How Google turns a query into results, per a Barcelona recap

At Google’s Search Central Live Deep Dive in Barcelona, John Mueller and Gary Illyes walked attendees through how a search query becomes a results page. John Campbell of ROAST recorded the talks in Google Search Central Live Deep Dive, Barcelona - Day 3 Recap. These are a recap author’s notes of live sessions. The slides are not on Google’s events page, so wording below is as reported by ROAST, not Google’s own.

How does Google read a query?#

Mueller suggested treating normal search and AI search separately. In normal search, per the recap, the first step is detecting the query’s language. That is harder for brand terms, since a brand name often is not a word in any language, so Google leans on other data such as the user’s location.

Next, stop words that do not matter are removed. Entity recognition then keeps them where they belong to a name: in “The Lord of the Rings”, the “of” and “the” stay.

How do Google synonyms work?#

The recap says Mueller called expansion with synonyms a very important part of Google’s ranking system. It bridges the gap between the words a user types and the words on your page, and works a lot like searching with the OR operator. The example given: “fried chicken place in barcelona” is rewritten so that “place” also matches area, location or restaurant.

The same happens for photograph, image, picture and photo, and in other languages, such as Foto, Bild and Fotos in German. Some synonyms depend on context: “GM” means General Motors next to “car” and genetically modified next to “barley”.

There are also siblings, words that play a similar role but are not interchangeable, like Canon vs Nikon. Per the recap, Google spots these from “X vs Y” searches, treating a comparison as a sign the two are not synonyms. For language-specific sites, the advice was to focus on what your users actually search for.

What does AI search add?#

The AI flow adds a step. The query goes to an LLM, which creates fan-out queries and sends them back to the search engine, and the results ground the answer, which returns with links. Each fan-out query still goes through query understanding. Our report on Google’s R4T fan-out research covers the research side.

The recap says Google’s fan-out queries do not show in Search Console, because they do not come from users. Other platforms might show them, such as the Gemini app or some APIs, and every AI system does this differently.

What did the summary slide say?#

The recap lists the slide’s points:

• Do not worry about typos and plurals.
• Synonyms are expanded, and they are not always language-based.
• Some languages do not use spaces.
• Query fan-out does even more expansion.
• There are many opportunities to find and show your content.

What happens next?#

Illyes covered retrieval, how Google finds candidate documents in the index. Each document carries signals, and the recap says language and country are important ones at this stage. Retrieval happens before ranking: Google narrows the index to candidates first, then ranks them, and quality is what makes the final decision.

He then used “where to eat orange” as an example. Orange could be a fruit, a colour or a brand, and data from previous users helps Google decide which meaning to serve.

Do AI results need structured data?#

Per the recap, rich results are generally driven by structured data, while AI results are different: they do not need it, because they take text and build the result from it. That matches what we found when we compared Google and Bing statements on schema and AI.

What should you take from this?#

Write for the words people use, not one exact phrase. Cover the related questions around a topic, since fan-out creates extra queries your page may answer. The AI answer readiness checker tests whether a page answers a question directly. Follow AI search and content strategy for more.