and ask for gaps, not a score. The last two lines matter: a rewrite teaches you nothing, and a number from a chatbot means little.

You are checking a UPSC Civil Services Mains answer. Question: [paste]. Marks: [10 or 15]. Word limit: [150 or 250].
My answer: [paste].
1. Does the answer do what the directive word asks (discuss / critically examine / analyse)? Quote the line that shows it, or say it is missing.
2. List the dimensions a complete answer would cover (social, economic, political, constitutional, ethical, environmental, international) and mark each covered or missing.
3. Name up to three examples, data points, committees, judgments or reports that would strengthen it. Only name ones you are sure exist; say "verify" next to any you are unsure of.
4. Is the conclusion forward-looking? Suggest one line.
Do not rewrite my answer. Do not give a mark.

Prompt 2: plan before you write

Planning is where a chatbot helps most. Toppers’ answers in the writeupsc bank run to about 175 words for a 10-marker and 251 for a 15-marker, so a plan has to fit five or six points, not fifteen.

Make a 60-second answer plan for this UPSC Mains question: [paste]. Marks: [10 or 15].
Give: the directive and what it demands, a one-line introduction, 4–6 body points under 2–3 sub-headings, one diagram or flowchart idea if it fits, and a one-line way forward. Bullet points only.

Prompt 3: make practice questions

Give me 5 UPSC Mains GS[1-4] style questions on [topic] in the pattern of recent papers: 10 and 15 marks, each with a different directive word. No answers.

Real past questions are better than generated ones: every GS and Essay question from 2013 to 2025 is in the Mains PYQs, with a link to write each one against the clock.

Where ChatGPT stops

• No reference answer. It compares your answer with what it thinks a good answer is, not with an answer that scored in UPSC.
• Marks that move. Ask twice and the mark can change; a mark that moves is not feedback.
• Made-up sources. A committee, a report or a Supreme Court case it names may not exist. Check every one before you learn it.
• Handwriting. UPSC is written by hand; a typed answer hides your real length, speed and layout.

A longer comparison is on writeupsc vs ChatGPT.

What writeupsc does differently

It reads the photo of your handwritten pages, finds the closest questions in a bank of 5,901 real topper answers, and names what yours is missing beside what the toppers wrote, without rewriting it. If no close topper answer exists, it says so instead of guessing. The first evaluation is free.

Questions