Key takeaways
- Keyword research is not about collecting single keywords. It is about finding topics people search for, judging whether you can compete, and grouping related queries onto one page.
- Start with a short list of seed terms, then expand them using the same places real searchers reveal intent: the live results page, related searches, and People Also Ask.
- Read difficulty from the SERP itself, not only from a tool’s difficulty score. Who ranks now tells you more than a number does.
- If your site already has traffic, your best keyword source is free: Google Search Console shows the queries where you already sit just off page one.
- Cluster closely related keywords into a single page. One strong page can rank for hundreds of variations, so you rarely need one page per phrase.
Most keyword research advice starts with a tool and ends with a spreadsheet of a thousand phrases nobody will ever turn into a page. That is backwards. Good keyword research is a decision process, not a collection exercise. You are trying to answer three questions in order: what are people actually searching for around my topic, can I realistically compete for it, and how should I group those searches into pages that earn traffic?
This guide walks through the workflow the way an experienced SEO runs it. It leans on the free data most people already have and ignore, and it treats paid tools as a way to add volume estimates, not as the starting point. Along the way it flags the judgment calls that separate research that ranks from research that sits in a tab forever.
Start with a seed, then expand
A seed is a short, obvious phrase at the center of your topic. If you sell running shoes, your seeds are terms like “running shoes,” “trail running shoes,” and “marathon training.” You do not need many. Five to ten good seeds per topic is plenty, because the goal of the seed is not to be your keyword, it is to be the doorway to the real ones.
Expanding a seed means finding the fuller, more specific ways people phrase the same need. The most reliable expansion sources are the ones Google shows you for free, because they reflect real queries rather than a model’s guess:
- Autocomplete. Type your seed into the search box and read the suggestions. These are drawn from real, popular queries.
- People Also Ask. The expandable questions on the results page show the adjacent questions searchers pair with your topic. They are a direct map of subtopics and a strong source of headings.
- Related searches. The block at the bottom of the results page shows sibling queries, often revealing modifiers and intents you would not have guessed.
This is also the reason you cannot rely on a keyword tool alone. Google has said for years that around 15 percent of the searches it sees each day are ones it has never seen before, a figure it has publicly reaffirmed more than once. Tools that build their databases from historical query logs cannot show you those. Reading the live results page keeps your research anchored to how people phrase things right now, including the longer, more conversational queries that are becoming more common as people search inside AI assistants. If you are planning for those surfaces too, it helps to understand how AI search differs from traditional Google search.
Only after you have a real list of phrases is it worth pulling volume estimates from a tool. Volume is useful for prioritizing, but treat the numbers as directional. They are modeled estimates, not meter readings, and two tools will happily disagree by a wide margin on the same term.
Read difficulty from the SERP, not the score
Every major tool gives keywords a difficulty score, usually zero to one hundred. It is a helpful first filter, but it is a model, and most of these models are built largely on the backlink profiles of the pages that currently rank. That leaves out a lot of what actually decides whether you can win a query. The faster read is to open the results page and look at it the way a competitor would.
When you look at a live SERP, ask:
- Who is ranking? If the top ten are all major brands and established authorities, a new page will struggle no matter what the difficulty score says. If you see forums, thin pages, or clearly dated content, there is an opening.
- What type of page ranks? The results tell you the intent Google has decided on. If every result is a product category page and you planned a blog post, you have misread the query. Match the format that already wins.
- Is the SERP mixed? When the results blend guides, product pages, and videos, Google is signaling it is unsure of intent. That is often an easier query to enter, because there is no single locked-in format to beat.
- How much of the page is Google’s own? Featured snippets, People Also Ask, and AI-generated summaries push the classic blue links down. If the real estate above the first organic result is crowded, the traffic on offer is smaller than the volume number suggests.
A difficulty score of 40 next to a page of weak, off-intent results is a better opportunity than a score of 20 in front of ten authoritative, perfectly matched pages. The score is a shortcut. The SERP is the evidence.
Mine Search Console for striking-distance keywords
If your site already gets some search traffic, the single best keyword source is not a paid tool. It is Google Search Console, and specifically its Performance report. Search Console reports the actual queries that showed your pages in search, along with clicks, impressions, and your average position for each one, all straight from Google (Search Console Help). No keyword tool can match that, because it is your real data rather than a sample.
The high-value move here is finding striking-distance keywords: queries where you already rank on the edge of page one, typically around positions 11 to 20, and need only a nudge to break through (Clearscope). These are the cheapest wins in SEO, because you have already proven Google considers your page relevant. You are optimizing an existing ranking rather than trying to earn a new one from scratch.
To find them in Search Console:
- Open the Performance report and select the Queries tab.
- Turn on the Average Position metric so it appears as a column.
- Filter position to greater than 10 and less than 21 to isolate the page-two band.
- Sort by impressions, high to low, so the queries with real demand rise to the top.
- Export the list. This export is your priority worklist.
Each row tells you a real query, roughly where you rank, and how much demand sits behind it (Search Console defines position as your topmost result for that query, averaged across the queries where you appeared, per Google’s Performance report documentation). Now cross-reference: which page is ranking for that query, and does the query represent something the page only mentions in passing? Often you will find a page ranking at position 14 for a term it barely addresses. Strengthening that section, or answering the exact question the query asks, is frequently enough to move it onto page one.
This is also the difference between research on a live site and research on a brand-new one. On a new site you are forecasting demand with external tools. On an established site, Search Console has already told you which battles you are almost winning.
Cluster keywords into one page, do not collect single ones
Common mistake. The most common beginner mistake is treating every keyword as a separate page.
“Best running shoes,” “best running shoes for beginners,” and “top running shoes 2026” are not three pages. They are one page, because they are the same search intent phrased three ways. Google understands this, and it will rank a single strong page across all the variations.
The scale of that is easy to underestimate. In an Ahrefs study of three million searches, the average page ranking number one also ranked in the top ten for nearly a thousand other related keywords (Ahrefs). You do not rank for a keyword. You rank for a topic, and a topic pulls in hundreds of related phrases automatically. That is why clustering matters: build one authoritative page per intent, and it harvests the long tail on its own.
To cluster your expanded list, group by intent rather than by wording. A practical test: open the results page for two keywords. If the ranking pages are largely the same, Google sees them as the same intent, and they belong on one page. If the results are mostly different, they are separate intents and need separate pages. This SERP-overlap check is more reliable than grouping by keywords that happen to share words.
Once clustered, each group becomes one target page with a primary keyword (usually the highest-volume phrase that captures the intent) and a set of secondary keywords you weave into headings and body copy. Those secondary terms are not filler. They are the People Also Ask questions and related searches you gathered earlier, which means your page answers the full shape of the topic rather than one narrow phrase.
Where each keyword source fits
No single source does the whole job. Here is how the main ones compare and when to lean on each.
| Source | What it gives you | Best for |
|---|---|---|
| Google Search Console | Real queries you already rank for, with position and impressions | Striking-distance wins on an established site |
| Autocomplete, People Also Ask, related searches | Real phrasing and adjacent subtopics, free and current | Expanding seeds and building headings |
| The live SERP | Who ranks, what format wins, how crowded the page is | Judging difficulty and confirming intent |
| Keyword tools (Ahrefs, Semrush, and others) | Volume estimates, difficulty scores, competitor keywords | Prioritizing and researching a brand-new site |
A simple workflow to put it together
- List five to ten seed terms for the topic.
- Expand each seed with autocomplete, People Also Ask, and related searches.
- If you have traffic, pull striking-distance queries from Search Console and add them to the list.
- Add volume estimates from a tool to help you prioritize.
- Open the SERP for your top candidates and judge difficulty and intent from what actually ranks.
- Cluster the survivors by intent using the SERP-overlap test.
- Assign one page per cluster, with a primary keyword and secondary terms mapped to headings.
That sequence keeps the work honest. It starts from real demand, filters with real evidence, and ends with a page plan rather than a keyword pile.
Frequently asked questions
Do I need a paid keyword tool to do keyword research?
No, though one helps. If your site already has traffic, Google Search Console gives you real queries, positions, and impressions for free, which is data no paid tool can replicate. Paid tools earn their place by adding volume estimates and competitor keywords, which matters most when you are researching a brand-new site with no data of its own.
What is a striking-distance keyword?
It is a query where you already rank just off the first page, usually around positions 11 to 20. Because Google already considers your page relevant, a small improvement can push it onto page one, where the clicks are. These are typically the fastest wins available, and you find them in the Search Console Performance report by filtering position to that band.
Should each keyword get its own page?
Usually not. Keywords that share the same search intent belong on one page. A single strong page routinely ranks for hundreds of related variations, so building a separate page per phrase splits your effort and competes against yourself. Cluster by intent, and check the overlap by comparing which pages rank for each keyword.
How do I judge if a keyword is too competitive?
Use the difficulty score as a rough first filter, then open the results page. If the top results are dominated by established authorities with pages that fully match the intent, it is hard regardless of the score. If you see weaker pages, forums, or off-intent results, there is room to compete. The live SERP is stronger evidence than any single number.
Written by Matthis Duarte, a senior SEO and organic growth expert with 10+ years of experience driving organic growth for international brands across highly competitive verticals. He is the founder of Knownful, an independent publication on SEO and organic growth featuring in-depth guides, best practices, playbooks and original analyses, including a free monthly study of which brands AI actually recommends across 10 industries.
