Search Intent: What Google Documents, and How to Read Intent on the SERP

On September 14, I searched Google in Brazil, where we are based, for “intenção de busca.” That is Portuguese for search intent. At the top, an AI-generated summary answered with four types: informational, navigational, commercial, and transactional. The source next to the list was a video from Semrush’s Brazilian channel. ¹

Now open the manual Google hands to its human search raters. Its list uses other names: Know, Do, Website, and Visit-in-person. “Commercial” does not appear among the intent types. ²

Google’s AI answer repeats the list from the SEO tools. Google’s own manual uses another one. Both live on the same screen, and almost nobody says where each one came from.

This piece puts the two side by side, each with its source. Then it shows how to read intent without leaning on any list. You read it from the format of what shows up on the results page.

What search intent is

Picture the counter at a hardware store. Three people walk up, one after another, and say the same words: “water heater element.” The first wants to buy the part. The second wants to know which one fits the water heater at home. The third wants someone to show them how to swap it out.

The words are identical. What each person wants to do with them changes. This is called search intent.

The rater guidelines separate two questions that usually get mixed up. The first is what the query means, and the guidelines call that the interpretation. The second is what the person is trying to accomplish when they search. That is the intent. ² The same text warns that a single person can have many intents for the same query. ²

Google’s page on ranking results puts intent at the very first step. To return relevant results, Google writes, it first needs to establish what you are looking for. That is the intent behind the query. ³

The first question, the one about meaning, already has a piece of its own. The piece on search volume traps shows that only the results page tells you what topic a search is about. This piece takes the second question. With the topic right, what does the person want to do?

The four intents in Google’s manual

The manual is called the Search Quality Rater Guidelines. The version in force when I checked is the one from September 11, 2025. ²

In the section on intent, the guidelines suggest thinking in four types. ²

Know. The person wants to find information or explore a topic. Inside it lives a special case, the Know Simple query. It is a question with a short, correct answer that fits in one or two sentences or a short list. The manual’s example is Barack Obama’s height. ²

Do. The person wants to accomplish a goal or engage in an activity. The manual lists downloading, buying, obtaining, being entertained, and using a website or app. Paying a parking ticket is among the examples. ²

Website. The person wants to reach a specific website or page. The manual calls that page the target of the query. ²

Visit-in-person. The person is looking for a place to go, near where they are. The examples include “pizza,” “dentists,” and “car repair.” ²

The manual does not treat the list as a set of sealed drawers. It says the line between Know and Do is sometimes blurry, and that many queries do not fit neatly into a single type. ² In its example, a search for “walmart” can go either way. One person wants to go to the store. Another wants to shop on the website. ²

The manual’s yardstick: Needs Met

Picture a waiter coming back to the table. The customer ordered a diet soda. Bringing a diet soda meets the need. Bringing a regular soda meets it halfway. Bringing the wine list meets nothing at all.

The manual rates every result this way, on a five-point scale. This is called Needs Met. ²

The scale runs from Fully Meets to Fails to Meet. The top rating is reserved for queries with a clear intent to find one specific result. It goes to the result that is that target. ² The middle ratings measure how much a result helps each interpretation of the query. A result that serves only an unlikely meaning gets a low rating. One that goes off-topic lands on the bottom rating. ²

Interpretations work like steps on that scale. The dominant interpretation is what most people mean. A common interpretation is what many or some people mean. A minor interpretation is what fewer people have in mind, and it can be reasonable or unlikely. ² There is also the no chance interpretation. In the manual’s example, almost nobody searching “hot dog” in the United States wants a result about an overheated pet. ²

One caveat, which the piece on volume already recorded. No single rater score changes a page’s position by itself. The ratings measure whether search is working and supply examples to improve it. ² The manual describes the target Google is aiming at. How the systems get there, it leaves out.

The part that speaks most to today’s screen sits in the special blocks. The manual calls them Special Content Result Blocks, results that show content right on the search page. A weather forecast, the calories in a banana, a music video that plays in place. ² When most people would not click, the rating goes to the block alone. ² The need can be met without a visit to any website.

Where the tools’ four types came from

The list the industry repeats has a well-known grandparent. In 2002, Andrei Broder published “A taxonomy of web search,” with data from the AltaVista search engine. ⁴ He split searches into three classes. ⁴

Navigational, when the person wants to reach a site. Informational, when they want information they assume sits on some page. Transactional, when they want to carry out an activity through the web. ⁴ Back then, even finding a map counted as a transaction. ⁴

Two points from the paper appear in none of the texts I read on the first page of this search in Brazil. ⁷ Broder warns that he does not assume intent can be inferred from the query with any certainty. ⁴ And in his data, informational searches made up less than half of the total. ⁴

The fourth class, commercial, is missing from both Broder’s paper and Google’s manual. It shows up in Semrush’s knowledge base, which lists four main types: informational, navigational, commercial, and transactional. ⁵ The tool says it identified the intent of every keyword in its databases through automated analysis. ⁵

Semrush itself records the gap. A post on its blog from November 2024 first presents the four intents from Google’s manual. Right after that, it says SEO professionals use slightly different categories. ⁶

I did not find the document that gave birth to the commercial class. What can be stated with a source is short. The industry’s list of four is a working convention of the tools, and Google uses another.

What gets lost between the two lists

The two lists do not line up piece by piece. The manual puts buying inside Do: “buy ipad” has a clear purchase intent. Meanwhile “ipad reviews,” the search of someone doing research first, sits in Know. ²

The same passage says most product queries do not clearly point to one type. And it asks for high ratings on results that help people research, browse, and decide what to buy. ²

Visit-in-person has no seat in the industry’s list of four. On September 14, I read the thirteen pages that shared the first page for “intenção de busca” in Brazil. ⁷ Three of them added a local type to make room for the map. They called it “geolocalizada” (geolocated) and “busca local” (local search). ⁷ Google’s manual already had that spot. ²

Two of the thirteen cite the manual’s list and then carry on with the industry’s. One of them has a matching table with no row for Visit-in-person. ⁷ None of the thirteen explains the Needs Met scale, and none carries Broder’s caveat about certainty. ⁷

Google itself says the screen changes with intent

Google’s documentation for site owners has a precise example. A search for “bicycle repair shops” will likely show local results and no images. A search for “modern bicycle” is more likely to show images and no local results. ⁸

The page on ranking completes the picture. Words like “cooking” or “pictures” in a query lead the systems to show recipes or images. ³ Many queries have local intent, and a search for “pizza” brings results about nearby businesses that deliver. ³

The weight of each factor, the same page says, varies with the nature of the query. For a current news topic, fresh content counts for more than it does for a dictionary definition. ³

Google’s guide to its ranking systems names two mechanisms behind this work. BERT is the AI system Google uses to understand how combinations of words express different meanings and intent. ⁹ Freshness systems bring up recent content when the query calls for it. The example is “earthquake.” On an ordinary day, it brings preparation material. Right after a quake, it brings news. ⁹

One label, four screens

On September 14, I asked Semrush, in its Brazil database, for the intent label of four searches in Portuguese:

All four came back with the same label: informational. ¹⁰

Then I opened each search live on Google in Brazil, and in Semrush’s snapshot of the results. ^{10 11} The screen told four different stories.

pizzaria (pizzeria). Live, the top of the page was a map with three pizzerias in the neighborhood I was searching from. The web results below it all came from the same area. ¹¹ The manual has a name for this: Visit-in-person. ² It is also the pizza example from Google’s own page. ³

tênis de corrida (running shoes). Live, the top brought a carousel of sponsored products, with filters for price and gender. Seven of the eight web results were stores. The eighth was a store’s article on its five best models. ¹¹ In the Semrush snapshot, eight of the ten addresses were store pages. ¹⁰ This is the manual’s product query: research and compare, with the purchase in sight. ²

como trocar resistência do chuveiro (how to replace an electric shower’s heating element). Live, the top was a carousel of short videos from four social networks. Among the results, only two were written guides. ¹¹ The AI summary showed up in the middle of the page. Its step-by-step leaned almost entirely on YouTube videos. ¹¹ In the Semrush snapshot, five of the ten addresses were YouTube videos. ¹⁰ The person wants to do the job with their own hands, and the screen answers with people doing it.

altura do cristo redentor (height of Christ the Redeemer). Live, the AI summary answered in its first line. The monument is 38 meters tall in total: 30 for the statue itself and 8 for the pedestal. ¹¹ This is the manual’s Know Simple query, and the need can end right there, without a click. ²

By the manual, the four screens ask for different things. Going to a place, choosing a product, learning to do a repair, knowing a number.

The tool’s label is there to sort a list of a thousand keywords, and it does that job. When you decide what page to build, the screen is what answers.

What the screen does not tell you

Reading the SERP has limits, and the manual spells them out for its own raters. Anyone who researches the query on Google should not rely on the top results. A query may have meanings the results page does not show. ² Two of the thirteen texts I read claim the algorithm has already done this work for you. None of the thirteen carries the manual’s warning. ⁷

The screen also belongs to one place and one moment. In the Semrush snapshot, “pizzaria” brought addresses from the São Paulo area and no map among the page features. ¹⁰ Live, from another city, the map took the top. ¹¹ And the manual warns that the meaning of a query changes over time. ²

The snapshot ages, too. One of the ten addresses Semrush lists for “intenção de busca” now returns a page-not-found error. ^{7 10}

That is why our house rule calls for two readings. One is the tool’s, and one is the live screen’s, with the date and place written down.

The method: read the top before the label

The table below translates what fills the top of the screen into the manual’s list. It is my own reading, built from the screens of September 14 and from the manual’s examples. ^{2 11}

The step-by-step fits on one sheet of paper.

1. Search the term on Google, from your client’s city. Write down the date and the place.
2. Before you scroll, note what fills the top, in a single word.
3. Translate the top using the table. Where the screen mixes formats, write down both intents.
4. Look at the level. The type says what the person wants, and the level says where they stand on the path. The Inverted Intent Pyramid, from the SEO Master Prompt, separates three levels. Base is the person with the problem, Middle is the one comparing, and Tip is the one ready to hire.
5. Decide where to compete. If the top is a map, the main contest happens on Maps. That is a search surface with rules of its own, described in search marketing. If the top is video, your text competes for the few text spots, or you pick up a camera.
6. Give each distinct intent its own page. That is a house rule. Pages on the same topic belong together, and how to group them is in semantic silos.
7. Reopen the screen before you publish.

With the intent decided, what the page declares to Google, from the title to the subheadings, is in the piece on on-page SEO.

The question you ask from now on

From now on, when a tool marks a keyword as informational, open the screen before you agree. The label helps you sort the list. The top of the page tells you what the person wants to do.

Open your spreadsheet today. Next to the tool’s intent column, create two more: “what fills the top” and “intent by the manual.” Fill them in for the five keywords you most want to go after, and mark where the columns disagree. The rest of the topic is in our semantic SEO section.