How Google determines relevance and helpfulness

I have the feeling that I have studied so much in recent years. But today, as I advised a customer who was affected during the March -Kern -Update, Google called Google called A development in the way they determine the helpfulness of contentI had the feeling that I had to check again how Google determined what is relevant and helpful. This article got out! And I got ideas to help my customer. 🦾

The more we understand how Google determines which results are probably helpful and relevant for a viewfinder, the better we can do when improving our websites so that we are probably this result.

I used deep research to understand how Google determines relevance and helpfulness

I asked all 3 of Grok, Chatgpt and Geminis Deep Research Tools this request:

Research me to find credible information about how Google determines relevance and helpfulness. I am looking for Google research studies and official Google documentation. I am less interested in articles from SEOS.

Grok referred me to this article from Google from 2018. I found quite interesting:

The development of the search

My notes:

• Thanks to Progress in machine learningGoogle can evaluate high -quality content with greater accuracy.
• Neuronal embedding (also known as neural matching or vector codes) helps Google to understand the underlying concepts that people are looking for without relying on certain words used.
• “So how do you use intentions? Search queries are a good starting point, since the phrasing of a query can result in a lot about the intention of a user. “
• “Brands that use these intentions to anticipate customer needs and to react with relevant experiences. “”

Grok also recommended Google’s announcement on Bert. Did you know that Bert was a voice model? At the time it was announced that most SEOs had never heard of a voice model before. I didn’t have!

And of course another very important document for reading Google is Word 2 VEC Paper from 2013, which explains the beginnings of conversion of the language into numerical vectors.

Gemini sent me to Google How search document works. If you haven’t read it lately, I would encourage you to study it again!

I read this several times, but for some reason this part got me to the point:

Signals.

Google brought us with the core update in March 2024, which they brought with them New signals To the core systems:

So now we know that the important signals:

• Meaning
• relevance
• Quality
• User -friendliness
• context

The search work document explains all of you in detail.

Meaning

You use voice models to understand what the user really means when you enter a query.

relevance

The systems analyze the content to assess whether it contains information that is relevant to what the viewfinder is looking for. Keywords in the headings or in the main text body are important.

You use “aggregated and anonymized interaction data To assess whether search results are relevant for queries. “In the doj VS Google test we learned that clicks are used to train Google’s system. I would encourage you to read more for more Pandu Nayak testimonyOr if you don’t have time to read the full Doc, Ctrl-F for “clicks”:

My understanding of how Google evaluates the relevance is that you use conventional ranking systems to first evaluate the results. Then use the vector search (the neuronal embedding that I mentioned before) to further restrict the results to those who are likely to match the user intention. And then these neural embeds are over time Finely coordinated Based on what searchers have actually decided to click and deal with it.

Make machines predict. Then the actions of real people help to improve the machines so that they make better predictions. I wrote much more about it in my book SEO in the Gemini -ära -The story of how AI has changed the Google search.

At the end of this article I will share a little more how we can use an understanding of the vector search to improve our content. However, I believe that Google’s relevance has continued to be used than just using the vector search.

At the end of 2024, Google published a paper entitled “It’s all relative! -A approach to the production of synthetic queries to improve the relevance forecast of Zero-Shot”. This was specific to training language models to understand what is relevant, but I do not understand why such techniques would not be used for the search. In this paper you gave a LLM a document and asked to generate two queries for the document – one relevant and one that is related but is not really relevant. By focusing on the difference between the two, it was better to understand what is a bit relevant.

Quality: Google says: “After we have identified relevant content, our systems want to prioritize those that appear most helpful. For this purpose, you identify signals that can help to demonstrate the content of specialist knowledge, authority and trustworthiness. Did you know that 2019 the The rater guidelines replace many of the cases of “food” with the words “side quality”?

One Signal used here are links. “One of the factors used to determine quality is to understand whether other prominent websites link the content or refer to the content.” I would encourage you here to think less about left in the context of Pagerank and more about the fact that links are recommendations or voices from other known sources in your area.

Google’s predictions about the quality are also something that is finely coordinated over time – in this case by quality applicants. “The aggregated feedback from our search quality assessment process helps to refine how our systems recognize the quality of the information. ““ Here too, machines predict which content is probably of high quality, and demonstrate the suitable EEAT that are expected for the request, and the people assess the results and helps the machines to improve.

From the rater guidelines:

Using friendliness: Usability is about Side experience. You are probably thinking about Kern -Web -Vital and side speed, which is definitely part of the side experience. However, I think that many people overlook this line:

Can users find the main content? The Rater guidelines Talk in detail about the main content. It is the part that helps the page to achieve their purpose:

The The last update of the QRG Notes that Google wants the learned systems to go further here to determine whether there is too much filler content to distract from the main content.

Context: You use keywords along with the language that are used for the search, location of the user and even current events to add more context to the understanding of the user’s request.

Chatgpt’s deep research was excellent here. It goes more detailed on Rankbrain, mom and more. I would encourage you to do so Read it!

How can we use this information?

You are likely to read this article because you want to know how you want to make it so that your content is what Google presupposes in order to be most likely and reliable. I will tell some thoughts on how to do this, based on the information we have just learned.

However, there is something important that I have to say … and it is uncomfortable.

We all guess when it comes to maneuvering Google.

There was a time when SEO was essentially a science. Carry out keyword research. Write a good title day. Use the headings well. Get some links. Monitor, optimize and repeat. This simple strategy used to be very effective.

Google’s systems have developed immensely. Google started to transform the language into numerical vectors twelve Years ago. Bert radically changed how the search works Seven Years ago. And who knows what innovations Google uses in recent years that you have not published any information about!

So much of the SEO council we give and read, based on what worked in the past or what we should work for, but everything is really guessed.

I have experimented with some customers and in some cases I was able to improve the keyword performance.

These are good -looking graphics, but I also had several cases in which changes made to better fulfill and match the user intent. That is because there are many signals that Google intervenes. I have seen many cases of websites that used to classify well on the basis of excellent SEO skills that have not been successful since the March -Kern -Update, as Google rates the results. If you lack Eeat, I suspect that optimizing your text, in order to look better, does not have much to do.

Against this background you will find some tips and thoughts from me here:

Concentrate more on purpose than on keywords

You can take the top keywords with you that send the data traffic to your side and give them to an LLM (I prefer Gemini Thinking Experimental 01-21 in AI studio at the moment.)

Prompt: Here are the keywords that bring traffic to this side. And here is the content (add it to it). What is the main purpose of this page? What do users try to achieve? How can I better structure or add this page so that it meets these requirements?

Play around with the resemblance with cosine … maybe?

The similarity of the cosine is a way to calculate, like similar or closes elements in a vector room. Google will probably announce its content or in other words, it turns it into vectors when they are indexed. You can use Python to determine the Cosinus similarity of a query to a content. If you optimize the content to improve the similarity of the cosine, you should. TheoreticallyLook better for vector search systems.

I played with it. If you only know a little coding, you can ask Chatgpt to guide you through the writing of code that finds the cosine similarity of a key word to a content. You can then ask for help to optimize the content so that this is rather similar to the keyword.

However, we have to take a few things into account. First, it is unlikely that Google will carry out a simple review of the similarity with cosine between keywords and content. Google probably does much more with its keyword search to understand the intention and to adapt it to the relevant content. As we have already explained, the vectors that are used for the neuronal agreement are finely coordinated, either through what Google has learned from surveillance, how people click on the Internet and get involved on the Internet, and also through the ratings of quality confirmation. (Note that your website is not necessarily influenced if an assessor evaluates this. You give the examples of machine learning systems that then help to improve the systems as a whole.)

The vector search looks good is one thing. But what is more important is Do you really have the content that people satisfied more than their competitors and high quality, as described in the rater guidelines.

Nevertheless, I had a certain degree of success in the use of an understanding of similarity to Cosinus similarity to help some of my content in AI overviews. However, this is again a system that is constantly changing so that there is still no repeatable method that I know for ranking in Aios.

In some cases, my changes, which look better in relation to the similarity of the cosine, put me in an AIO ranking, and some time later these ranking lists were lost. There is more about ranking than just looking good for machines!

Pay attention to your main content

I have exercises in mine QRG workbook This allows you to determine what your main content is and whether it is appropriately placed on your side. The title is also part of the main content. Think for every page you have on your website: “What is the purpose of this page? How well satisfies a viewfinder that lands here? “

It can also help compare your website with competitors. Do you better do the viewfinder to the answer you are looking for?

Many websites affected by the helpful content system, and the Kern update in March 2024 also contained pages with many filler information that the reader needed to find the part that meets their needs.

Do you lack Eeat?

I understand So many Website owners who strive to evaluate content so much that they do not have the specialist knowledge or experience to write. I was recently contacted by a location owner who has all possible documents on questions of animal health. You are well written and look helpful, but if you look at the search results, you can find that you are dominated by relevant websites for pets.

No amount of rewriting and restructuring content helps this site rank.

EEAT is usually not something that you can fix with changes on the page. If you think you may have an EAT problem, I have homework for you. Read that Rater guidelines and Ctrl-F for “known for”. The word “call” is 213 times in the QRG. EEAT is not what they say about themselves, but what the world says online about them.

Again mine QRG workbook You can work with exercises to evaluate your website for EAT problems. There is also a link to a GPT that I have created with many advice and thoughts to improve your reputation online.

In some cases, however, it may not be possible to improve EEAT.

Be really original – in a way, seekers will be helpful to fulfill your intention

Gary Illyes from Google said at a recent event“Originality is something we will concentrate on this year. That will be important. “

The first advice that Google gives in yours Instructions for helpful content IS, “offers the content Original informationPresent Reporting, research or analysis? “

Once again I have homework for you: Read the rater guidelines and Ctrl-F for “originality” and also “paraphrased”.

If you want to publish a content, it must be something that complements knowledge of knowledge on the Internet. It is no longer enough to publish something similar to competitors, but a little better.

Uh! There it is. I knew that writing this report would help me with my advice on the website that I check that has been sinking since March -Kern -Update. Together we will brainstorming how you can add original content on your topic. I have so many ideas here.

Here is a good invitation for you to try your favorite -Lllm:

I have a website on the subject (topic). I am looking for ideas to write about original information, reporting, research or analysis.

Or you can use the helpful content brainstorming GPT in my QRG workbook. It’s so good! I’m doing it now to use it …

If you found this helpful, you will also enjoy my community – The search bar. I would love it if you join us!

This originally started as Maries notes Entry and transformed into an article.