How Google Rewrites and Expands Your Queries Behind the Scenes
Google query rewriting is the step where Google changes what you typed before it searches. It fixes spelling, adds synonyms and related terms, drops words that block good results, and can split one question into many searches. Google says one in ten queries is misspelled, so pages should cover a topic and its variants, not one exact phrase.
What is Google query rewriting?
Query rewriting is how Google turns the words you typed into a better search. The goal is to find pages that meet your need, even when those pages use different words. You see the results of the rewrite, but rarely the rewrite itself.
People type fast, misspell words and use their own terms. Page owners write in their own terms too. Rewriting closes the gap between the two vocabularies.
This step sits early in the process. Google reads the query, works out what it means, and only then retrieves and ranks pages. Our guide on how search engines understand search queries covers the full journey. This post zooms in on the rewriting step.

Rewriting, expansion, relaxation, reformulation and fan-out: what each means
These terms overlap, but each describes a different move. Rewriting is the umbrella, and expansion, relaxation, segmentation, correction and fan-out are the tools under it. Reformulation usually means the person changing their own search, not Google.
|
Technique |
What it does |
Example |
|
Spelling correction |
Fixes typos before searching |
A search for “restaraunt” is read as “restaurant” (illustration) |
|
Synonym expansion |
Adds words with the same meaning |
“change laptop brightness” also matches pages that say “adjust laptop brightness” |
|
Related term expansion |
Adds close concepts |
A “jogging shoes” search can return pages that say “running” |
|
Relaxation |
Drops a word that blocks good results |
A long query returns pages that miss one less important term |
|
Segmentation |
Groups words that belong together |
A place name with two words is read as one unit, not two |
|
Reference resolution |
Fills in words from earlier context |
A follow-up search borrows the topic of the previous search (described in a Google patent) |
|
Fan-out |
Splits one question into many searches |
AI Mode breaks a question into subtopics and searches each one |
The synonym and “jogging shoes” examples come from Google’s own pages. Google has not published relaxation, segmentation or reference resolution as named steps, so treat those rows as the general techniques search engineers use.
Why Google does this: precision and recall
Every rewrite trades two things. Recall is how many good pages a search finds. Precision is how many of the found pages are actually good. Expansion and relaxation raise recall. Segmentation protects precision.
A rewrite that goes too far drifts away from what you meant. That risk is highest for names, numbers and exact terms, which is why we cover how to protect them later.
What Google says it does to your query
Google’s How Search Works pages, Search Help and blog posts describe spelling correction, synonyms and language models. You do not need a patent to know rewriting happens; Google says so directly.
Spelling correction
In a 2021 post on spelling, Pandu Nayak wrote that 1 in 10 queries is misspelled. Google’s spelling model has more than 680 million parameters and runs in under 2 milliseconds.
Synonyms
Google’s How Search Works page describes a synonym system. Its example: a search for “change laptop brightness” can match pages that say “adjust laptop brightness”. The words differ, but the need is the same.
Google’s Search Help adds that the most helpful results may not contain all your search terms. Back in 2010, Steven Baker explained that synonyms depend on context. “Arm reduction” and “arms reduction” mean different things, so Google cannot swap words blindly. He also noted that matching synonyms are bolded in results.
Language models and new queries
How Search Works says language models improve results in over 30% of searches across languages. In its 2019 BERT post, Google also said that 15% of daily queries are new. Google has never seen those searches before, so it cannot rely on past data alone. It has to understand the words.
The systems behind it in plain English: RankBrain, neural matching, BERT and MUM
In a 2022 Google blog post, Pandu Nayak described four AI systems that help Search understand language. Each does a different job, and not all of them rank pages.
- RankBrain (2015) relates words to concepts, so a page can be relevant without the exact words.
- Neural matching (2018) helps Google retrieve documents by concept rather than by exact words.
- BERT (2019) reads words in context, including small words like “to”. Google uses it for both ranking and retrieval.
- MUM is a newer model. In that 2022 post, Google said it was not currently used to rank pages.
The 2019 BERT post used the query “2019 brazil traveler to usa need a visa”. The word “to” matters here. The person is a Brazilian going to the USA, not the other way round. At launch, Google said BERT helped with 1 in 10 English searches in the US.
For a deeper look at how these models read pages as well as queries, see how search engines understand content.
Rewriting from context: earlier searches in the same session
Google may also use your earlier searches to rewrite a new one. This idea appears in a Google patent, not in Google’s public documentation. Treat it as a possible method, not proof of what Google does today.
The patent is US 10,387,437 B2, “Query rewriting using session information”. It was filed on 13 January 2017 and granted on 20 August 2019. It describes taking a new query, building candidate rewrites from earlier queries in the session, and scoring them.
This is the idea behind reference resolution. If you search for a stadium and then type “how many seats”, the second search is unclear on its own. Context from the first search fills the gap.
A patent shows what a company has explored, not what it runs today. We read them as clues, not as a manual.
How to see a rewrite yourself
Most rewriting is hidden, but some of it shows on the results page. Look for correction messages, bolded words you did not type, and results that ignore one of your terms.
- “Showing results for” means Google has already searched for a corrected version. A “Search instead for” link lets you go back to your original words.
- “Did you mean” is a suggested correction you can click.
- Bolded words in snippets that you did not type are often synonyms or close variants Google matched.
- Results without one of your words are expected. Google’s Search Help says the most helpful results may not contain all your search terms.
To make Google match your exact words, put them in quotes. Google’s Search Help recommends quotes when you need exact words in the results. This is useful when you check how your own pages rank for a precise term.
From one query to many: how AI Mode uses query fan-out
Fan-out takes rewriting a step further. Instead of changing one query, Google turns it into many. AI Mode then combines what it finds into one answer.
At Google I/O in May 2025, Elizabeth Reid explained that AI Mode uses “query fan-out”. It breaks a question into subtopics and issues many searches at once. Deep Search can run hundreds of searches for a single request.
For your pages, this means a long question may be answered from several pages, each covering one part. In our view, a page that covers one subtopic well has a better chance of being used than a page that skims all of them. We explain the method and what to do about it in query fan-out in Google AI Mode. Our AI SEO services page explains how we plan content for AI answers.
Arabic, Arabizi and spelling variants: what changes for UAE searches
UAE searches mix scripts and spellings. People type in Arabic, in English, and in Arabizi, which is Arabic written in Latin letters. Place names also have several English spellings, such as Jumeirah and Jumeira, next to the Arabic جميرا.
Spelling correction and synonym matching can help Google connect these variants, but Google has not published how it handles each one. It still pays to use the spellings your customers use, in both languages where it makes sense.
AI search is also moving into Arabic. In October 2025, Google launched AI Mode in Modern Standard Arabic, using the same query fan-out method. Google said AI Mode questions are 2 to 3 times longer than classic queries. In our view, longer questions give Google more context to work with. For bilingual sites, see how we approach international and multilingual SEO.
What query rewriting means for your pages
Write for the topic, not for one exact phrase. Google’s spelling and synonym systems already connect many variants to the same need, so the page that answers that need best has the edge.
One page per topic, not one per synonym
Separate pages for “adjust” and “change” versions of the same topic compete with each other. Google’s own synonym example shows it can match both words to one need. In our experience, one strong page beats two thin ones. Our guide to types of search intent helps you decide when two queries truly need two pages.
Cover variants naturally
Use the words your customers use, including common alternatives. Answer the related questions they ask next. This helps both classic search and fan-out, where each subtopic may be searched on its own.
Protect exact names, prices and model numbers
Rewriting works on meaning, which is risky for exact details. State brand names, product codes, prices and dates clearly and in full. Put them in headings, tables or lists where they are easy to read. When we run SEO services audits, we check that key names and codes are spelled the same way everywhere.

How to spot rewrites in Search Console
Search Console shows the queries your pages appear for. Some of those queries will use words that are not on the page. Those are signs that Google matched your page through synonyms, spelling or related concepts.
Follow these steps:
- Open the Performance report and select Search results.
- Click a page you know well and filter to that page.
- Open the Queries tab and sort by impressions.
- Read the list and mark queries that use words your page never uses.
- Use a query filter with a custom regex to group variants, such as misspellings or synonyms of your main term.
- Compare clicks and average position for the exact term against its variants.
If a variant gets many impressions but few clicks, your title may not reflect it. Consider adding the variant naturally to the title, a heading or the intro. If a page ranks for a variant with a different intent, that may point to a separate page worth creating.

Myths to drop
Three common beliefs about rewriting do not match what Google has said. Here is what the sources actually show.
“Google swaps queries for ads”
In 2023, a Wired opinion piece claimed Google replaces queries with ones that make more money. Google’s Danny Sullivan replied “we don’t”. Search Engine Roundtable reported that the author later said she may have misread Google’s explanation.
“You need LSI keywords”
LSI keywords do not appear in the Google posts cited here. What Google describes is language models, synonym systems and neural matching. Write naturally about the topic and its related ideas instead of chasing keyword lists.
“Exact match still wins”
Google’s Search Help says the most helpful results may not contain all your search terms. Exact words still matter for names and precise terms. But a page can rank for a query it never states word for word, as your Search Console data will show.
FAQ
In RAG, or retrieval-augmented generation, query rewriting changes a user’s question before the system searches its documents. It may fix wording, add related terms or split the question into parts. The goal is the same as in Google Search: find better source material for the answer.
Query expansion adds synonyms and related terms to a search so it finds more relevant pages. It raises recall, which means fewer good pages are missed. Google uses it so that pages written in different words can still match your need.
In information retrieval, query expansion is the practice of adding terms to the original query before searching. The added terms can be synonyms, related concepts or corrected spellings. It helps bridge the gap between the words people type and the words in documents.
A query is a short request to a search engine to find pages. A prompt is an instruction to an AI model to produce an answer. In AI Mode the two meet, because a long prompt-style question is turned into many search queries.
Google changes your words to fix typos, match synonyms and find pages that meet your need. It says one in ten queries is misspelled, so correction is common. If you need your exact words, put them in quotes.
Yes. Google’s systems match pages by meaning, so a page can appear for synonyms and close variants it never states. You can see this in the Search Console Queries report for that page.
Sources
- Google: How Search Works, ranking results
- Google Search Help: refine web searches
- Google blog: The ABCs of spelling in Google Search
- Google blog: Understanding searches better than ever before
- Google blog: How AI powers great search results
- Google blog: Helping computers understand language
- Google patent US 10,387,437 B2: Query rewriting using session information
- Google blog: AI Mode update at I/O 2025
- Google MENA blog: Introducing AI Mode in Arabic
- Search Engine Roundtable: Google does not delete queries to replace them
