Keyword Research: The Complete Practical Guide for 2026

In short: keyword research is the process of identifying what your audience actually searches for, how much demand each term carries, what intent sits behind it, and which page on your site should answer it. It is not a word list, it is a demand map that turns into site structure and a content plan. The practical process is seven steps: map what the business sells, collect terms broadly, measure demand, classify by search intent, assess competitiveness realistically, group into clusters, and map clusters to pages. In 2026 there is an eighth step: making sure the page also answers the sub-questions AI engines generate from the original query (query fan-out). An Ahrefs study published on 2 March 2026, covering 863,000 keyword SERPs and 4 million URLs cited in Google AI Overviews, found that only 37.9% of cited sources came from the top 10 results of that query, down from roughly 76% in July 2025. Research that looks only at the head term is no longer enough.

What keyword research really is, and why it is not a word list

Keyword research is the process of collecting, measuring and classifying the terms people type into search engines in your category, and then deciding what to build on your site and in what order.

The most common mistake is believing the deliverable is a spreadsheet with two hundred terms and their search volumes. That is the raw material, not the output. The real output of good research is a decision: which pages already exist, which are missing, which page owns which group of terms, and what the priority order is for the next six months.

That distinction is also the difference between research that produces results and research that stays in a folder. A site where every page was built to answer a defined need earns relevant traffic over time. A site built around a word list ends up with near-identical pages competing against each other, and none of them rank. If you are just starting out, read what SEO is and why your site needs it first, then come back here.

The four types of search intent, and the page that fits each one

Before volumes, before tools, before anything else, you need to understand what the searcher is trying to achieve. The same term can require a completely different page depending on intent. This is the classification that determines what to build:

The practical rule: an informational term does not belong on a service page, and a transactional term does not belong in a blog post. If you push a commercial term into a tutorial, Google sees that the page does not match what users expect and you pay for it in rankings. The cheapest way to check intent is to search the term and look at what type of result ranks: if all ten results are guides, Google has already decided the intent is informational.

What changed in 2026: keyword research in the AI search era

This is the part most guides have not been updated for, and it changes the process in practice.

AI-driven search engines and chat assistants do not search the query as typed. They break it into several sub-queries, run each one separately, and assemble an answer from the sources that cover each part. That process is called query fan-out. It means your page can be cited because of a sub-question that never appeared in your classic keyword research.

The data on this already exists and is not speculation:

What do you do with this? Add a step you can call sub-question mapping: once you have picked a core term, ask which sub-questions an AI engine would generate to answer it fully, and make sure the page answers each one explicitly, under its own subheading, with a direct answer. The simplest free way to find them is to combine the "People also search for" block at the bottom of the results page, Google autocomplete, and the questions customers actually ask you on the phone.

It is worth saying plainly: keyword research did not die, it expanded. The unit of research is no longer a single word but a topic, a question and an entity.

The full process: seven steps to keyword research

Step 1: start from the business, not the tool

Before opening any tool, write down what you actually sell: services, products, problems you solve, audiences and geographies. This is the seed keyword list everything else derives from. A provider who skips this step ends up with a generic list that fits every competitor in the market.

Step 2: collect broadly, without filtering

During collection the goal is quantity, not quality. Every term that comes up goes on the list, even if it looks too small or off-topic. Filtering happens later, once there is data. Filtering too early is the most common reason research misses exactly the profitable terms.

Step 3: measure demand

Now push the list into a tool and check average monthly search volume, trend over time and seasonality. Note that search volume is a yearly average, so a seasonal term can look weak on average yet be enormous in the two months that matter.

Step 4: classify by intent

Every term gets tagged using the table above. This is the step that turns a list into a map. Without intent classification you cannot tell whether a term with 2,000 monthly searches deserves a service page or a paragraph inside a guide.

Step 5: assess competitiveness realistically

A difficulty score in a commercial tool is a starting point, not a verdict. The real test is to open the results page and ask: who ranks there, how deep is their content, when was it last updated, and what is missing from it. If all ten results are large, established sites, the head term is not your entry point.

Step 6: group into clusters

Terms that share intent and return roughly the same results are the same term as far as Google is concerned, so they need one page. "keyword research", "how to do keyword research" and "keyword research in Google" are one cluster, not three pages.

Step 7: map to pages and set priorities

Attach a page to every cluster, existing or new. Then order by a combination of demand, proximity to revenue and likelihood of winning. An existing page sitting at position 14 is usually worth more than a new page starting from zero.

A worked example: an online jewellery store

Look at the last row. A high-volume term with no purchase intent is exactly the kind of decision good research is meant to prevent: traffic that does not generate enquiries costs time and money. Both paid and organic rest on the same decision, as our guide to ecommerce site promotion shows.

Eight places to find terms, even without paid tools

• Google Keyword Planner - the official source for search volumes, free with a Google Ads account.
• Google Search Console - shows the real queries your site already appears for. The most accurate source you own.
• Google autocomplete - type the term and do not press enter. The suggestions are real searches.
• "People also ask" and "People also search for" - a free map of the sub-questions around the topic.
• The search terms report in a paid campaign - if you run ads, this is the richest source available: real searches with conversion data attached.
• Competitor sites - page titles, navigation and category structure reveal which terms they built pages for.
• Google Trends - for seasonality and trend. We covered this in our guide to using Google Trends for advertising.
• Sales and support calls - the questions customers ask on the phone are real human phrasing, and usually exactly what gets typed into search.

The best tool you already own: Search Console and striking distance

This is the method we run for clients before touching a paid tool, and it almost always produces the fastest win.

In the Search Console performance report, enable the average position metric alongside clicks and impressions, then sort by impressions. Now look for rows with many impressions, very few clicks, and an average position between 11 and 30. These are terms Google is already willing to show you for, but you are sitting on page two.

The advantage of this method over any external tool is that these are not estimates but real data about your own site: queries Google has already associated with you. Instead of guessing at a new term you would start from zero on, you identify a term that already has momentum and push it onto page one.

The tool table: what each tool is actually good for

Check a term yourself

You can type in English or Hebrew:

Our practical advice: do not buy a tool before you have exhausted the free ones. Keyword Planner, Search Console and Google Trends together are enough to build serious research for a small or mid-sized business. Paid tools pay off when you manage many sites, need systematic competitor analysis, or work in an unusually competitive market.

How to read search volumes correctly

Before making decisions on a number, understand what it actually says. Google Ads documentation states that historical stats such as average monthly searches are shown for exact matches only, and that search volume statistics are rounded. That is why the same term shows one number in Google's tool and a completely different one in a commercial tool.

• Volume is a 12-month average. Always check the monthly graph before drawing a conclusion about a seasonal term.
• Wide ranges signal an account with no spend. A Google Ads account with active spend receives more granular data.
• Volume is not profit. A term with 200 searches and buying intent is worth more than one with 20,000 searches and learning intent.
• Zero searches does not always mean zero demand. Many long-tail phrases sit below the measurement threshold and still bring traffic.

Keyword research in Hebrew: problems that do not exist in English

Most guides were written for the English market and skip what makes research in Hebrew harder. These are the gaps we hit on every project, and the same logic applies to any non-English market:

That last point matters most: it is easy to reject a term because it shows 90 searches a month, while the cluster as a whole generates 700. Always look at the sum, not the single row.

Organic research versus paid campaign research

We run both organic SEO and paid campaigns, so we see the difference up close. The two processes start in the same place and end differently:

The most practical thing you can do with this: run a small Google campaign for two weeks on a cluster you are considering promoting organically. Within days you will know which terms produce enquiries and which only waste clicks, and you take that into the organic content plan. It shortens the feedback loop by months. For more on the paid side, see our guide to paid advertising types and real costs, and our article on Quality Score, which explains how the match between term, ad and landing page affects what you pay.

Clusters and page mapping: how not to cannibalise yourself

Internal cannibalisation is one of the most common problems on sites that produced a lot of content without a map. It happens when two or more pages target the same search intent, forcing Google to choose, hesitate, and sometimes rank both of them low.

• The clear signal: in the Search Console report, two URLs alternate getting impressions for the same query.
• The rule: one cluster, one page. If two terms return nearly the same ten results, they are one cluster.
• The fix once it has happened: pick a primary page, merge the good content from the second into it, and either redirect the second or turn it into a supporting page that links to the primary.
• The right structure: a broad pillar page for the core cluster, surrounded by supporting pages for specific questions, all linking to the pillar and back.

That structure is what builds topical authority, and it is also what raises your odds of being cited in AI Overviews, because it covers more sub-questions around the same topic.

How to know the research worked

Keyword research is not a one-off project but a living document. These are the metrics we check, in the order they move:

The last row is the only one that really counts. Organic traffic that does not turn into enquiries usually means the research produced informational terms only, or that the page they landed on was not built to convert. We covered that side in our landing page guide and in our article on user experience and SEO.

Eight mistakes that repeat in almost every research project

Who we are and why it matters here

At SFB Digital Marketing we run organic SEO and paid Google advertising for businesses in Israel and abroad, with over 11 years of experience and more than 50 million shekels in managed media budgets. That combination is why the research we build rests on both organic search data and real conversion data from paid campaigns, rather than on external tool estimates alone. Want us to build the research and content map for your site? Talk to us, or take a look at our SEO packages.

Frequently asked questions

How long does keyword research take?

Basic research for a small business with a single service takes 3 to 6 working hours. Full research for an ecommerce site with dozens of categories, or a business with several service lines, takes one to two weeks including competitor analysis, intent classification and page mapping. If someone promises full research in an hour, that is a tool export, not research.

Can keyword research be done for free?

Yes, and to a good standard. Google Keyword Planner, Google Search Console, Google Trends, autocomplete and "People also ask" together provide most of what a small or mid-sized business needs. Paid tools such as Semrush or Ahrefs save time and add competitor analysis and difficulty scoring, but they are not a prerequisite for starting.

How many keywords should each page target?

One page targets one cluster, meaning one core term plus the group of variants and sub-questions around it. In practice a good page ranks for dozens or even hundreds of terms, but all of them share the same search intent. There is no fixed number of repetitions and no magic keyword density percentage.

What is the difference between a keyword and a key phrase?

In practice there is none. A keyword is usually a single word or short combination and a key phrase is longer, but both terms refer to the same search query. They are used interchangeably, and Google treats them the same way.

Is keyword research still relevant in the AI era?

Yes, but it has expanded. AI engines break every query into sub-questions, so a page has to cover a whole topic rather than a single term. According to a Surfer SEO study from December 2025 analysing 173,902 URLs, a page ranking for both the head term and at least one fan-out query was 161% more likely to be cited in an AI Overview. The conclusion is not to stop researching keywords but to add a layer of questions and entities.

How often should research be refreshed?

Core clusters, every quarter. The whole site, once a year or whenever the business changes materially, such as adding a product line or entering a new market. On top of that, open the Search Console performance report monthly and look for new queries that started collecting impressions without you planning for them.

What if every term in my industry is too competitive?

Enter from the side. Instead of the head term, pick long terms with specific intent: by problem, by audience, by geography or by stage in the buying process. Build a cluster around them, accumulate topical authority, and only then target the head term. In parallel, a focused paid campaign can cover the head terms while organic matures.