Keyword Research: The Practical 2026 Step-by-Step Guide

What Keyword Research Really Decides

Search for "keyword research" and you get a definition: the practice of finding the words people type into Google. That is true, and almost useless. What the process actually decides is four things about your business: which pages get built, which ones you deliberately skip, the order you build them in, and where your writing time goes. The keyword list is the by-product. The decisions are the work.

Every term you consider has to answer three questions before it earns a page:

• Who types this, and how close are they to buying? Someone searching "what is a chatbot" and someone searching "chatbot integration agency" sit at opposite ends of a buying decision. Both are valid targets. They belong on different pages, with different calls to action.
• Can you answer it better than the pages already ranking? If the top results are thin, outdated, or written for another country, there is room. If the top ten are ten excellent guides from major brands with teams behind them, there probably is not, at least not this quarter.
• Does it have a logical home on your site? A keyword with no page to live on becomes a random article that links to nothing and converts nobody. If you cannot say which page serves it, you are not ready to write it.

One thing worth saying early, because it shapes everything below: for a small or young site, search volume is the least useful number in the process. It gets the most attention because it is the easiest to look at. Difficulty and relevance decide far more about whether a page ever pays back the hours it took to write.

Start With Seeds, Not With Tools

The most common way this goes wrong is opening a keyword tool first. You type something broad, receive a list of eight hundred phrases, feel productive, and end up writing about whatever had the biggest number printed next to it. That is how sites end up with content that has almost nothing to do with what they sell.

Start with seeds instead, collected from places that reflect real demand:

• Your customers' own words. Sales calls, support emails, chat transcripts, contact form messages. The exact phrasing someone used to describe their problem is a keyword, and usually a better one than anything a tool suggests.
• Your objections. Every objection you answer on a call is a search: "is X worth it", "X vs Y", "how long does X take", "how much does X cost".
• Autocomplete and People Also Ask. Type your topic into Google and read the suggestions. Then read the questions Google expands at the bottom of the results page. That is Google telling you what people ask next, for free.
• Your competitors' structure. Their navigation, their service pages and their sitemap show how they have chosen to divide the market. You are not copying. You are checking whether they have already found demand you have missed.
• Your own Search Console. The strongest seed source you have. Queries where you already appear on page two with a handful of impressions are proven demand pointed at your site, and most of them are quick wins.

From those seeds, build a working list of 100 to 300 candidate phrases. Do not filter by volume yet. Volume filters are excellent at deleting the specific, low-volume phrases that later turn into a small site's best pages, and at keeping the broad ones you will not rank for this year.

Search Intent Comes Before Volume

Intent is what the person wants, which is not always what the words look like they want. The same three words can carry completely different intent depending on context, and the results page tells you which one dominates. Read the SERP before you open any keyword tool.

• Informational: the person wants to understand something. The top results are guides, tutorials, definitions, videos. Convert with a clear next step, never a hard sell.
• Commercial investigation: the person is comparing options and is close to a decision. The results are comparison pages, reviews, "best X" lists, pricing pages. For a service business this is usually the most profitable intent on the list.
• Transactional: the person is ready to act. The results are product pages, service pages, booking forms. Send them straight to the page where the action happens.
• Navigational: the person wants a specific brand or destination. Mostly irrelevant unless it is your own brand name, which is always worth holding.

The results page is your brief. If the first page is ten long-form guides, a short opinion piece will not rank, however well written it is. If it is ten videos, you need video. If it is a tool page plus two tutorials, the term has mixed intent, and the honest move is to decide which half is more valuable to you and serve that properly rather than half-covering both.

Modifiers do much of this work for you. "How to", "best", "versus", "cost", "for beginners", "near me" and "for small business" each shift intent in a predictable direction. Two phrases that look nearly identical in a spreadsheet can deserve two different pages, or one page that answers both questions in order.

Reading Volume and Difficulty Honestly

Search volume is directional data, not a measurement. Any tool's monthly number is an average across a year, rounded, modelled, and in some cases estimated from advertising data rather than counted. Treat 1,000 as "a meaningful number of people", 100 as "a small but real audience", and 10 as "too small to judge". Never build a strategy on the difference between 880 and 1,000.

Two factors matter as much as the headline number:

• Country and language. A term that pulls thousands of searches in the United States can pull a few dozen in your market. Always run your target country rather than the global default.
• Seasonality. Some terms spike and collapse. A trend graph costs nothing to check and stops you from reading an annual peak as normal demand.

On difficulty, the honest position is that third-party scores are estimates. They are built from the link profiles of the pages currently ranking, which makes them reasonable as a relative scale and unreliable as an absolute one. A score of 30 does not mean you will land in third position. It means the pages there are weaker than the pages sitting at 60.

So use difficulty as a first filter, then look at the results page yourself. Two checks take a minute each and tell you more than any score:

• Who is ranking? Forums, community threads, directories and thin pages in the top five are an opening. Nine results out of ten from major brands with dedicated SEO teams is not.
• Are the results actually good? A page ranking first for a phrase it barely answers is a page you can displace with a better one. Weak results at the top are the clearest opportunity in SEO.

Long Tail Keywords: Where Smaller Sites Win

Long tail keywords are the longer, more specific phrases: fewer searches each, far more of them in total, and usually the only realistic entry point for a site without a strong link profile. Four to seven words, often with a modifier attached. "Keyword research" is a head term. "How to do keyword research for a local service business" is long tail.

Three things make them work:

• They are winnable. The competition for a specific phrase is a fraction of the competition for its two-word version, and the pages ranking are often generic.
• They convert better. Someone typing seven words knows exactly what they want. Specificity in the query is specificity in the intent.
• They build the authority that later wins the head term. Twenty useful pages around one topic make your site the obvious source for the broad term as well. Head terms are usually won as the final step of a cluster, not by attacking them directly.

One caveat worth keeping: "low volume" and "no volume" are not the same thing. Plenty of specific phrases show as too low to display in a tool and still send steady traffic, question phrases especially. Judge those on whether the phrase sounds like something a real person would type, whether it appears in autocomplete, and whether Google expands it in the questions section of a results page. If it fails all three, skip it.

Local modifiers deserve their own cluster if you serve a city or region, because geography changes which pages rank and the signals Google wants are different enough to be worth learning separately. If that is your situation, read this alongside a dedicated guide to local SEO.

Clustering Keywords Into Pages

The mistake I see most often is one keyword per page. It sounds organised and it produces a site with sixty thin pages, several of which compete against each other for the same query. Google then has to pick between them, and it usually picks neither for long.

Cluster instead. The rule is to group by intent, not by how similar the phrases look:

1. One primary keyword per page. The phrase you want that page to own. It goes in the title, the H1, the URL and the opening paragraph, once each. Repetition beyond that adds nothing.
2. Five to twenty supporting terms per page. Synonyms, sub-questions, related comparisons, the same phrase with a city attached. These do not need pages of their own. They need headings and a paragraph of honest coverage.
3. Group phrases that answer the same question. "How to do keyword research", "keyword research process" and "keyword research steps" are one page. Splitting them creates three weak pages instead of one strong one.
4. Split only when the intent genuinely diverges. "Keyword research tools" and "keyword research for ecommerce" each deserve their own page because the job to be done is different. That is a real split, not a cosmetic one.
5. Check existing pages before building new ones. In most audits, a third of the planned new content can be handled by improving a page that already earns impressions. Updating a page Google has already crawled is faster than waiting for a new URL to be discovered.

Once the cluster is mapped, order the work by opportunity rather than volume: existing pages with impressions first, then terms with low difficulty and clear commercial intent, then everything else. Write the mapping down. Three columns are enough, primary keyword, supporting terms and target URL, and they stop you from writing the same page twice six weeks apart.

The Tools and the Workflow

You need fewer tools than the industry suggests. Four cover almost everything:

• Google Keyword Planner. Free with a Google Ads account, and you do not have to run ads to use it. It is the closest thing to first-party volume data and its country and language filters are reliable. Very low-volume terms show as ranges rather than numbers, which is a real limitation.
• Google Search Console. The only place where query data about your own site is accurate. Sort by impressions, then look at the queries sitting between positions five and twenty. Those are the pages where a modest improvement pays immediately.
• A paid platform, if the budget exists. Ahrefs or Semrush for difficulty scores, competitor gap analysis and link data. Useful, not mandatory. Export the gap, close the tool, then do the thinking yourself.
• The free sources nobody uses properly. Autocomplete, People Also Ask, Google Trends for seasonality, YouTube suggestions for video phrasing, and forums for the words people actually use instead of the words marketers use.

A 90-minute session that produces a plan

1. 0 to 15 minutes: pull the data. Export Search Console queries with impressions for the last 90 days, and collect ten seeds from customers, competitors and autocomplete.
2. 15 to 45 minutes: expand and de-duplicate. Run the seeds through Keyword Planner for your country. Merge near-duplicates. Delete anything unrelated to what you sell.
3. 45 to 75 minutes: check intent and competition. Search each cluster's primary phrase. Note the format of the top results and whether the leaders are beatable.
4. 75 to 90 minutes: assign and prioritise. Write the mapping: primary keyword, supporting terms, target URL and a rough order of work. Anything without a URL is not a priority, it is a note to yourself.

Done monthly, that rhythm keeps a content plan honest. When you would rather have the map built against your own site and search data, that is the first piece of work I do as an SEO consultant in Dubai, and the wider picture is covered in my complete SEO guide for 2026.

Measuring and Refreshing

The work does not end when the page is published. It ends when you decide the page is finished, which is either when it ranks and converts, or when you confirm it never will and redirect or rewrite it.

Track three things and ignore the rest:

• Impressions and clicks by cluster, not by page. A group of twelve pages serving one topic is a unit of progress. Individual positions move for reasons that have nothing to do with your work.
• Which queries a page earns clicks for. A page ranking for fifteen phrases you never targeted is a page that should be expanded and retitled. Search Console tells you which ones, so look.
• Whether the page is indexed at all. A surprising share of new content never gets a fair run because of a technical problem, a near-duplicate or a missing internal link. Confirm the page is in the index before concluding the keyword was wrong.

Watch for two specific failure modes. Cannibalisation looks like two of your pages swapping positions every week for the same query, and the fix is to merge them and redirect one. Stagnation looks like a page with steady impressions and a click-through rate under one percent; the content is probably fine and the title and meta description are the problem.

Finally, accept that some informational terms will never send meaningful traffic, because the answer now appears directly on the results page. Those pages still earn their place. They support the cluster, they attract links, and they feed the commercial pages. Just do not measure them with the same yardstick as a page written to sell something.