Keyword research for content: demand before drafts

By Matija Konjić

Keyword research is where most content programs are decided, and most teams do it backwards: export a thousand rows, sort by volume, and write toward the biggest numbers they recognize. The result is a blog optimized for a spreadsheet, chasing queries the business cannot win or would not want, while the terms that actually buy things sit unclaimed three tabs over.

The backwards version also explains a familiar postmortem: a year of publishing, decent writing, real effort, and rankings on nothing that matters. The autopsy almost always finds the same cause of death, targets chosen by volume rather than by buyer, winnability and value, which means the failure was decided before the first draft and no amount of writing was going to undo it.

That is the stake this guide plays for: research is the only stage where a content program can still choose its battles, and every stage after merely fights them. Choose well and the writing gets easier, the links get cheaper, and the reporting starts agreeing with the plan, which is what a program built on demand instead of guesswork feels like from the inside, month after month.

Done forward, keyword research is a filtering discipline: from everything people type, to what your buyers type, to what you can win, to what pays. This guide is that discipline end to end, the way it feeds every roadmap in our content strategy work, with the judgment calls the tools cannot make for you spelled out.

The three filters

Every export shrinks through the same three questions, in order.

Buyer reality first: does your actual customer type this, in this language, at any point on the way to buying. Winnability second: given the domain’s current authority, is this term reachable inside a year, which is an honest reading of who holds the positions now. Value last: would ranking here produce customers rather than visitors. A term needs all three yeses, and the discipline is letting big volumes die at any filter without ceremony.

The order of the filters is itself the safeguard. Winnability judged before buyer reality produces a list of easy irrelevancies; value judged first produces a wishlist of unwinnable money terms. Buyers, then winnable, then paying, and the sequence keeps every later argument short, because a term that failed an earlier filter never reaches the meeting.

Where the seed list comes from

Tools expand lists; they cannot start them well. The seed list comes from the business: what sales gets asked, what support keeps explaining, what the last ten customers said they searched before finding you, and what problem each service actually solves, phrased the way customers phrase it rather than the way the industry does. An hour of internal listening produces seeds no competitor export contains, and the expansion from real seeds stays anchored to money.

Competitor exports join the pool as evidence rather than gospel. The terms rivals rank for reveal the market’s proven demand, and the terms they all ignore reveal either graveyards or gaps, distinguishable by whether anyone types them. Stealing a competitor’s keyword list wholesale imports their strategy’s mistakes along with its wins, and their capacity assumptions along with both, which is why the exports feed the filters instead of bypassing them.

Intent before volume, every time

Volume tells you how many people type a phrase; intent tells you what they want when they do, and intent decides everything downstream.

The classification is read from the results page rather than guessed: whatever page types hold the positions are the market’s verdict on what the query wants. A term whose results are all product pages will not rank a blog post, however good, and a term whose results are guides will resist a service page. Matching page type to intent is the difference between competing and colliding.

Mixed-intent terms, where the results page splits between guides and product pages, deserve a special flag: they usually mark a market mid-shift, and the winning move is often both pages, built in the same cluster, each serving its half of the split, with the internal link between them doing the routing the searcher could not. The tools will report one keyword; the market is holding two doors open.

Seasonal and emerging terms

Two term species need their own handling. Seasonal queries get judged on their season’s numbers and scheduled with runway, a November term ships in September. Emerging queries, the ones tools report as zero because the tools lag reality, get judged by the seed sources instead: when sales hears a phrase weekly and the export says nobody types it, trust sales over the export, because being early to a real term is the cheapest ranking anyone ever wins.

The volume trap, quantified

High-volume heads are crowded, generic and often informational; long-tail phrases are specific, numerous and closer to money. A hundred visits from software for accountants pricing beat a thousand from what is accounting, because the first query is a buyer and the second is homework, and homework rarely buys anything on the day. Programs built on long-tail clusters routinely out-earn programs chasing heads, and they get to winnable a year sooner.

The arithmetic deserves one worked line. A head term at five thousand monthly searches, position eight, sends fewer visits than forty long-tail terms at eighty searches each held at position two, and the long-tail visitors arrive knowing what they want. Volume is a ceiling, position is the multiplier, and specificity is what buys position on a young domain.

Reading competition honestly

Difficulty scores are a starting estimate, and the real read is opening the results and looking. Who holds the top ten: recognizable authorities with deep pages, or thin content and forum threads. What would it take to beat the weakest page on page one: better depth, fresher data, cleaner intent match, or authority your domain does not hold yet. Ten minutes of reading answers what no score can, and it also hands you the angle, because the gaps in the ranking pages are the brief.

Signals hiding in the results page

While reading, collect the free intelligence: the People Also Ask questions are subtopics buyers phrase themselves, the ads reveal which terms carry money, and a results page full of aged content is a market waiting for one current answer. The same ten minutes that grades difficulty hands over half the outline, which is why the reading pass beats any export-only workflow on both speed and quality.

Note what the assistants answer for the term as well, because AI summaries now sit above many of these results. A query the assistants answer thinly is a citation gap on top of a ranking gap, worth double to whoever writes the page both surfaces prefer, and invisible to teams still sorting by volume.

The quadrant sorts every surviving term into a move: claim the high-demand weak-competition gaps first, collect the easy long-tail wins alongside, build deliberately toward the strong-rival terms as authority grows, and let the low-value corners go without guilt. That ordering is the roadmap’s skeleton, and it is also a budget: terms in the build-toward quadrant are where link investment gets scheduled, because content alone will not close those gaps.

Revisit the quadrant each quarter, because terms migrate. Authority gained moves build-toward terms into reach, competitors shipping moves gaps into contests, and the map that steered last quarter mis-steers this one exactly in proportion to how much the program achieved. Winning re-prices the next targets, which is the pleasant version of the problem.

From list to clusters

Surviving keywords group into topics, and topics are what actually get planned. Twenty related phrasings are one page with sections, or one cluster with a hub and supporters, and the grouping decision is the same intent-reading exercise at smaller scale: phrasings whose results pages look alike share a page; phrasings whose results diverge deserve their own. Tools cluster by word overlap, which is wrong exactly often enough to require the human pass.

Each cluster then gets its priority from the quadrant, its page types from intent, and its slot on the roadmap from capacity, which is the moment keyword research stops being research and becomes the plan.

Volume estimates get one final demotion on the way in: they are directionally useful and individually unreliable, tool figures for the same term routinely disagree by half, so the plan ranks terms against each other rather than trusting any absolute count. Relative demand survives bad estimates; absolute promises built on them do not, and forecasts pinned to tool volumes are how research loses rooms it should have owned.

Research as maintenance

The market keeps typing after the plan ships. Quarterly, the same pass runs lighter: Search Console shows the queries pages already attract, including the ones nobody planned for, and those accidental rankings are the cheapest expansion signal there is, demand the site already half-owns and can usually claim with one section added to an existing page. New seeds from sales and support join the pool, dead terms leave, and the roadmap absorbs the changes at the same quarterly review everything else obeys.

The one metric worth watching on the research itself: share of shipped pieces that reached the top twenty within two quarters. It measures whether the winnability filter is calibrated, and when it drifts low the filter needs tightening more than the writers need blaming, which is a diagnosis most programs never think to run.

Run the whole discipline once properly and it stays cheap forever: the first pass takes days, the quarterly refresh takes an afternoon, and the roadmap it feeds decides thousands of hours of production. Few afternoons in marketing carry that leverage, which is exactly why this one should never be delegated to a sort button.

And when the plan is built, close the loop with the writers: every brief carries its term, its intent and its quadrant, so the person drafting knows what the piece must beat and why it was chosen. Research that never reaches the brief was tourism, and the brief is where the afternoon of filtering finally turns into pages that pay. Revisit the research quarterly rather than annually; search language drifts faster than most calendars assume, and the drift is where early positions get won.