Keyword Research in 2026: How to Find Topics Your Website Is Missing

A few years ago, an article about keyword research would almost inevitably begin the same way: choose a core topic, open a keyword tool, collect a few hundred or a few thousand phrases, sort them by search volume, and build a semantic core.

I worked with lists like that for many years too. They have not disappeared, and they are still useful. What has changed is the way I look at them.

SEO in 2026 is in a rather unusual place. At professional conferences, the discussion is increasingly no longer limited to how to reach Google’s Top 3 or Top 10. A growing part of the conversation is about something else: how to make a website appear in AI Overviews, ChatGPT, Gemini, and other systems that generate an answer for the user instead of simply presenting a list of links.

At first glance, that might suggest that traditional keyword research is becoming less important. I would argue the opposite. We are finally seeing more clearly what it was supposed to be used for in the first place.

Once you look at keyword research this way, most of the familiar tools remain useful. We simply use them differently.

I No Longer Start With a List of a Thousand Keywords

A large semantic core always looks impressive, especially when the spreadsheet contains search volume, competition, CPC, current rankings, keyword difficulty, and several carefully colour-coded columns.

The problem is that the number of rows tells us very little about what should actually change on the website.

A hundred similar queries may require one good page. Two almost identical phrases may require two completely different pages if the intent behind them is different.

In the first article in this series, we looked at what keywords are and how search intent changes the meaning of similar phrases. In the second, we focused on why even a carefully collected keyword list can fail when queries are grouped incorrectly, pages compete with each other, or copy turns into a collection of exact-match phrases.

Here I am interested in a different question: where do we find the topics the website is genuinely missing?

In my own work, I usually start with three sources: Google Search Console, Google’s own suggestions, and Google Keyword Planner. Only after that does the most important part begin — checking the SERP and deciding whether a new page is needed at all.

1. Google Search Console: Let the Website Show You What Is Missing

For a website that already has some search history, Search Console can often be more useful than a third-party keyword tool.

The reason is simple. Keyword Planner or an SEO platform tells you what people search for in general. Search Console tells you which queries Google already associates with your website.

I often look for queries that receive a large number of impressions but very few clicks. If Google has already shown the site hundreds or thousands of times for a particular phrase, there is clearly some thematic connection. But if almost nobody clicks, it is worth finding out why.

Sometimes the explanation is simple: the page is ranking on page three or four. In another case, the position may be reasonable but the title does not match the query particularly well.

The more interesting situation is when Google keeps showing a page that only touches the topic briefly because the website has no better answer.

That is the scenario I find most useful. A long article, for example, may contain one paragraph that briefly answers a particular question. Google notices that fragment and starts showing the page for the query, even though the article was written for a completely different purpose.

At that point, a natural question appears: does this topic deserve a dedicated page of its own?

Some of the best content ideas come from situations like this, not from another list titled “100 Popular Keywords in Your Industry.”

A Lot of Impressions Does Not Automatically Mean “Write a New Article”

It is easy to overcorrect here. Seeing 10,000 impressions and 20 clicks does not mean that a new page should immediately be created.

First, I want to know which page is receiving the impressions, where it ranks, whether its title matches the query, and what Google is actually showing near the top of the SERP. Then I check whether the phrase represents a genuinely separate intent and whether a new page would compete with something that already exists.

If the topic is already covered well and the page simply cannot move higher, content may not be the problem at all. Technical quality, competition, domain authority and backlinks may be limiting its ability to compete. We have a separate explanation of what SEO links are and why backlinks can still influence a page’s ability to rank.

So Search Console does not give me a ready-made article title. It gives me a signal worth investigating.

2. Google Autocomplete: See How Users Expand a Topic

The second tool does not require a subscription or even a separate interface.

I simply start typing a topic into Google. The search engine immediately suggests continuations. To an ordinary user, they are just convenient autocomplete suggestions. To someone working with content, they can reveal useful directions in which people develop the original topic.

If the starting topic is SEO, for example, Google may suggest variations involving cost, timelines, DIY promotion, specific tools, or common problems.

Again, I do not copy everything into a spreadsheet with the assumption that every phrase must somehow appear on the site. I ask a different question: does the website already provide a good answer to this need?

If it does, fine. If the issue is mentioned only briefly, perhaps the existing material should be expanded. If there is no answer at all, we may have found a useful new direction.

Autocomplete still does not define the structure of the website. People may search for “how to choose keywords,” “how to find keywords,” “SEO keyword research,” and “how to choose keywords for a website.” These are different phrases, but they may all represent exactly the same need.

If that is the case, one strong article is usually better than four weak ones. This is the point where keyword research begins to turn into topic research.

3. People Also Ask: What Does the User Want to Know Next?

It is also worth looking beyond autocomplete and paying attention to related questions in the SERP.

They show how a topic develops in the user’s mind. Someone starts with one question, receives an answer, and immediately has another. That is why a strong page rarely answers only one line from Keyword Planner.

A topic such as keyword research naturally leads to questions about search volume, search intent, clustering, how many keywords belong on a page, when a separate article is needed, and how the results should be evaluated later.

If all of those questions belong to the same intent, they can form the structure of one substantial article. If one question consistently produces a different type of SERP and represents a different need, it may deserve its own page.

This is also why I do not particularly like the practice of building an article outline mechanically from the H2 headings of competitors. A question should not be included merely because several competitors mention it. It should make sense as the next step in the user’s journey through the subject.

4. Google Keyword Planner: Not the Beginning, but a Way to Expand a Direction

I use Google Keyword Planner somewhat differently from the way it was often presented in older SEO guides.

I do not open it with the goal of collecting the maximum possible number of phrases. Usually, I already have a topic. Search Console, Autocomplete, or the website itself has pointed me toward it.

Now I want to understand how else people describe the need, how much demand exists, where commercial intent appears, and whether one broad phrase is actually hiding several different needs.

A broad query may have very high search volume while more specific versions reveal different products, buying intent, informational intent, or different use cases.

In the previous article, we used this kind of example to show why high search volume does not automatically mean high business value. Here the same data answers a different question: it shows how one broad topic begins to split into several distinct directions.

Some queries describe a product type. Others reveal an intention to buy. Others are clearly informational. Only after seeing that distinction can we decide what the website actually needs: a product category, a landing page, an article — or nothing new at all.

5. The Most Important Tool Is Still Google Itself

After Search Console, Autocomplete and Keyword Planner, I still return to ordinary Google Search.

No spreadsheet answers this question as clearly: what does Google currently consider a good answer to this query?

If the top results are articles, that tells us something about intent. If the SERP is dominated by service pages, ecommerce categories, calculators or videos, that tells us something different.

Two queries can differ by a single word and still produce substantially different results.

“Web design”

A broad query can produce a mixed SERP with informational content, tools, inspiration, courses and service providers.

“Web design cost”

Adding a single word can shift the results toward pricing guides, agencies, budgets and commercial pages.

This is why I would never treat automatic keyword clustering as the final decision. A tool sees similarity between words. Google gives us a much better indication of the type of answer users appear to expect.

6. One Keyword Does Not Mean One Page

This is probably one of the biggest changes in the way I now think about SEO.

When a keyword tool produced a long list, the natural question used to be: how many pages do we need to create for all of this?

Today, I would ask something else first: how many genuinely different intents are represented here?

Consider three queries: “what are keywords,” “keyword mistakes,” and “how to choose keywords.” They all belong to one large topic, but the person behind each query is trying to solve a different problem.

The first wants to understand the concept. The second is diagnosing a problem. The third wants to perform a practical task. That is why, in our own blog, they naturally became three separate articles.

Ten variations of “how to choose keywords,” however, do not automatically deserve ten different URLs.

This principle is especially useful before a new website is launched. If we already understand that several intents need different page types, the future structure can be planned around them from the beginning instead of being patched later. That is one reason a good term of reference for website development should describe not only design and functionality, but also the logic of the future sections and page types.

7. Before Creating a New Page, Check Whether the Answer Already Exists

This step becomes particularly important on older websites.

After several years, a blog can accumulate dozens or hundreds of articles. It is surprisingly easy to forget that the topic you have just “discovered” was already discussed five years ago.

Creating another page may then make the situation worse rather than better.

Sometimes I find a promising query and discover that the website already has an older page with history, backlinks, indexation and existing rankings, but the content itself is outdated. In that situation, updating the existing page usually makes far more sense than creating another URL.

In fact, that is exactly how this article came about. The old page about search queries had existed for years. But the way I now approach the subject has changed so much that a cosmetic update would not have been enough. The URL and its history can remain. The article itself needs to be rebuilt.

On large or long-neglected websites, this kind of check is easier to do systematically. A thorough website audit can reveal not only technical problems, but also duplicate intent, weak landing pages, outdated content and areas where the site already has visibility but is not using it effectively.

8. How I Decide Which Topic Should Come First

After all the checks, there are usually still more possible topics than there is time to produce them. So priorities matter.

I would definitely not sort the list by search volume alone. I would look at several signals together: how closely the topic relates to the business, whether the site already has some visibility, whether there is a genuine information gap, how realistic the competition is, and whether the content can attract the type of user the business actually wants.

Informational traffic by itself is not the final objective of a commercial website. A good topic should at least have a logical connection to the product, service, expertise, or trust the company is trying to build.

9. In 2026, We Are No Longer Researching Keywords Alone

This is where things become especially interesting.

Traditional SEO was built around a fairly simple model: there is a query, a search volume, a page, and a ranking position.

Today, users increasingly receive an answer before visiting a website at all. Search engines and AI systems can synthesise information from several sources and present the explanation directly.

This is no longer only a conference topic. Google now provides a separate Generative AI performance report in Search Console. According to Google’s own documentation, the report measures organic impressions from generative AI features in Search, including AI Overviews and AI Mode, and these insights were rolled out worldwide as of August 31, 2026.

That changes the question a website owner has to ask.

The traditional question was: How do I get this page into the Top 10?

Now another question sits beside it: Is this page good enough to be used as a source when an answer is being generated?

That does not mean keywords, rankings or technical SEO have become irrelevant. A page will not become a strong source simply because someone labels it “AI-friendly.” It still needs relevance, crawlability, clear structure, useful information, and enough authority to be taken seriously.

But the emphasis is shifting from the isolated phrase toward the completeness and quality of the topic.

10. From Keyword Research to Topic Research

That is why I increasingly prefer to think in terms of topic research, not just keyword research.

Take the query “how to choose keywords.” An older approach could easily become: find the exact search volume, collect several variations, place them in the title, H1 and copy, then start promotion.

Today, that is not enough.

We need to understand what the person really wants to know, what questions they are likely to ask next, where they may make mistakes, what tools they need, how they should evaluate the result, and what they are supposed to do with the list they eventually collect.

A good page then stops answering one row in a spreadsheet. It answers the subject.

That kind of content has a better chance of remaining useful regardless of where the user eventually encounters the answer — in a traditional Google result or through an AI-driven search experience.

11. Can AI Help Find New Topics?

It would be strange to discuss keyword research in 2026 without mentioning another tool that is already useful in this process.

An AI system can be given a website structure, a list of existing articles, and the general subject area, and asked to identify possible gaps: what is already covered well, which questions are still missing, which topics overlap, and which adjacent areas may deserve further investigation.

That can be very fast. In a few minutes, AI can produce directions that might have taken considerably longer to find manually.

But one word matters here: hypotheses.

I would not treat a list produced by AI as a finished semantic core. AI may suggest a topic that sounds perfectly logical but has almost no real search demand. It may miss business-specific context. It can also recommend content that the website already covers under a different title.

So after AI analysis, I still want Search Console, Keyword Planner, the actual SERP, and ordinary professional judgement.

AI is becoming a very good tool for finding directions. It is not a substitute for validating them.

12. Turning Search Queries Into a Real Content Plan

At this stage, all the sources begin to work together.

Search Console shows which topics Google already associates with the website. Autocomplete and related questions show how users expand those topics. Keyword Planner gives us additional wording and an idea of demand. AI may identify gaps we did not notice ourselves. The SERP tells us which form of answer Google currently prefers.

Then I return to the website.

Only now do I decide whether to update an old page, expand an existing one, create something new, merge several weak articles, or do nothing at all.

A content plan should not live separately from the website itself. New pages have to be published, internally linked, updated, checked after CMS changes, and kept technically stable. On websites without an in-house team, that often becomes part of ongoing website support and maintenance rather than a one-time task after the content is written.

The last option — doing nothing — matters more than it sounds. Not every keyword deserves a new article.

Instead of a Conclusion

When I look back at older SEO articles today, what stands out is not necessarily that we used completely different tools.

Most of those tools are still here.

Search Console still works. Google still suggests queries. Keyword Planner still shows demand. We still analyse the SERP.

What has changed is our understanding of why we are doing all of this.

There was a time when a large spreadsheet with thousands of rows could look like the successful outcome of keyword research.

Today, I find a different result much more valuable: after the research, I understand which answers the website is missing and which pages should be created or improved.

Where the user eventually sees that answer — in traditional Google results, an AI Overview, or an AI-powered search system — becomes the second question.

First, the website needs content that is genuinely worth showing, citing, or using as a source.