From Hummingbird to AI Search: Entities and Semantic SEO

Key takeaways

• Google's language systems (RankBrain, BERT, passage ranking, neural matching) all aim to match meaning rather than exact words.
• Topic clusters work because they show the full set of entities and questions around a subject, linked in a way crawlers can follow.
• AI answer features pull passages, so each section of a page should answer one question clearly enough to quote alone.

If your content plan still starts with "one keyword, one page, repeat the phrase five times," it ignores a decade of change. Google, and now AI answer engines, try to understand what a page is about and which real-world things it describes. This page traces that shift from the Knowledge Graph to AI answers and turns it into a practical content method.

What changed with the Knowledge Graph and Hummingbird?

The Knowledge Graph (2012) gave Google a database of entities, and Hummingbird (2013) rebuilt query handling to use meaning instead of word-by-word matching.

Google introduced the Knowledge Graph in May 2012 with the line "things, not strings." The example was a search like "taj mahal," which could mean a monument, a musician or a casino. Knowing those are different entities lets the engine choose results and panels by meaning.

Hummingbird, announced the following year, applied the same thinking to the whole query. Longer, conversational searches ("where can I get my iPhone screen fixed near the office") could be read as an intent with entities and a location, rather than a string of words to match.

How did RankBrain, BERT and passage ranking build on that?

Each later system improved Google's ability to match the meaning of a query to the meaning of a passage.

Google's ranking systems guide still lists RankBrain, BERT, neural matching and passage ranking as active systems as of November 2026. None of them reward repeating a phrase. They reward pages whose content plainly matches the concept behind the question.

Why does this matter for AI answers?

AI answer features select and quote passages, so pages built from clear, self-contained answers about well-defined entities are easier to cite.

Google says AI Overviews and AI Mode may issue several related searches (query fan-out) and pull from a wider set of pages than a classic results page. ChatGPT search and similar tools also retrieve pages, then quote or summarize parts of them. Answers vary from run to run, and nobody outside these companies knows exact selection rules. Still, a passage that names the entity, states the answer and gives a supporting detail is easier for any system to lift. Our guides on Google AI Overviews and how ChatGPT search picks sources go deeper.

How do you build entity-first content?

Start from the entities and questions in a topic, then map them to a connected cluster of pages instead of a list of keywords.

A working method:

1. Define the core entity (your product, service or subject) and write a one-sentence definition you will reuse consistently.
2. List related entities: components, use cases, alternatives, standards, places, organizations.
3. Collect the real questions people ask about each, from Search Console queries, People Also Ask boxes, support tickets and sales calls.
4. Group questions by intent and assign each group to one page, so no two pages compete.
5. Link pages to each other with descriptive anchors that name the entity being linked to.
6. Mark up the organization, products and articles with structured data where it fits (see schema markup for AI search).

Hypothetical example: a solar installer builds a hub page on "home battery storage" with spokes on battery chemistry, sizing, incentives, installation steps and comparisons with generators. Each spoke links back to the hub and to the sizing page, which links to the quote form.

How do you know if it is working?

Track rankings and clicks for the whole query group, plus mentions in AI answers, rather than one head term.

Entity-first pages tend to rank for many long variations they never mention verbatim, so look at the count of ranking queries per page in Search Console. For AI visibility, run a fixed set of prompts monthly and record whether your brand or pages are cited, accepting that results fluctuate. Check that crawlers can reach you at all with our GEO readiness checker. If you want help building the cluster map and measuring AI citations, see our GEO and AI search service.

Frequently asked questions

What was the Hummingbird update?

Hummingbird was a rewrite of Google's core search algorithm announced in 2013. It was designed to interpret whole queries, especially longer conversational ones, by meaning instead of matching each word separately.

What is an entity in SEO?

An entity is a distinct, identifiable thing such as a person, place, organization, product or concept. Google's Knowledge Graph, introduced in May 2012 under the phrase 'things, not strings', stores entities and the relationships between them.

Do I still need keywords?

Yes, as evidence of how people phrase their needs. Use keyword research to find the questions and terms around a topic, then write naturally about the entities involved instead of repeating one exact phrase.

Does schema markup help with entities?

Structured data can make it clearer which organization, product or person a page is about, and it is required for some rich results. It does not replace clear writing, and Google says no special markup is needed to appear in AI Overviews.

Sources

1. A guide to Google Search ranking systems
2. Introducing the Knowledge Graph: things, not strings
3. Understanding searches better than ever before
4. AI features and your website