Understand your users through research and query semantics

Lazarina says: “Learn how to incorporate and scale query semantics and user research, with the help of AI and machine learning.”

What are query semantics?

“Query semantics means understanding anything that relates to how users search.

It's not just the words that are being typed, but also the context of where they're being typed, the device being used, the platform people choose, and why they choose it in different situations. It's also different types of user signals, like trends or the user’s search history.

It’s understanding the specific query path and the user journey: how they travel from the point of wanting to get some information to the point of being satisfied with whatever result they get. This is the broad concept of query semantics, and it also includes search intent, information gain, knowledge graphs, entities, and all of this fun stuff.

It's a little bit technical, but not entirely. It's everything that we've been doing as SEOs for as long as we can remember. User research is obviously very related to this; knowing exactly who your users are, how they break down in terms of personas, and how these different personas carry out search journeys online.”

How do query semantics impact knowledge retrieval?

“They impact knowledge retrieval because Google uses all of these concepts in how they choose what information to serve.

Some of these data points we have access to, and some we do not. For instance, we can classify queries based on search intent. We have access to some queries, including synthetic queries now as well, but we don't have access to the user’s search history or the personalised context of the user. We don't have access to a lot of the data that Chrome collects, and that Google uses in the retrieval process to personalise the results that each user gets.

So, yes, they do impact search and how knowledge is retrieved. We can demonstrate this with the help of a lot of the patents that Google has released, but as SEOs and people working in marketing, we don't actually have the full picture.

With the help of AI, however, we can certainly get to a much better place in our understanding than we had five years ago.”

What are synthetic queries?

“Any query that has not been initiated by a person is a synthetic query. For instance, AI Mode’s query fan-outs are synthetic queries.

Of course, there will be some overlap between the queries that are generated by Gemini/AI Mode/ChatGPT (if it’s doing any research for you) and the queries that you get in the normal SERP when Google recommends next steps for your search, like featured snippets. There is some overlap with autocorrect as well.

Essentially, if the user initiates the query (i.e., they click on a featured snippet or within People Also Ask), that is a user-initiated query, even though it's assisted by Google giving the suggestion. However, if Google carries out query fan-outs in AI Mode, that's a synthetic query.

Both of these are important to understand because we're now working with AI systems as methods of search. It's important to understand the queries that they would carry out, and not only to think about the users, but also to think about the generative AI agents that people are using to get to information. We'll see how long that is relevant for.”

What different types of queries do people tend to use in traditional Google versus AI Mode versus ChatGPT, and other AI search engines?

“With the ability to contextualise a lot of the conversations that we have with ChatGPT, Gemini, AI Mode, etc., there is a lot of nuance that people share within the query. That's why it's important to be aware of that difference.

First of all, the queries that we submit to generative agents are a lot longer; a lot more context is shared, and a lot more personal information is shared. For instance, ‘I am doing a project on X, and your role is to do Y’.

It’s similar to prompt engineering, but when people use it for search, it's more like, ‘Give me hiking recommendations, but bear in mind, I only have these shoes. Also, give me tips about bringing my child with me – and we really want to see a waterfall!’ It's completely different from searching for ‘best hiking routes’.

People tend to bundle up queries into one prompt because they know that the generative agent is a lot more capable at handling that context and generating something that's really personalised. Now, the fallacy here is that people expect whatever is being generated to be completely truthful, and that they can completely rely on the plan that ChatGPT will give them. In practice, it doesn't yet work that way.

So, they are longer, and they include a lot more context and personal details. They describe more of what they're looking for, who they are, and why that's important to them. In traditional search, these might be broken down into different queries that follow a path – either in one session or broken down into separate sessions.

With AI systems, users can bundle those into one, and then the AI system does the research and spits out a response. Those are some of the key differences.”

Why is it important to understand how different search engines work and where they pull semantic data from, to map against semantics?

“It means you can understand where different search engines get their information, how they are sorting through results, and how they are building a corpus of documents that they pick answers from.

When we talk about traditional search, the approach is different from generative search. Here, the importance is to understand how queries are generated and whether the approach is probabilistic or deterministic. Is there a right answer, or is there a series of answers? You want to learn more about embeddings as well.

Semantics is an overarching umbrella term for understanding text and the meaning behind text. It's more about thinking about the different ways that text is being used and also combined with other metrics that Google has: from Chrome and the type of device people are using to where they're making the search from, how often they search for this topic, etc.

Previous preferences about specific content types also come into play. We know from patents that Google suggests topics to people and different content pieces that have resonated with them in the past, in terms of formatting and style.

Long story short, search is very personalised – not only traditional search, but AI search even more so. It's important to understand the context and the semantics behind all of these signals, and how to analyse text at scale.”

Is understanding additional concepts like the impact of context, personalisation, journey mapping, and personas related to personalised search results within AI platforms?

“It's also about traditional search. It's everything.

My main advice is to consider keyword research in the traditional form as almost obsolete. You need to move away from traditional thinking, where you look at terms, and move towards thinking in terms of concepts. Which concepts are relevant for your brand to compete in? Who are the people that you're trying to attract? Where are they?

It might make sense for your brand not to even try to compete in generative agents, because your audience isn’t there. Instead, you might want to rethink your content strategy around the changing audience that you have.

If you haven't already, it’s important to start thinking about these things because they are making search more personalised, and competition is increasing with the ability for us to scale content.

It's important to know how you are positioned, what makes you stand out, where to find the people that you're speaking to, and present yourself in a way that's relevant to them in their search journey – not bombard them everywhere and wonder what will stick.”

How do you make sure that the right piece of content is surfaced at the right time in the user journey?

“Start with the research to see, first of all, what type of content is presented at different stages, to understand the concepts that people search for and how they typically structure query paths within one session or within separate sessions.

Once you have that information, you will be able to reverse engineer from what Google presents in traditional search via the SERP snippets and the competitors that are presented. What are the different articles that are being surfaced? What is the format that people prefer for these types of searches?

With AI Mode, that's a little bit more complicated because the results are so much more personalised, but you can still reverse engineer based on the type of query. What's important here is to think about the concepts (we’re talking about entities here) and your topical map (which is something that can now be automated to a large extent), and then take this entity topic map and make it really personalised to your audience.

Every query you traditionally run through keyword research can now be contextualised with a tonne of nuance about your audience, what their pain points are, and what's really relevant to them.

This is where user research is very important, because the human element and that connection to the real problems your users have is going to be what makes your brand stand out – in both AI search and traditional search.”

How do you scale your semantic keyword research to ensure that you can do this effectively and efficiently on a website with a large number of pages?

“There are a few different approaches.

First, you can break down different tasks and different parts of understanding query semantics, and automate separate tasks with custom-trained APIs. For instance, if you're looking at entity analysis, you can use Google's Natural Language API or Amazon Comprehend: an API that's trained specifically for this task.

After that, you can use Google's Knowledge Graph API to understand what entities are related to that particular entity. You could then understand, for example, that when users are searching for X type of information, the next thing that is close to you and your brand is Y entity. Then, you could internally link articles based on that, include it in sections, and so on – basing your content strategy around all of these steps.

If you want to think about user search query paths, you can take those from featured snippets. What next steps does Google recommend for the user journey? What are the related queries, People Also Ask, what to search for next, etc? You can also take some query data from synthetic query-mimicking agents. I know LOCOMOTIVE has one, and Dan Petrovic is using a custom-trained one, which he will hopefully release.

You can take all of these different components to build a system. Your semantic keyword universe should include not only the terms that you get from tools like Semrush, but also your entities, paths, suggested FAQs, and how those are approached for your specific audience. You should also expand your entity map with attributes/variables that you may not be seeing in the queries, but you know your users are asking for.

Then, based on all of this information (if it lives in a database or a spreadsheet), you can build a program/script to pull that data into separate content briefs and topical maps.

I know I'm throwing a lot of information at people. There are resources that you can find online for all of these steps. Definitely consider looking more into these things because they're more relevant in 2026 than ever before.”

Lazarina, what's the key takeaway from the tip you shared today?

“Invest in understanding your users better through the way they search online.”

Lazarina Stoy is Founder at MLforSEO. Find out more over at LazarinaStoy.com.