Most keyword research fails at the first step, and it fails the same way every time.
Someone opens a tool, sorts by search volume, picks something with a big number and a difficulty score that looks manageable, and writes a post. Six months later it's on page four, and the conclusion is usually "SEO doesn't work for us."
What actually went wrong: they picked a keyword before understanding what people wanted when they searched it, and before checking whether a site like theirs could plausibly win it.
Both of those are free to check. Neither requires a subscription. Here's the process.
The short version
Work in this order:
- Intent first — what does the searcher actually want?
- SERP second — who's ranking, and can a site like mine displace them?
- Demand third — is anyone searching this at all?
- Fit last — does this connect to something I sell or want to be known for?
Volume comes third, not first. That single reordering fixes most keyword research.
And the free stack is genuinely sufficient for most sites: Google itself, Search Console, autocomplete, People Also Ask, Reddit, and your own customer conversations.
Why volume-first fails
Search volume tells you how many people search a term. It tells you nothing about:
- Whether those people want what you have
- Whether you can realistically rank
- Whether ranking would produce anything of value
A term with 20,000 monthly searches dominated by Wikipedia, Amazon, and three sites with a decade of links is worth less to you than a term with 40 searches where the current top result is a four-year-old forum thread.
There's also a specific trap for young sites. High-volume head terms are high-volume because they're broad — and broad queries pull in people at every stage of intent, most of whom aren't looking for you. "Email marketing" gets searched constantly by students, job hunters, and the idly curious. "Why do my emails go to spam after a domain migration" gets searched by someone with a problem you can solve today.
Intent: the four types
Before anything else, work out what the searcher wants. The same words can carry completely different intent, and mismatching it makes everything downstream irrelevant.
Informational — "what is DMARC," "why do emails bounce." Wants an explanation. Ranks: guides, explainers, definitions.
Commercial — "best email verification tools," "Klaviyo vs Mailchimp." Wants to compare before buying. Ranks: comparisons, reviews, roundups.
Transactional — "Klaviyo pricing," "buy SPF checker." Wants to act. Ranks: product and pricing pages.
Navigational — "Klaviyo login." Wants a specific destination. You can't win these unless it's your brand.
How to check intent: search the term in an incognito window and look at what Google is showing. That's Google reporting the aggregate behaviour of everyone who's ever searched it — a far better signal than your assumption.
If the whole first page is comparisons and you planned a definitional explainer, you've learned something valuable before writing a word.
Read the SERP before you commit
Intent tells you what to write. The SERP tells you whether it's worth writing at all.
Open the query and look for:
Who's ranking. Wikipedia, Amazon, and major publishers across all ten results means walk away — those aren't beatable positions for a small site. A mix of smaller sites, forums, and older content means there's room.
What format wins. If eight of ten are step-by-step guides, that's the expected shape.
How good the content actually is. Click through to two or three. Genuinely comprehensive, current, and useful? Hard. Thin, outdated, or clearly written to a word count? That's your opening.
How much of the page you can even reach. If an AI Overview, a featured snippet, a video carousel, and four ads sit above the organic results, position one is a long way down the screen. Some queries are no longer worth winning. Reading a SERP properly.
Six free sources for ideas
1 — Google autocomplete. Type your topic and watch the suggestions. These are real queries, ordered roughly by popularity. Add a letter to fork the list: "email deliverability a," "email deliverability b."
2 — People Also Ask. Real questions, and it expands as you click — one query can yield twenty related questions. These map directly to FAQ sections and supporting posts.
3 — Related searches at the bottom of the results page. Adjacent concepts Google associates with your topic.
4 — Search Console. The best source you have, and it's specific to you. Queries you already get impressions for are validated demand with confirmed relevance. How to mine it.
5 — Reddit and forums. How people describe problems in their own words, before they've learned the jargon. "My emails are going to junk" rather than "inbox placement optimisation." That vocabulary gap is where uncompeted long-tail queries live.
6 — Your own inbox and calls. The questions clients actually ask are the highest-value keywords you'll ever find, because they come from people who pay you. Keep a running note.
Notice that none of these cost anything, and the last two produce better queries than any tool because they're rooted in how real people talk.
Keyword difficulty scores and what they miss
Every tool offers a difficulty score. They're useful as a rough filter and misleading as a decision.
What they measure: mostly the link profiles of currently-ranking pages, sometimes with domain authority mixed in.
What they miss:
Content quality. A page with strong links and terrible content is more beatable than the score suggests.
Intent match. If the ranking pages don't quite answer the query, a page that does can outrank stronger domains.
Freshness. Four-year-old top results on a topic that's changed are vulnerable regardless of their links.
Topical authority. A site that's covered a subject thoroughly can outrank a stronger generalist domain on that subject specifically.
SERP crowding. Nothing in the score accounts for an AI Overview taking the top third of the screen.
The practical position: use difficulty scores to filter out the obviously hopeless, then judge the shortlist by reading the actual results. Ten minutes of looking beats any number.
Long-tail: where a young site starts
The counterintuitive move that works: on a new site, deliberately target queries with small search volume.
"Email deliverability" is unwinnable for a new domain. "Why did my emails go to spam after changing DNS providers" might get forty searches a month and be entirely winnable this quarter.
Three reasons this is a better strategy than it sounds:
You can actually rank, which means traffic now rather than hypothetically.
Intent is sharper. Someone searching a fifteen-word question has a specific problem. They convert far better than someone searching two words.
Coverage compounds. Twenty specific pages on one subject build the topical authority that eventually lets you compete for the head term. That's the actual route to "email deliverability" — not attacking it directly, but earning it. The full case.
Is search volume even accurate?
Worth being honest: not very.
Volume figures are estimates, derived from clickstream data and Google Ads data that was never designed for this purpose. Different tools give different numbers for the same term, sometimes by multiples. They're rounded into buckets, they lag, and they miss seasonality.
Use volume as a rough order of magnitude, not a precise figure. The useful question isn't "does this get 320 searches" — it's "is this in the tens, hundreds, or thousands." That distinction is reliable. The specific number isn't.
Also worth knowing: a large share of Google searches are unique queries never seen before. A tool reporting zero volume for a long-tail question doesn't mean nobody searches it — it means the tool doesn't have data. Some of my best-performing pages target queries every tool called zero.
Mapping keywords to pages
One page, one primary query, plus its close variants.
The variants belong together. "How to set up DMARC," "DMARC setup guide," and "how do I configure DMARC" are the same question. One page serves all three, and trying to split them produces three thin pages competing with each other.
Genuinely different questions get different pages. "What is DMARC" and "DMARC not working" are different intents and need separate pages.
The test: would the ideal answer to these two queries be the same page? If yes, one page. If no, two.
Get this wrong and you get cannibalisation — two of your pages competing for one query, splitting your signals, and both underperforming. How to spot and fix it.
Building a publishing queue
Research produces a list. A list isn't a plan.
Score each candidate on three things, roughly:
Winnability — from your SERP read. Can you realistically rank within a few months?
Business value — does this connect to what you sell? A high-traffic post about nothing you offer is a vanity metric with hosting costs.
Cluster fit — does it support a pillar you're building, or is it an orphan topic?
Then sort by winnable-and-valuable first. Publish those. Momentum matters more than optimality — three pages ranking beats a perfect spreadsheet.
One structural point: build clusters rather than scattered posts. A pillar plus four or five supporting pages on one subject outperforms ten unrelated posts, because the cluster demonstrates topical command and the internal links compound. The hub-and-spoke model.
Frequently asked questions
How do I find low-competition keywords?
Look for queries where the current results are weak rather than where a tool reports a low difficulty score. Specific long-tail questions, problem-phrased queries using everyday language rather than industry jargon, and topics where the top results are several years old on a subject that's changed. Reddit and forums are unusually good sources because people describe problems there before they've learned the professional vocabulary.
Is search volume accurate?
Not precisely. Volume figures are estimates derived from clickstream and ads data, they vary substantially between tools for the same term, they're bucketed and rounded, and they lag. Treat them as orders of magnitude — tens, hundreds, thousands — rather than exact counts. And zero reported volume often means the tool has no data rather than that nobody searches it.
How many keywords should one page target?
One primary query plus its close variants — different phrasings of the same question. Genuinely different questions need different pages. The test is whether the ideal answer to both queries would be the same page. Splitting variants across multiple pages causes cannibalisation, where your own pages compete and both underperform.
Are free keyword tools good enough?
For most sites, yes. Google's own autocomplete, People Also Ask, related searches, and Search Console cover the majority of what you need, and Search Console gives you something no paid tool can — validated demand specific to your site. Paid tools mainly buy you speed at scale and competitor analysis. If you're publishing a few posts a month, the free stack is genuinely sufficient.
What to do next
Open Search Console, go to Performance → Queries, and sort by impressions.
Look for queries you're getting impressions for but never wrote about. Those are the cheapest opportunities you have — validated demand, confirmed relevance, and no guessing about volume, because Google is already showing you to those people.
That list will usually beat anything a keyword tool gives you this week.
Free: The SEO audit checklist.
Related guides
- How to read a SERP — judging winnability before you write
- Long-tail keywords — where new sites should start
- Find keywords you already half-rank for — the cheapest wins
- Keyword cannibalization — when your own pages compete
- On-page SEO — what to do once you've chosen
- SEO when AI answers the question first — which queries are still worth winning
Join the Newsletter
Get practical marketing tactics delivered straight to your inbox.
Written by
Muhammad Basim
Related Articles
Keyword Cannibalization: How to Spot It and Fix It
You wrote a good post about email authentication in 2023. Then a better one in 2024, because you'd learned more. Then a third in 2025 covering DMARC specifically, which overlapped with both. Now none of them rank well, and Google keeps swapping which one it shows. That's cannibalisation. Your own pages competing for the same […]
Mining Search Console for Keywords You Already Half-Rank For
There's a list sitting in your Search Console right now of keywords Google has already decided you're relevant for, where you're ranking just badly enough that nobody sees you. Nobody looks at it. Everybody buys a keyword tool instead. Those queries — roughly positions 8 through 20 — are the cheapest wins available on any […]
Long-Tail Keywords: Where a New Site Should Actually Start
Every new site owner wants to rank for the two-word term. "Email marketing." "Project management." "Running shoes." And every new site owner is competing there against domains with fifteen years of links, full-time content teams, and a brand people already search by name. You will not win that this year. You may not win it […]