How does Google Search work?
From crawling to AI answers

Google Search works in three stages: crawling, indexing and serving results¹.

Table: The three stages of Google Search (Google Search Central)

Searches a year, according to Google (2025)

5 trillion+

Size of Google's index, per Google

100,000,000+ GB

Year PageRank was published

1998

Monthly AI Overviews users at I/O 2025

1.5 billion

There is no central registry of web pages, so Google has to discover them. Its crawlers, collectively called Googlebot, start from pages they already know and follow links to new ones; site owners can also submit sitemaps listing their URLs¹. Googlebot then fetches each page and renders it with a recent version of Chrome, running its JavaScript, so Google sees roughly what a visitor sees.

Crawling is polite by design: Googlebot adjusts how fast it requests pages so it doesn't overload servers, and it obeys a site's robots.txt file, which can tell crawlers to stay out of certain areas.

Indexing is understanding and storing. Google processes a page's text, title, headings, images, video and structured data, and works out whether it duplicates other pages; from a cluster of duplicates it picks one canonical version to show¹. The index itself is enormous: Google says it holds hundreds of billions of pages and is well over 100,000,000 gigabytes².

At its heart is a structure computer scientists call an inverted index, like the index at the back of a book: for each word, a list of the pages it appears on. When you search, Google doesn't read the web; it reads the index, which is why answers arrive in a fraction of a second.

Ranking is done by many automated systems, not one formula. Google's own guide lists systems for relevance, link analysis, freshness, reliable information, spam detection, passages within pages, helpful and original content, reviews and more⁴. Broadly, they consider:

Google uses human Search Quality Raters to check whether results are good, guided by a public handbook that describes these qualities; raters don't rank pages directly, but their feedback helps evaluate changes to the systems¹¹.

In 1998, two Stanford PhD students, Larry Page and Sergey Brin, published "The Anatomy of a Large-Scale Hypertextual Web Search Engine", describing a prototype called Google³. Its key idea, PageRank, treated a link from one page to another as a vote, and gave more weight to votes from pages that were themselves well linked. It made Google's results far better than rivals that relied mainly on matching words.

From 2015 Google added machine-learning systems that understand language rather than matching strings. In 2019 it applied BERT, a language model that reads words in the context of the whole sentence, to search; Google said at the time it affected about one in ten English searches in the United States⁵. A query like "can you get medicine for someone pharmacy" could now be understood as being about picking up a prescription for another person, where earlier systems would have matched the keywords.

In May 2024 Google began showing AI Overviews in the United States: short generated summaries at the top of some results, with links to the pages they draw on⁶. By Google I/O in May 2025 they had 1.5 billion monthly users in 200 countries and territories, and Google launched AI Mode, a chat-like search experience for longer, more complex questions⁷.

For site owners, Google's guidance is that there are no extra requirements or special optimisations to appear in AI features: the same crawling, indexing and quality systems apply, and pages need to be indexed and eligible to show a snippet⁸.

• Make pages crawlable — no robots.txt blocks on important pages, no content hidden behind logins, working internal links.
• Submit a sitemap — and verify your site in Google Search Console to see what is indexed and why.
• One clear URL per page — avoid duplicates; use canonical tags where duplicates are unavoidable.
• Answer the query fully — write for the person searching, with the answer early and the detail after.
• Show who you are — clear authorship, sources and contact details support trust.
• Keep it fast and mobile-friendly — usability is part of ranking and of whether visitors stay.
• Earn links, don't buy them — links from relevant, reputable sites are a quality signal; paid or spammy links can be penalised.

Google's own summary of the strategy is to create helpful, reliable, people-first content⁹. If your traffic arrives but visitors don't convert, the problem has moved from search to experience: that is where a SaaS UX audit helps.

Crawler (Googlebot) — a program that automatically fetches web pages by following links.

Index — Google's database of the pages it has processed and can show in results.

Inverted index — a data structure listing, for each word, the documents that contain it.

Canonical URL — the version of a page Google chooses to show when several URLs have the same content.

Ranking systems — the automated systems that order results for a query.

AI Overview — a generated summary shown at the top of some Google results, with links to sources.

The description of crawling, indexing and serving follows Google Search Central's documentation and Google's "How Search works" pages; historical facts come from Brin and Page's 1998 paper and Google's announcements for BERT (2019), AI Overviews (2024) and I/O 2025. The PageRank figure uses the published formula applied by us to a six-page example. Google changes its systems often, so treat details as accurate to October 2026. Last reviewed October 2026.