We analyzed 1,300 UPSC Prelims questions (2013–2025). Here's what UPSC actually asks.

We took every GS Paper I question from UPSC Prelims 2013 through 2025 (1,300 questions), verified every answer key against the official UPSC keys, classified each question by subject, topic, format, and difficulty, and counted.

Here is what the data says: Polity & Governance (24.1%) and Economy (20.5%) together are nearly 45% of the paper. Roughly three-quarters of the paper is static syllabus rather than current affairs.

Most questions reward coverage outright, and elimination skill when coverage runs thin. Most prep folklore survives none of this.

Which subjects does UPSC weight?

Is the weightage changing over time?

Yes, and in a direction most aspirants haven't noticed. Comparing the first five years of our window (2013–2017) with the last five (2021–2025):

The loudest finding: Art & Culture has collapsed, from ~6 questions a year in the mid-2010s to 1–3 in recent papers. If your plan still budgets a month for it because a 2016-era strategy article said so, the data disagrees. Polity, meanwhile, keeps growing: it hit 35 questions in 2024, over a third of the paper.

What format are the questions?

Recall, or reasoning?

We labelled every question by the thinking it demands: pure recall, analytical combination, or applying a concept to a scenario. Drawing those lines involves judgment (the honesty note below says how much), so we'll state what the classification supports without a decimal on it.

The winning behaviour is boring: cover the standard sources, revise until recall is fast, practise the formats. Coverage regularly rescues average brilliance; brilliance doesn't rescue thin coverage.

How much is current affairs, really?

The elimination finding

We also asked of each question: could a prepared aspirant reach the answer by confidently ruling out options, even without producing it cold? That is the most judgment-laden label in the dataset, so this finding stays qualitative too; it's the second claim in the card above. What it means in practice: identical knowledge can score fifteen marks apart on the same paper, trained test-taking against reading alone. Every question in our arena carries an explanation written to show the elimination path.

Which specific topics repeat most?

Money & Banking alone has produced 70 questions in 13 years, more than all of Art & Culture. Four of the top six topics are Polity. These ten rows are where scarce revision time earns the most.

What should you change in your plan?

1. Order your effort by the weightage table rather than by subject size or fear. Polity + Economy first, always current.
2. Retire the month-long Art & Culture project. Cover it from PYQs and a thin source; spend the freed weeks on Polity.
3. Train statement-based questions deliberately. Half the paper is this one format.
4. Cap current affairs at a fixed weekly hour budget. Roughly three of every four questions come from the static core.
5. Practise elimination as a skill. Most questions reward it, and it trains.

How to read these numbers honestly

Before you rebuild your plan around this post, here is how far the data goes, and where it stops. We'd rather you trust us slightly less and plan slightly better.

• The classifications are ours. Deciding whether a question is "Economy or Polity", "recall or applied", involves judgment. We applied one consistent taxonomy across all 1,300 questions, but a different reasonable taxonomy would shift most percentages by a point or two. Treat the digits after the decimal as texture rather than precision.
• Weightage is a prior, not a promise. These are patterns from 13 papers. UPSC owes them nothing. The tables tell you where to place your hours on average; they cannot tell you what the next paper holds.
• Scope: GS Paper I only. Nothing here covers CSAT, Mains, Essay, or the Interview.
• This analysis stops at 2025 — Prelims 2026 has since been held and is not yet in this dataset; every table will be refreshed once it is. We are not affiliated with UPSC. This is our independent analysis, based on our understanding of the exam. Where the data changes, our advice will change with it — that's the deal.

Methodology: 1,300 questions = GS Paper I, 2013–2025 (100/year). Every answer key verified against official UPSC keys; 19 widely-circulated wrong keys corrected in the process. Two-level taxonomy (7 subjects, 60+ topics); IR classified under Polity. Computed directly from the dataset in July 2026; the 2026 paper is pending inclusion.

What this data can't tell you

Everything above measures one thing: what UPSC asks in Prelims GS Paper I. That is the half of the exam a dataset can see. The other half, Mains, is decided by what you can produce: an answer, from memory, against the clock, in your own words. No count of questions can measure whether you can do that, and coverage doesn't quietly turn into that ability on its own.

So use this data for what it's good at (placing your Prelims hours where the paper pays) and run the writing alongside it from the first topic. That's how the six-step topic loop works: map, learn, practise, write, evaluate, revise, so each topic feeds both papers before you move on. DOJO trains the writing skill itself; PRISM tells you the truth about what you wrote: honest marks, kind tone.

And because claims should stay inside their evidence: a Mains counterpart to this analysis (the themes and directives those papers reward) is on our roadmap. When it exists, it will come with the same honesty note.

Quick answers