How to do keyword research

I sat down with a team selling accounting software, and their quarterly blog numbers looked healthy: sessions up, pageviews up. Then we opened the pipeline. Not one meeting had come from any of it. The reason was in their spreadsheet. Months earlier they had sorted every keyword by search volume, taken the ones at the top, and written articles for them. Most of those searches came from students doing homework. The list was accurate. What it measured was the problem.

Keyword research is not the act of producing a table sorted by volume. It is the act of mapping the questions in your customer's head, the words they use to ask them, and the answer they expect back. This piece covers where seed keywords come from, how to read search intent, how to balance volume against difficulty and upside, and how a raw list turns into a page plan you can hand to a writer on Monday.

Seed keywordsRaw long listFilter by intentDifficulty & upsideContent plan

The output is a page plan, not a word list

If research ends with five hundred rows in a spreadsheet, the work has not finished — it has not started. That table answers none of the questions that matter: which page gets written, who it speaks to, which queries belong together, and what ships first.

A useful output looks different. Every row is a page, carrying the main question that page answers, the secondary queries it covers, the format it should take (guide, comparison, product page), the funnel stage it serves, and a rough publishing order. When five hundred raw terms collapse into thirty rows like that, the research is done. That gap is the gap between raw data and a decision.

So judge a research session by how many page ideas got sharp, not by how many terms you collected. For the wider picture of how search works, our guide to SEO fundamentals sets the ground.

Where seed keywords come from

Seeds are the starting points you expand from, and they come out of your business, not out of a tool. Five sources cover nearly every case.

• Product and service names: What you sell, its category, and the synonyms your industry uses — your customer may not use your word for it, so write down both.
• The customer's own sentences: Call notes, proposal threads, and support tickets hold the exact phrasing people use for their problem, and "our stock counts keep breaking in Excel" is already a query.
• Objections your reps keep hearing: Repeated questions like "how do we move data off the old system" are unwritten article titles wearing a disguise.
• Support and FAQ traffic: The third time the same question hits your inbox, you have cheap proof it is being typed into search too.
• Competitor menus and blog archives: Their navigation labels reveal concept clusters you missed — you read them to find gaps, not to copy headlines.

Expanding the list with what search tells you for free

Once you have seeds, expansion costs almost nothing. Type a seed into the search box and write down the autocomplete suggestions; those are built from queries real people typed. Scroll to related searches and the "people also ask" block. Then run the seed again with modifiers around it: how, what is, pricing, alternatives, best, example, template. If your site is already live, the richest source is your own: Google Search Console shows which queries you appear for and which of those nobody clicks. Impressions without clicks are the fastest win on the board.

Search intent: same words, different expectation

Two people type the same phrase and want completely different things. Content written without reading intent gives the wrong answer to the right query. Four intent types cover the work.

Informational searchers want to learn: "what is…", "how to…". They deserve an explainer or a step-by-step walkthrough. Commercial searchers are narrowing options: "x vs y", "alternatives to…", "best…". They deserve an honest comparison split by use case. Transactional searchers are ready to move: "demo", "sign up", "near me", and deserve a product page, a form, a booking link. Navigational searchers are looking for a specific brand — yours or a competitor's.

The shortest way to read intent is to run the query and look at what ranks. Whatever kind of page Google promoted is what searchers expected. Forcing a product page into a results set full of guides is rowing against the current for months.

Volume tells you how many people asked. Intent tells you what they do after they get the answer. Only the second one shows up in your pipeline.

The volume, difficulty and upside triangle

Weigh every term on three axes. Volume is how many people search it, and alone it is the most misleading number in the file. Difficulty is the strength of the pages already ranking; the honest way to measure it is to open page one and ask whether you can genuinely write something better. Upside is the odds the person typing this becomes your customer.

For a new or small site the center of gravity sits on the upside axis. Ranking first for a quiet query your exact buyer types beats page three of a busy one by a margin that is not close. The first sends a few qualified conversations a month. The second sends nothing at all, forever.

Intent	Query shape	Page that fits	Action you expect
Informational	"what is…", "how to…"	Guide, glossary, walkthrough	Newsletter signup, resource download
Commercial	"x vs y", "alternatives to…"	Comparison table, buying guide	Demo request, trial start
Transactional	"demo", "quote", "near me"	Product page, form, booking link	Signup, proposal request
Navigational	"brand + feature", "brand + reviews"	Product page, help content	Purchase, support resolution

Long-tail and question-shaped queries

Long-tail means the longer, more specific queries — usually four words and up. Individually they look tiny; together they are large, and their intent is unmistakable. You cannot tell what someone searching "CRM" wants. Someone searching "CRM for a small team that pulls in WhatsApp messages" has told you exactly what to sell them.

Question-shaped queries are the most productive part of that tail. They are easy to write, because you already answer them on sales calls, and their share keeps growing alongside voice search and AI assistants. A heading that asks one question with a clean answer underneath reads well for humans and for the systems that quote you — a point we covered in why question-and-answer content gets cited.

Local searches and the language people actually type

People do not type the way brands write. They drop capitals and apostrophes, misspell product names consistently, and abbreviate anything long. The same concept gets searched under both its formal name and its shorthand — "customer relationship management" and "CRM" reach genuinely different audiences, and a list with only one of them is half a list.

Geography adds another layer. Queries carrying a city, a neighborhood, or "near me" behave differently; if you serve specific areas, local search visibility is a separate workstream. And if you sell in several languages, never build the list by translating it: start from seeds in each language and collect the questions that language actually asks.

Grouping one intent onto one page

Finding five near-identical terms and writing five articles is the most expensive mistake here. If "what is a CRM", "what does a CRM do", and "CRM meaning" ask the same thing, they are one page. Split across separate URLs, your own pages compete and none gets strong — that is cannibalization.

The practical test: run the queries and compare the results. If largely the same URLs rank across them, those queries belong together. If the results diverge clearly, they have earned separate pages. Grouping terms into clusters and hanging them off one authoritative page is covered in our topic clusters and pillar content guide; for many near-identical pages from one template, programmatic SEO is the pattern.

Tying keywords to the funnel and the calendar

Every term sits at a funnel stage. Awareness queries describe a symptom ("why do deals keep slipping"), consideration queries compare approaches ("spreadsheet vs CRM"), decision queries pick a vendor. A healthy plan feeds all three. Stack everything at the decision stage and traffic dries up; stack everything at awareness and revenue does.

None of it moves without a calendar. Keep which page ships which week, who writes it, and which cluster it belongs to in one place — the structure in our content calendar and publishing plan piece is enough. Building the page once the term is chosen is a separate craft, handled in how to write an SEO-friendly blog post.

How AI search changed keyword thinking

Someone asking an AI assistant does not type the way they type into a search box. Instead of "CRM pricing" they write: "can you suggest a system for a ten-person real estate office that also collects inquiries over WhatsApp?" That does not kill the keyword; it changes its unit. You now target the question the term lives inside, not the term alone.

Practically: keep full-sentence questions in the list beside short queries, use them as headings, and answer each one underneath in a passage that stands on its own when lifted out. The method for collecting what buyers ask assistants is in our prompt research guide — a sibling discipline to classic keyword work, not a replacement.

Prioritizing, measuring, and keeping the list alive

With thirty page ideas in hand you still cannot write them all at once. Score each row against four questions: Could this searcher become a customer? Can I write this better than what ranks now? Is page one plausible for my site? How close is this to a buying decision? Highest total goes first. Volume is the tiebreaker, not the sort column.

On measurement, watch behavior rather than rankings: which queries bring people in, what they do once they land, and how many ask for a conversation. That link only exists if the lead your content creates carries its source into the CRM. In Rocketly a web form submission lands straight in the pipeline with its origin attached, so you can answer "which keyword produced a proposal" instead of "which keyword produced traffic."

Treat the list as a living document. Every quarter, reopen Search Console, harvest the impressions that never earned clicks, reread the underperformers for intent mismatch, and add the phrases your sales team started hearing. Markets change vocabulary; your list has to change with them.

The mistakes that keep repeating

Five failures cover most of what goes wrong: sorting by volume and never asking about intent; opening separate pages for near-identical terms and competing with yourself; chasing queries far above your site's weight; building the list once and letting it rot in a folder; and targeting internal jargon instead of the words customers say out loud. All five share a root cause — treating research as a spreadsheet exercise.

Done properly, keyword research is the cheapest customer research a marketing team will ever run. It tells you what people ask, in their words, and where they get stuck. To see where those inquiries stall and measure which content actually creates revenue, create your Rocketly account and put your first form live this week.