Products SEO & GEO Agent Paid Ads Agent Website Agent Conversion Agent
How it works Pricing FAQ Blog Book a call

Semantic search: how Google's NLP reads meaning

Updated · 6 min read

Short answer Semantic search means Google works out what a search means, not just which words it contains. Language systems such as BERT, RankBrain and neural matching connect a customer's wording to pages that say the same thing differently. Your search words still matter. Write how customers talk, answer their real question, and skip pages for every phrasing.

Want help applying this to your business? Talk it through on a free 30-minute call.

Book a call

It's search that matches meaning instead of exact strings of letters. Google says the first job is to find "the intent behind your query", and it builds language models for that (Google Search: How Search works, ranking results). Its own example:

"For example, you might have searched for “change laptop brightness” but the manufacturer has written “adjust laptop brightness.” Our systems understand the words and intent are related and so connect you with the right content." (Google Search: How Search works, ranking results)

The technology behind it is called natural language processing (NLP): software that reads human language for meaning, grammar and context. For an owner, the practical change is this:

Old keyword matchingSemantic search today
A page ranks for the exact words on itA page can rank for searches worded differently
"Stand" matches "stand-alone""Stand" is read in context
Small words like "to", "for" and "no" are ignoredSmall words can flip the meaning, and Google reads them
One page per phrasingOne good page answers many phrasings

Keywords did not stop mattering. Google still calls matching words "the most basic signal that information is relevant", especially in headings and body text (Google Search: How Search works, ranking results). Semantic search widens the net; it doesn't remove the words.

How did Google move beyond exact keywords?

Through a series of systems. All but the Knowledge Graph appear in Google's guide to its ranking systems. Dates are given only where Google published them.

SystemWhat Google says it does
Knowledge Graph (announced May 16, 2012)A model of "real-world entities and their relationships to one another: things, not strings"
RankBrainUnderstands "how words are related to concepts"
Neural matchingUnderstands "representations of concepts in queries and pages and match them to one another"
BERT (announced October 25, 2019)Understands "how combinations of words express different meanings and intent"
Passage rankingLooks at single sections of a page to judge how relevant the page is
MUMUnderstands and generates language; Google says it's "not currently used for general ranking"

An entity is a specific thing Google can tell apart from others with the same name. Google's 2012 example was Taj Mahal the monument versus Taj Mahal the musician. Your business is an entity too: a name, an address, a set of services.

RankBrain is the one that most directly explains why a page can rank for words it never uses:

"It means we can better return relevant content even if it doesn't contain all the exact words used in a search, by understanding the content is related to other words and concepts." (Google Search Central: A guide to Google Search ranking systems)

What changed with BERT?

BERT reads each word together with the words before and after it. Google said in 2019 it would help with "one in 10 searches in the U.S. in English", mostly long, conversational ones. Its examples show what keyword matching used to get wrong:

  • "2019 brazil traveler to usa need a visa": old results were about Americans going to Brazil. BERT reads the "to".
  • "do estheticians stand a lot at work": the old system matched "stand" to "stand-alone". BERT ties "stand" to the physical demands of a job.
  • A parking-on-a-hill search with "no curb": the old system ignored "no".

Google's advice to searchers that year was to "let go of some of your keyword-ese" and type naturally (Google blog: Understanding searches better than ever before). Your customers took it. Pages written in stiff keyword strings now match their questions worse than plain sentences do.

Do exact keywords still matter for SEO?

Yes, but as a starting point, not a formula. Put the words customers use in your page title, main heading and opening lines. Beyond that, Google's SEO Starter Guide says you don't need every variation:

"Google's language matching systems are sophisticated and can understand how your page relates to many queries, even if you don't explicitly use the exact terms in them." (Google Search Central: SEO Starter Guide)

What still trips owners up is chasing a number: a keyword percentage, a list of "LSI keywords", or an "NLP score" from a tool. None of these is a Google measure. See keyword density for why.

  1. Answer the question behind the search. "Furnace making a banging noise" wants causes, whether it's safe, and what a repair costs. Answer those in plain sentences.
  2. Use your customers' words, not your trade's. If they say "AC not cooling" and you say "refrigerant system diagnostics", use both. Keyword research for a local business shows how to find their words.
  3. One page per job, not per phrasing. "Water heater repair", "fix hot water tank" and "no hot water" belong on one page. Google's AI search guide warns that pages for "every possible variation of how people might search", made mainly to manipulate results, violate its scaled content abuse spam policy.
  4. Name the entities clearly. Say the business name, the service, the town and the brand of equipment the same way everywhere, including your Google Business Profile. That helps Google tell your business apart from others. See Google knowledge panel for your business.
  5. Give each section a clear heading. Passage ranking reads single sections of a page, so a heading like "How much does drain cleaning cost?" followed by the answer makes that section easy to match.

Is it the same in AI Overviews and AI chat answers?

For Google, yes. Its guide to generative AI search says:

"AI systems can understand synonyms and general meanings of what someone is seeking, in order to connect them with content that might not use the same precise words." (Google Search Central: Optimizing for generative AI search)

It adds that you don't need to worry about having enough "long-tail" keywords, and that you don't need to "write in a specific way just for generative AI search". The same pages that work for semantic search work there.

In Canada, Google's systems work the same way, with one note: Google says the language of the query decides the language of most results, so a search in French gets French pages. If you serve French-speaking customers, see multilingual website SEO.

Example

Illustrative cases with invented numbers.

A plumber with eight thin pages (US dollars)

The site has separate pages for "drain cleaning", "clogged drain", "drain unclogging" and five more like them, each 200 words. Merge them into one page that covers causes, the $150–$300 price range, what's included and how fast a crew arrives. Redirect the old addresses to the new page.

A salon whose page lists services as keywords

The services page reads "balayage highlights color correction keratin Vancouver". Rewrite it as short sections in customer language: "What is balayage?", "How long does it take?", "How much does it cost?" with prices in Canadian dollars.

A bookkeeper who uses only formal terms

The page says "accounts reconciliation services". Clients search "catch up on my books". Add a section in their words that explains what catch-up bookkeeping involves and how long it takes.

FAQ

Is semantic search the same as natural language processing?

No. NLP is the technology that reads language. Semantic search is what Google does with it: match a search to pages by meaning.

Can I optimize for BERT?

Not directly. BERT is how Google reads language, not a checklist to meet. Clear, natural writing that answers the question is what it matches well.

Do I need an NLP or "semantic SEO" tool?

No. Such tools can suggest topics you missed, but their scores are their own. Reading your page as a customer would tells you more.

Will semantic search let me rank for a city I don't serve?

No. Matching meaning doesn't change where you work. Pages stuffed with town names can break Google's spam policies; see city pages for service areas.

Different for you?

Keep learning

NextKeyword prominence: does putting keywords first still matter?Where your search words should go on a page, what Google actually says about placement, and why prominence and proximity scores are a tool's math.

Do it yourself

On-page SEO checklist: every item to check before publishingA pre-publish on-page SEO checklist for one new page: content, title, headings, URL, images, links, indexing, mobile and speed, plus what to do after launch.

Ask us one question

One line is enough. We reply by email.

Sources

  1. Google Search: How Search works, ranking results
  2. Google Search Central: A guide to Google Search ranking systems
  3. Google blog: Understanding searches better than ever before (BERT, October 25, 2019)
  4. Google blog: Introducing the Knowledge Graph: things, not strings (May 16, 2012)
  5. Google Search Central: SEO Starter Guide
  6. Google Search Central: Optimizing for generative AI search