How to Do Keyword Research and Decide Which Page to Write
See where search phrases come from, what the numbers really tell you, and how to decide which single page you can win, without paying for a tool.
/ On this page13 sections
Keyword research is how you work out which page to make next. You gather the phrases people type, look at what already ranks for each one, and commit to the handful you have a real chance at.
None of that needs a subscription. The phrases and the results are free, and the paid numbers are coarser than almost anyone tells you.
What Keyword Research Is
Keyword research is the work of finding the phrases people type into a search engine and choosing which of them you are going to answer.
What you should have at the end of it is small and specific: three or four phrases, a page committed to each, and a date against them.
Say you run a bike shop. "Bike helmets" is a topic and nothing follows from it.
"How to tell when a bike helmet is too old to use" is a commitment, because you can already see the page it becomes and the week you would write it.
Most of the difference between people who get something out of this and people who do not is what they think they are making.
A list of four hundred phrases feels like progress and changes nothing. Three phrases with a page attached to each is a week of work you can start on Monday.
The choosing matters because publishing costs you something every time.
An article takes a day to write and a year to be judged. Pick the wrong phrase and it does not cost you a spreadsheet row, it costs you the day and the year.
Why It Is Worth Doing
Search demand is worth working with because it is the one part of your marketing that already exists before you show up. People are typing something today, whether or not you have written anything about it.
Research tells you two things you cannot guess. The first is what they type, which is rarely what you call it internally.
The second is whether anyone typing it could plausibly end up on your site.
Skip it and you write from intuition.
Intuition is fine for the topics you know cold and terrible for the phrasing. You are the last person alive who talks about your own subject the way a stranger does.
Where the Candidate Phrases Come From
You need a pool of phrases before you can choose one, and four sources will fill it.
Work through them in this order. The first two cost nothing and take an afternoon; the second two need an account somewhere, and one of them needs a paid one.
Thirty or forty phrases is plenty. You are looking for the raw material of a decision, not a database.
Start From What You Sell, Not From a Tool
Write down what you do, in the plainest words available, one line each.
A tool needs a seed phrase before it can suggest anything, and the quality of everything downstream depends on that seed being real. Feed it your category and you get your category back.
Say you run a two-person bookkeeping practice. The honest seeds are bookkeeping, payroll, sales tax filing and the name of the software you set up for people, not "financial operations".
Add the questions clients ask you on the phone. Those are already phrased the way a stranger phrases them, which is the whole thing you are trying to find out.
The Search Box Itself
Type a seed phrase into a search engine and stop before you press enter.
The suggestions underneath are the completions other people chose, ordered by how often they chose them. Press enter and you get more: the questions in the People Also Ask panel, and the related searches at the foot of the page.
It is free, it is exact, and no vendor sits between you and it.
Work one level down each time. Click a People Also Ask entry, read the questions that appear underneath it, and take anything that sounds like something a customer would say.
What You Already Rank For
If your site has been up for a while, some of it already appears in search results for phrases you never chose.
Search Console shows them: every query that put one of your pages in front of somebody, and how often anyone clicked.
The interesting rows are the ones where you sit on the second page. The engine has already decided you are relevant and has not decided you are best.
Find the report that lists your queries and sort it by impressions. Menu routes change, so the name of the screen matters less than the column you sort on.
What Your Competitors Rank For
The last source is other people's phrases. Find two or three sites that sell roughly what you sell, and look at what brings them traffic.
This needs a tool, and it is the one job where a paid tool earns its money, because nobody gives away a competitor's whole keyword profile.
Take it as a source of ideas rather than a plan. What works for a site with ten years of links behind it is not automatically available to you.
What the Numbers Mean, and How Far to Trust Them
Three numbers come back beside every phrase, and each one answers a narrower question than its name suggests.
Learn what each is measuring before you sort by it. Sorting a column you have misread is how a whole quarter of writing ends up pointed at the wrong thing.
Search Volume
Search volume is roughly how many searches a phrase gets in a month. It is the size of the room, not the size of your share of it.
Four things about that figure are worth holding onto, and none of them is a reason to ignore it.
- It counts searches, not people. One person checking the same thing six times is six searches.
- It is an average across twelve months. Ahrefs makes the point with a phrase searched 120,000 times in December and never again: the reported figure is 10,000 a month, and every one of those months is wrong.
- It is rounded, and it is not only your phrase. Google Ads Help says its search volume statistics "are rounded", and that the average monthly searches figure covers a keyword "and its close variants". Which is why two phrases you think of as different often come back at exactly the same number: they were counted as one thing.
- Ranking first does not hand you the volume. Ahrefs puts the traffic from a single top-ranked keyword at rarely more than about 30% of it.
So the honest use of the number is comparative. It tells you that one phrase is much bigger than another, and nothing at all about what next month looks like.
Two phrases a few hundred searches apart are the same phrase for your purposes. Two an order of magnitude apart are not.
Which makes it a tiebreaker rather than a sort order.
Keyword Difficulty
Difficulty is a score out of a hundred estimating how hard a phrase would be to rank for.
Every tool publishes one and no two of them mean the same thing.
Ahrefs is unusually clear about what its own score counts: the number of separate websites linking to the pages currently in the top ten, and what you would need to join them there rather than to lead them.
That distinction does a lot of work.
Reaching position nine and reaching position one are different projects, and the score is quoting you for the cheaper one.
Other tools weigh other things, so the same score in two products is two different claims about the world. Advice to keep under some threshold means nothing until somebody says whose scale it sits on.
Use it inside one tool, as a ranking of your own candidates. Do not carry the number between tools, and do not treat it as a verdict on whether you can win, which is a question the results page answers better.
Cost Per Click
Cost per click is what advertisers pay to appear above the results for a phrase.
It sits in your keyword report because the underlying data was built for advertisers, not for you.
For organic work it is one thing only: a read on how commercial the searcher is. Nobody bids heavily on a phrase typed by people with no intention of buying anything.
Treat a high figure as a hint that the searcher may be close to buying, and let the results page settle what kind of page to write.
Then leave it alone. What to bid is a different job with a different budget.
Use this chart — embed code and citation
<a href="https://neerajjivnani.com/blog/keyword-research/"><img src="https://neerajjivnani.com/infographics/keyword-research/one-month-reported-as-twelve.png" alt="A chart of one year of searches for a single phrase, drawn to show what a reported search volume is averaging. One tall orange bar stands at December, labeled 120,000 searches, and the other eleven months are flat hairlines labeled as never again. A dashed line runs across the whole chart close to the baseline, at the level a keyword tool reports for the phrase, 10,000 a month, which is the twelve months divided by twelve. The vertical scale is searches in that month, from zero to 120,000. Three notes underneath carry the rest of what the figure is measuring: that it counts searches and not people, since one person checking the same thing six times is six searches; that Google Ads Help says its search volume statistics are rounded and that the average monthly searches figure covers a keyword and its close variants, which is why two phrases you think of as different often come back at exactly the same number; and, highlighted, that Ahrefs puts the traffic from a single top-ranked keyword at rarely more than about 30% of it. A band at the foot says the honest use of the number is comparative, that it tells you one phrase is much bigger than another and nothing at all about what next month looks like, and that this makes it a tiebreaker rather than a sort order." width="1200"></a>
<p>Chart: <a href="https://neerajjivnani.com/blog/keyword-research/">Neeraj Jivnani</a></p>Neeraj Jivnani, "How to Do Keyword Research and Decide Which Page to Write", neerajjivnani.com, https://neerajjivnani.com/blog/keyword-research/Free to republish with a link back to this page.
Reading the Results Page
Type the phrase in and read the results page. That is the most useful minute in the whole process, and almost everyone leaves it until last.
What is on that page is not an estimate of anything. It is the searcher's actual experience of the phrase, and it is the thing every number above was approximating.
Two answers are sitting there for free.
What the Results Tell You About Intent
Intent is what the person wanted. You can classify it without a tool by looking at what kind of page the engine chose to reward.
The four kinds are conventional and they are useful.
- Informational means somebody wants to understand something.
- Navigational means they want a particular site.
- Commercial means they are comparing before buying.
- Transactional means they are ready to buy.
Reading them off the results is quicker than reasoning about them.
- Ten guides and how-to articles means informational. Publish a product page here and it will not rank, however good it is.
- Product and category pages means transactional or commercial. Publish a blog post here and the same thing happens in reverse.
- A mixture means the engine is not sure either, which is usually an opening.
One mistake is worth naming: treating intent as a property of the words.
"Best running shoes" reads commercial and is answered almost entirely by editorial reviews. The words suggest; the results decide.
How Much of the Click Is Left
Volume counts searches. What you care about is the clicks still available once the results page has taken its share.
Look at what sits above the first ordinary result. An AI answer, an ad block, a video carousel, a set of images, a panel of questions: each of them is a place the searcher can be satisfied without going anywhere.
A busy phrase whose answer is given in full above the results can be worth less than a quiet one that nothing else answers.
You cannot measure this precisely and you do not need to.
Look at the page, ask how far down the first ordinary result sits, and let the answer move a phrase up or down your list rather than settle it outright.
Head Terms and Long-Tail
Phrases divide roughly into two kinds, and the split matters because it tells you what page you would have to write.
A head term is short and broad: two words, enormous demand, no clue about who is typing it.
A long-tail phrase is longer and more specific. Far fewer people type it, and it tells you almost everything about the one who did.
Say a shop sells coffee equipment. The specificity is the whole value, and it works in two directions.
It tells you what to write, because "coffee grinder" could become fifty different pages and "burr grinder for a small espresso machine" could only ever become one. And it tells you who arrives, because nobody types the second phrase idly.
The arithmetic is the other half. Twenty specific pages, each drawing a little, routinely add up to more than one broad page you never manage to rank, and they arrive sooner.
None of this makes head terms a mistake. It makes them a later project.
They are worth writing once the specific pages around them are already working, because by then you have the links and the coverage that ranking for one requires.
If your site is new, the specific phrases are not a consolation prize. They are the only door that is open.
Choosing the One You Write Next
The one you write next is the phrase you could beat, for a reader you want, on a results page with clicks still left on it. Everything above was preparation for that call.
Take your candidates and ask three questions about each, in this order. The first no ends it, and you move to the next phrase.
- Is the person typing this somebody you want? Not "is it related to my business", which almost everything is. Would you be pleased if they landed on your site?
- Could you publish something better than what is at the top, this month? Open the first three results and read them. Not skim, read.
- Is there a click left? You read this off the page in the previous section. If there is not much, the phrase drops down the list.
Whatever survives all three, sort by volume and write the biggest one.
Why the Order Is the Whole Point
Volume comes last on purpose. Sort by it first and you get a ranked list of things you cannot win, in confident descending order.
The binding constraint is not demand. It is whether you can beat the page already there.
And that is a question about your afternoon, not about a score.
Sometimes the top result is thin and years old, and the answer is plainly yes. Sometimes it is an excellent piece from somebody with ten years of links behind them, and this one is not yours yet.
Saying that out loud is most of the skill.
Use this chart — embed code and citation
<a href="https://neerajjivnani.com/blog/keyword-research/"><img src="https://neerajjivnani.com/infographics/keyword-research/volume-is-the-last-question.png" alt="A four-step sequence for choosing which search phrase to write about next, with volume placed last. Step one asks whether the person typing the phrase is somebody you want, and notes that this is not the same as whether the phrase is related to your business, which almost everything is. Step two asks whether you could publish something better than what is at the top this month, and says to open the first three results and read them rather than skim them, because this is the binding constraint and it is a question about your afternoon rather than about a score. Step three asks whether there is a click left, since a phrase whose answer is fully given above the first ordinary result is worth less than its volume says. Each of the three carries a marker reading that a no ends it. Only after all three does the fourth panel, highlighted in orange, say to sort whatever survives by volume and write the biggest one, because volume is the size of the room rather than the size of your share of it. A band at the foot gives the other way round: sort by volume first and you get a ranked list of things you cannot win, in confident descending order." width="1200"></a>
<p>Chart: <a href="https://neerajjivnani.com/blog/keyword-research/">Neeraj Jivnani</a></p>Neeraj Jivnani, "How to Do Keyword Research and Decide Which Page to Write", neerajjivnani.com, https://neerajjivnani.com/blog/keyword-research/Free to republish with a link back to this page.
Grouping the Rest, and Giving Each One a Page
Grouping the rest is the tidy-up, and the rule is one page per group. You will finish with more phrases than pages, which is the normal and correct outcome.
The work splits three ways: deciding which phrases are the same question, giving each group exactly one page, and writing the whole thing down somewhere you will find it again.
None of it takes long. It is the part that stops you doing this twice.
Telling Two Phrases Apart
Most of your leftovers are not separate topics. They are the same question asked in different words, and the engine already knows that.
Here is how to tell.
Search two phrases you suspect are the same and compare the results. If eight of the ten pages are the same, in roughly the same order, one page can serve both.
Group them that way, by what the results say rather than by which words they share.
So "burr grinder" and "what is a burr grinder" are one page, and "burr grinder vs blade grinder" is a second, because the second returns comparisons rather than definitions.
Then pick the phrasing you will write toward, and treat the rest as things the page should cover.
One Phrase, One Page
The rule that follows is worth being strict about, and almost nobody breaks it on purpose.
Point two pages at the same phrase and you are not doubling your chances, you are splitting them.
The engine picks one of yours to show. The clicks and the links divide between them, and both do worse than either would have alone.
This is the failure people call cannibalization, and it is self-inflicted rather than competitive.
Keeping the Record
Keep it somewhere dull. A sheet with four columns will do: the phrase, the page it belongs to, what the searcher seemed to want, and the date you last looked.
The value of the sheet is not the research. It is knowing, eighteen months from now, that you already covered something.
Doing It Free
Yes, and doing keyword research free is not a compromised version of doing it. Four of the five things you need cost nothing, and the fifth is a view of what your competitors rank for.
What you cannot get free is precision, and precision is the part you needed least.
The split runs roughly like this.
What Costs Nothing and Is Exact
- The search box. Autocomplete, People Also Ask, related searches. Real phrases from real searches, with no vendor in between.
- The results themselves. Intent, the standard you have to beat, and how crowded the page is. This is the input the whole decision turns on.
- Search Console, if the site is yours. Every query that showed one of your pages, and whether anyone clicked.
- Google Trends, for whether interest is rising or falling, which is the one thing an annual average cannot tell you.
What the Free Tiers Hold Back
Google's own Keyword Planner is free and built for advertisers, which shows in what it returns: rounded figures, close variants counted together, and a strong pull toward what a click would cost rather than whether you could earn it.
There is one condition worth knowing before you plan an afternoon around it. Google Ads Help says you must complete your account setup by entering your billing information before you can reach basic features like getting keyword ideas.
The independent tools mostly meter the free tier by the day, and the allowances are small.
Zapier's 2026 test found three searches a day at AnswerThePublic, five at KWFinder and ten analytics reports a day at Semrush, with a fourth tool metered at ten lookups a month.
Enough to check a shortlist, nowhere near enough to explore.
That is the real trade, and it is smaller than the marketing around it.
Paying buys you speed and a competitor's keyword list. It does not buy you the decision, and the decision is the job.
If you are choosing where to spend, spend on the competitor data. That is the one thing the free layer genuinely cannot give you.
Use this chart — embed code and citation
<a href="https://neerajjivnani.com/blog/keyword-research/"><img src="https://neerajjivnani.com/infographics/keyword-research/four-of-the-five-cost-nothing.png" alt="Two columns splitting keyword research into what costs nothing and where the free layer stops. The highlighted left column lists four things that are free and exact: the search box, meaning autocomplete, People Also Ask and related searches, which are real phrases from real searches with no vendor in between; the results themselves, which give intent, the standard you have to beat and how crowded the page is, and which are the input the whole decision turns on; Search Console if the site is yours, for every query that showed one of your pages and whether anyone clicked; and Google Trends, for whether interest is rising or falling, which is the one thing an annual average cannot tell you. The right column lists three limits: Keyword Planner is built for advertisers, and Google Ads Help says its search volume statistics are rounded and that the average monthly searches figure covers a keyword and its close variants; the independent tools mostly meter the free tier by the day, and Zapier's 2026 test found three searches a day at AnswerThePublic, five at KWFinder and ten analytics reports a day at Semrush, with a fourth tool metered at ten lookups a month; and a competitor's keyword list is the one thing the free layer genuinely cannot give you, because nobody gives away a competitor's whole keyword profile. A band at the foot states the real trade: paying buys you speed and a competitor's keyword list, it does not buy you the decision, and the decision is the job." width="1200"></a>
<p>Chart: <a href="https://neerajjivnani.com/blog/keyword-research/">Neeraj Jivnani</a></p>Neeraj Jivnani, "How to Do Keyword Research and Decide Which Page to Write", neerajjivnani.com, https://neerajjivnani.com/blog/keyword-research/Free to republish with a link back to this page.
Keyword Research and AI Search
Keep the method and change two things inside it. AI answers move the arithmetic, not the process.
What shifted is the share of a search that ends in a click, and the shape of the phrases people type.
What Happens to the Clicks
An AI answer now sits above the results for a large share of informational phrases, and a question fully answered up there does not need a click. The gap between a phrase's volume and the traffic it can send you has widened.
It has widened most on the simple explanatory questions that used to be the easiest wins. "What is a burr grinder" is answered before you get anywhere near a result.
That is a reason to weigh the third of the three questions more heavily, not a reason to stop writing explanations.
What Happens to the Phrasing
People type differently into a chat box than into a search bar. They write a whole sentence, include their situation, and follow up rather than starting again.
So "grinder" becomes "is a burr grinder worth it for one cup a day". That pushes in the same direction the long tail always did: specific phrases, plainly answered, with the answer near the top of the page where it can be quoted.
We would resist one thing: treating this as a new discipline with its own research process.
The phrases you find are the same phrases. What moved is how much of the click survives the page, and you were already supposed to be checking that.
What Goes Wrong
Four failures account for most wasted keyword research, and none of them is subtle once you know to look.
Sorting by volume and starting at the top. The biggest phrases are the ones everybody wants, so the top result is usually somebody who has been working on it for a decade. You lose a year finding out.
Writing the page the words suggest instead of the page the results show. A phrase that reads commercial and returns ten guides wants a guide. Getting this wrong is not a ranking penalty; it is a page that never appears at all.
Treating the numbers as measurements. They are rounded, averaged and vendor-specific, so a decision that turns on a small gap between two of them is a decision built on nothing.
Never revisiting the sheet. Research done once and filed is research you will redo, badly, in a year, because nobody remembers what was already covered.
There is a fifth that is rarer and easier to fix: stuffing a chosen phrase into a page repeatedly, on the theory that frequency helps.
It does not work, and the page reads like it is trying to. Nobody has ever finished one of those.
When to Go Back to the Sheet
Three things send you back to the sheet, and each of them is an event you will notice rather than a date you have to remember.
- A page you own slips. Search it, read what overtook you, and decide whether to improve the page or leave it.
- The results for a phrase you own change shape. New features above the fold, or a different kind of page winning, means the intent moved and your page may no longer match it.
- You start selling something new. New offer, new seeds, new pool, and it takes an afternoon.
Search Console is the one thing worth a calendar entry, because the phrases you rank for by accident never announce themselves. Those are the cheapest pages you will ever write.
And one rule is genuinely fixed: do it before you write, never after. Research done on a finished page is a search for justification.
Choosing, Not Collecting
Done properly this is choosing, not collecting. It is a series of small refusals: not that one, not yet, not until the specific pages work, and the data is the least of it.
The phrases are free. The results page is free and exact.
What the paid layer sells you is speed, and a view of your competitors. Neither one makes the choice for you.
So the whole method fits in your head. Gather the phrases people type, open the ten results for the ones you like, and write the one where you could do better than what is there, for a reader you want.
Then write it, and pick the next one.