# Design a web search engine like Google: system design interview question

> Search across billions of pages with relevant results, fresh content, site access limits, and low query latency.

Source: https://sysround.com/questions/design-web-search-engine-like-google
Updated: 2026-11-05


Difficulty: hard. Levels: Senior, Staff+.

## How the interviewer opens it

"Design a web search engine that retrieves and ranks relevant web pages. It must support a massive collection, fresh results, and low query latency."

## What to clarify first

- Corpus
- Queries and traffic
- Fetch volume and read:write mix
- Freshness
- Latency
- Relevance
- Consistency and availability
- Durability
- Site access
- Scope limits

## Scope of a strong answer

Core requirements:

- Discover and fetch public web pages within each site's limits
- Add, update, and remove pages from the search index
- Rank pages for keyword queries
- Return titles, URLs, and text snippets with low latency

Usually out of scope:

- Ads and payments
- Personal results
- Image and video search
- JavaScript execution
- Model training

## What interviewers probe

### Finding matching pages

Common terms match millions of pages. The service must select useful results without scoring every page.

### Page changes in search results

Page changes must reach results quickly. Queries must not mix terms, titles, and snippets from different page versions.

### Page discovery and site limits

The web contains duplicate URLs and endless paths. Important changes compete with slow sites and limited fetch capacity.

### Search at scale

A query can require work across a large corpus. One slow dependency can delay all results, while spare capacity increases cost.

## Common mistakes

- The query service scans every page and relies on a cache to meet latency targets.
- The design omits page discovery and index updates.
- The crawler gives every page the same fetch schedule.
- The capacity estimate ignores work per query and slow dependencies.
- The result combines term matches from one page version with titles or snippets from another.
- The candidate spends the interview on model training instead of search system behavior.

## The bar by level

- **Senior:** Connects site limits, durable updates, complete page versions, and query deadlines. Preserves matched versions through metadata reads. Explains comparable scores, replica capacity, and worker recovery.
- **Staff+:** Quantifies query distribution costs and allocates the freshness budget. Defines overload behavior and capacity after zone failure. Defines version retirement across replicas, active queries, and caches. Explains trade-offs among relevance, completeness, latency, and cost.

Practice this question with an AI interviewer: https://sysround.com/practice/new?problem=web-search-engine


---
SysRound (https://sysround.com) is an AI practice tool for system design interviews. It has a voice or typed AI interviewer and a shared whiteboard. At the end, you get feedback on each skill and a model answer. The AI feedback can be wrong.