The Difference Between Traditional and Semantic SEO: Three Google Shifts, With Dates and Documents

Someone from the agency said in the meeting that the work is now semantic SEO. You nodded along. On the drive home, you realized you did not know what had changed.

You looked it up. You found the same sentence in several places, worded almost the same way. Traditional SEO looked at the word, and semantic SEO looks at the meaning. The sentence is right. It just does not tell you what happens to your page on Monday.

What is missing is what every serious change comes with: a date and a document. Google made three shifts in how it reads the internet, and put each one in writing. From the word to the thing. From the page to the entity it talks about. From the list of links to the answer.

I will take the three in order, with the paperwork for each one in hand. After that comes what they change in your copy.

The old game, described by Google itself

Remember the word search puzzles in the checkout-aisle puzzle books? You get a list of words and a grid of letters. You find HOUSE on the diagonal and circle it. Nobody asks whether you know what a house is. The game is finding the letters in the right order.

On September 26, 2013, Google Search turned 15. Amit Singhal, senior vice president of Search, opened the anniversary post by recalling 1998. Back then, you would “type in some keywords, and get 10 blue links to websites that had those words.” ¹

That is the word search at the scale of the internet. The machine looked for the letters of the query in pages and handed back a list. This is called keyword matching.

Traditional SEO was born for that game, and it played well. The keyword in the title, the keyword in the first paragraph, the keyword in the URL. One page for each term. And the fight for a spot among the ten.

The game is still on. Google keeps a page that explains how its own search works. On it, the most basic signal of relevance is content with the same keywords as the query. ² The word search is still running. It just lost its place as the only game.

A side note that doubles as a test for what you read out there. You often read that semantic search arrived with Hummingbird, in 2013. The name came out on the day of the fifteenth anniversary, through the press. Danny Sullivan, of Search Engine Land, reported what Google announced at a press event. ³ Google’s official post from that day does not contain the word Hummingbird even once. ¹ When someone gives you that date, ask which Google document backs it up.

First shift: from the word to the thing

2012: the name started pointing to something

More than a year before that post, the first shift had already been announced. On May 16, 2012, the same Singhal introduced Google’s Knowledge Graph. The post’s title carried the whole shift: things, not strings. ⁴

His example fits in a single search. Someone types taj mahal. It could be the monument, a Grammy Award-winning musician, a casino in Atlantic City, or the nearest Indian restaurant. ⁴ In a word search, all four answers are the same row of letters.

From 2012 on, each one became a separate thing, with its own identity and relationships to other things. This is called an entity.

What goes into that catalog, and where the facts come from, is in the piece on how Google records companies, people, and places. It also explains why there is no sign-up form. What matters here is the consequence for anyone who runs a company. The company’s name needs to point to one thing only, kept apart from everything else with the same name.

My book SEO de Entidade, published in Portuguese, calls this step the Google registry, where strings become things. ⁵ That work has a section of its own, Entity SEO, and this page lives in it.

2015 to 2019: the query started being read for its meaning

Walk into a hardware store and ask for the piece that holds a shelf up on the wall. The clerk does not search the stockroom for that sentence. He comes back with a shelf bracket. You never said bracket, and he found it anyway.

In a February 2022 post, Google explains that it brought this kind of reading to Search in 2015. The system is called RankBrain, and it was “the first deep learning system deployed in Search.” ⁶

Google’s official example works like the bracket. Someone asks “what’s the title of the consumer at the highest level of a food chain.” The system understands that the person is looking for what is usually called an “apex predator”. ⁶

Neural matching came in 2018. According to the same post, it “looks at an entire query or page rather than just keywords,” to get at the concepts behind them. ⁶

On October 25, 2019, it was BERT’s turn. And the example in the announcement is about a Brazilian. The search was “2019 brazil traveler to usa need a visa.” Before, Google returned results about U.S. citizens traveling to Brazil. ⁷

With BERT, the little word “to” started to carry weight, and Search understood the direction of the trip. The announcement said this would help Search understand one in 10 searches in the U.S. in English. ⁷

Beneath the announcements sit patents with numbers. US 10,452,978, on attention networks, describes each position in a sentence looking at all the others. It was granted on October 22, 2019, three days before the BERT announcement. ⁸

US 12,086,720, granted in September 2024, goes further. It describes a passage’s relevance to a query even when the two share no terms at all, or very few. ⁹

A patent describes a mechanism. It does not prove the mechanism is running today. Neither of these two names BERT or RankBrain. The link between them and the announced systems is my own reading. ^{8 9}

For your page, the consequence is practical. The machine can reach your copy without the exact word from the query. And repetition stopped being enough. Google itself writes that “when you search for ‘dogs,’ you likely don’t want a page with the word ‘dogs’ on it hundreds of times.” Its algorithms look for what exists beyond the word, like photos, videos, or a list of breeds. ²

Second shift: from the page to the entity

Take a wedding photo. Forty people show up in it: the aunts and uncles, the bridal party, the photographer reflected in the ballroom mirror. Nobody wonders whose photo it is. Everyone is in the frame, and the photo belongs to two people.

Traditional SEO treated a page like the list of everyone in the picture. If the word was there, it scored. A semantic reading asks the wedding photo question: which thing is this page about?

Google offers a tool that answers that, the Google Cloud Natural Language API. For each entity in a text, it returns a number from 0 to 1. The documentation says that number tells you the importance or centrality of the entity to the entire text. ¹⁰ This is called salience.

The reasoning has a patent behind it. US 9,619,457, filed in 2014 and granted in 2017, describes how to identify the salient entities in a document. ¹¹ Among the clues it lists are the sentence where the thing first appears and how many times it comes back. It also counts whether each mention uses the name, a common noun, or a pronoun. ¹¹

Three limits, and they matter. The patent expired in 2021 over an unpaid fee. The application it describes is predicting which ad to show each person. And the API is a cloud product: its number is no ranking score, and Google does not say Search uses it. ^{10 11}

What stays standing is the wedding photo question. John Mueller, of Google, gave an example in 2023 that any business owner will get. Picture a homepage that only says you “add pizazz to places.” It shows some beautiful houses, and nothing else. Neither people nor search engines know what you offer. “If your business paints houses, then just say that.” ¹²

The reasoning goes up one floor, from the text to the whole site. What is known about how Google sees a site’s focus, and with which caveats, is in the piece on semantic silos.

Third shift: from the list to the answer

Ask a friend who knows cars to recommend a mechanic. He does not hand you the phone book. He calls two people he knows, asks about price, asks about turnaround. Then he comes back with one name and the reason.

The 2013 post already pointed that way. Singhal closed it with a promise of “the best possible answers.” ¹ The shift arrived in parts, and each part has a date.

On October 15, 2020, Google announced it would start understanding the relevance of specific passages on a page. The post explains why: “sometimes the single sentence that answers your question might be buried deep in a web page.” The forecast was to improve 7% of search queries, across all languages. ¹³

Google’s official guide to its ranking systems keeps that description to this day. ¹⁴ The slice of a page that answers a question on its own has a name. This is called a passage.

On March 5, 2025, Google introduced AI Mode. The announcement describes a technique that runs several related searches at once, across different subtopics and sources. Then it brings the results together into one answer. ¹⁵ That is the friend calling around. This is called query fan-out.

Google’s guide to generative AI features, updated on July 10, 2026, gives an example. Someone asks “how to fix a lawn that’s full of weeds.” The fanned-out searches could include “best herbicides for lawns,” “remove weeds without chemicals,” and “how to prevent weeds in lawn.” ¹⁶

Put the two documents together and look at your page. It can be pulled in by a question nobody typed, asked by the machine on the person’s behalf. And it can make it into the answer through a single paragraph. That combination is my own reading, based on the two texts. ^{13 16}

The same guide says where the answer comes from. Google’s generative AI features are “rooted in our core Search ranking and quality systems.” ¹⁶ The answer changed shape. What picks the material is still the same Search as always.

What this asks of a page that wants to be cited is in the piece on LLMO, GEO, and AEO. It also covers what the industry sells under new acronyms.

What stays the same

Semantic SEO is not the opposite of traditional SEO. It is traditional SEO aimed at the right thing.

Keywords still count. In the same 2023 answer, asked about the ideal keyword density, Mueller said “Google does not have a notion of optimal keyword density.” He added that being explicit does matter. ¹²

Forced repetition has a name and a rule. Google’s spam policy calls it keyword stuffing: filling a page with keywords or numbers to manipulate rankings. Often, the policy says, those keywords appear in a list or out of context. ¹⁷

The technical side has not moved either. Crawling, indexing, titles, and speed still live in the technical SEO section, and they come before everything on this page.

What gets sold as new is often the old thing under a new name. The 2026 guide has a list of what you can ignore. ¹⁶ Breaking your content into tiny pieces so AI understands it: Google says there is no requirement. Writing in a special way for AI search: not needed either. New machine-readable files, AI text files, markup, or Markdown: Google says you do not need to create any of them to appear in Search. ¹⁶

What changes on your page, starting Monday

Pick one page, the one that matters most to your business. Set aside a sheet of paper and ten minutes. No tool comes into this.

1. At the top of the sheet, write the question the page answers. Write it the way a customer would say it at the counter, with no keyword. That is the first shift.
2. Find the paragraph on the page that answers that question on its own. Read it outside the page, as if it were the only passage anyone would see. If it does not exist, you have found the work. That is the third shift.
3. Underline the thing the page is about: the company, the service, the product. See in which paragraph it first appears by name. Count how many times it comes back by name, and how many times it turns into “we” or “the company.” That is the second shift, read through the patent’s clues, and that reading is mine. ¹¹
4. Last, mark every repetition of the keyword you added only to fit it in. Cut those.

The question you take to the next meeting

From now on, when someone says the work is semantic SEO, you have three questions to ask. Which question each page answers. Which thing it is about. Which passage in it answers on its own.

If the answer comes back as nothing but context, intent, and meaning, ask to see the page. The three shifts have dates, and all of them end in a paragraph you can point to with your finger.

The whole map, with the other fronts, lives in the semantic SEO section. Do the four-step sheet today, with the page that pays your business’s bills.