What is GraphRAG and which kind of question is it built for?

Attempt it

GraphRAG builds a knowledge graph of entities and relationships from the corpus offline, clusters it into communities, and writes an LLM summary of each community. This lets it answer global, aggregative questions — like "what are the main themes across all these documents?" — by reasoning over community summaries. Standard top-k vector retrieval only fetches locally similar chunks, so it can handle specific fact lookups but structurally cannot synthesize across an entire corpus. The trade-off is that building and refreshing the graph is token-expensive.

Your answer

Quick reference

TL;DR

GraphRAG turns a corpus into an entity-relationship graph, clusters it into communities, and pre-summarizes each — so it can answer global "what are the themes?" questions that top-k vector RAG can't.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Imagine a giant pile of meeting notes and someone asks "what were the big themes this whole quarter?" Plain search just hands you the few sticky notes that mention the exact words you typed — useless for a question that needs the whole pile. GraphRAG does prep work ahead of time instead. It reads everything, draws a map of who talked to whom and what connects to what, groups that map into neighborhoods of related stuff, and writes a short summary for each neighborhood. Now when the big-picture question arrives, it reads those neighborhood summaries instead of hunting for matching words. The downside: drawing the map and writing all those summaries takes a lot of reading up front, so it is slow and pricey to build.

Key concepts

Retrieval-Augmented Generation Agents

Detailed answer & concept explanation

~6 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.

Sign in for 5 free unlocks/day or go Pro for unlimited. See Pro pricing.

Quick scan · 2–4 min

Everything important, quickly.

3-4 min: the offline pipeline (extract, cluster, summarize), why top-k can't do global questions, the local versus global query split, and the build-cost trade-off.

Real-world examples

Real products, models, and research that use this idea.

Microsoft's open-source GraphRAG project is the canonical implementation of entity-graph plus community-summary retrieval.
Sensemaking over incident-report or research-paper collections, where users ask for cross-document themes, not single facts.
Legal discovery over a connected document set, where the question is about patterns across the whole corpus.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QHow does GraphRAG's global query mode actually combine community summaries into one answer?

Sign in or go Pro to see the approach for this follow-up.

QHow would you keep the graph fresh when source documents change frequently?

Sign in or go Pro to see the approach for this follow-up.

Don't say this

Red flags and common mistakes that signal junior thinking. Click to expand.

Most common mistake

Thinking a bigger k fixes global questions. No value of k makes top-k synthesize a corpus-wide theme — the answer lives in no single chunk, so you need a precomputed structure, not more chunks.

Sign in or go Pro to see all red flags and common mistakes. See Pro pricing.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.