Finding the terms your audience uses — volume, difficulty, intent — and turning them into a content plan.
Keyword research is how you find out what your audience actually searches for, in their words rather than yours — and how you turn that into a content plan. It rests on three measures: how much demand exists, how hard the query is to win, and what the searcher is actually trying to do. Get the third one wrong and the other two do not matter, because a page that matches the words but not the intent satisfies nobody.
Keyword research exists to answer a practical question: what should we publish, and for whom? It surfaces the terms and questions real people use when looking for what you offer, quantifies roughly how many of them there are, estimates how difficult each would be to compete for, and reveals what the searcher actually wants. The output is not a list of words but a prioritised content plan grounded in evidence about real demand.
The discipline matters because organisations naturally describe things in their own vocabulary — internal product names, industry jargon, aspirational framing — which frequently differs from how customers describe the same thing. Keyword research corrects that gap by grounding your content in observed language rather than assumed language. Understanding what the research is for keeps it from becoming an end in itself: the point is a plan you can act on, not a spreadsheet of terms.
Search volume estimates how many searches a term receives in a period, and it is the most visible metric in every tool — which is why it is the most commonly misused. Volume figures are estimates, often aggregated across variations, and they say nothing about whether the searcher is a plausible customer. A high-volume term attracting people with no interest in buying is worth less than a low-volume term used by people ready to act.
The practical correction is to read volume as one input among several rather than as a ranking of what matters. Match real demand and intent, not just search volume: ask whether the people behind a query are the people you want, whether the term reflects a need you genuinely serve, and whether winning it would produce anything of value. Understanding how to read volume properly is what prevents the most common failure in keyword research — chasing big numbers that convert into nothing.
Search intent is what the person is actually trying to accomplish, and classifying it is the single most consequential step. The familiar categories are informational (learning something), navigational (reaching a specific site), commercial (researching options before a decision), and transactional (ready to act). The same words can carry different intents, and the intent determines what kind of page can possibly satisfy the query.
The practical rule is to classify queries by intent before targeting them, because intent dictates the content format. An informational query needs a genuinely useful explanation; a commercial query needs comparison and evidence; a transactional query needs a page that lets people act. Targeting an informational query with a sales page fails regardless of optimisation, because it does not do the job the searcher came to do. Understanding intent is why classification precedes planning.
Keyword difficulty estimates how hard it would be to compete for a term, based largely on the strength of what currently ranks. It is a directional estimate rather than a precise measure, but it serves an essential purpose: preventing you from planning content that has no realistic prospect of visibility. A term dominated by highly authoritative sources is not a good target for a site without comparable standing, however attractive its volume.
The practical approach is to weigh difficulty against your own authority and pick the winnable fights. A new or specialist site should concentrate on terms where its depth and specificity can compete, building the standing that makes harder terms reachable later. An established authority in a domain can target more competitive terms within it. Understanding difficulty as relative to your position is what turns a keyword list into a realistic plan rather than an aspirational one.
Research begins with seeds — the obvious core terms describing what you do — and expands outward. Keyword tools generate related terms and questions from those seeds along with their estimated volume and difficulty. Search autocomplete reveals what people commonly type. People Also Ask boxes expose the related questions users are asking around a topic. Your own site search, sales conversations, and support tickets reveal the language customers actually use.
The practical method is to expand systematically from seeds using several of these sources, since each surfaces terms the others miss: tools give scale and estimates, autocomplete and People Also Ask give authentic phrasing and questions, and internal sources give the vocabulary of people who are already your customers. Understanding how to expand seeds broadly is what produces a comprehensive picture of demand rather than a narrow list reflecting only what you already assumed.
Volume alone is the most misused number in SEO. Match real demand and intent, weigh difficulty against your actual authority, and classify what the searcher is trying to do — because intent decides what page could possibly satisfy them.
Long-tail queries — longer, more specific searches with lower individual volume — are where most sites should concentrate, for three reasons. They are less competitive, so they are winnable without established authority. They carry clearer intent, since a specific question reveals precisely what the person wants. And collectively they represent a large share of all searching, because specific needs vastly outnumber generic ones.
In the AI era there is a fourth reason: long-tail questions are disproportionately where AI citations are won, because engines decomposing a query into sub-questions retrieve content that answers specific points precisely. A page that answers a narrow question well is exactly what such retrieval looks for. Understanding why the long tail matters is why new sites and specialists should build there first — it is both the winnable ground and the ground AI answers reward.
The research is only useful once it becomes a plan. The translation involves grouping related terms that a single page could satisfy — since many variations share one underlying intent and should not become separate pages — deciding what kind of page each group requires based on its intent, and sequencing by a combination of winnability and business value. The output is a set of planned pages, each with a clear purpose and target audience.
A common error is treating every keyword as a page, which produces thin, overlapping content that competes with itself. Grouping by intent prevents this: if several queries want the same thing, one strong page serves them all better than several weak ones. Understanding how to translate a list into a plan is the step that converts research into action, and it is where most of the judgment in keyword research actually lies.
People type keywords into search and full questions into AI assistants. DUNkē tracks which prompts you’re cited for across eight AI engines — against competitors — so your research covers both how people search and how they ask.
Keyword research has not become obsolete, but it has become incomplete. Traditional search still runs on queries, so understanding the terms people use remains foundational to what you publish and how you frame it. What has changed is that a growing share of information-seeking happens through full conversational questions posed to AI assistants, which are longer and phrased differently from the compressed keyword strings people type into search boxes.
The practical consequence is that keyword research now needs a companion discipline: question and prompt research, covered in its own piece, which maps the conversational questions people actually ask AI systems. The two are complementary rather than competing — keywords tell you how people search, prompts tell you how people ask. Understanding this is why modern research covers both, rather than treating traditional keyword work as either sufficient or obsolete.
Keyword tools are useful and worth using, but their limits are worth understanding. Volume figures are modelled estimates, not counts, and can be substantially wrong for individual terms. Difficulty scores are proprietary approximations of competitiveness, useful directionally but not authoritative. And tools systematically under-represent very specific long-tail queries and newer terms, because low-volume data is unreliable and recent shifts take time to appear.
The practical stance is to use tools for scale and rough prioritisation while treating their numbers as directional, and to supplement them with sources they cannot see: autocomplete, People Also Ask, community discussion, customer conversations, and your own analytics. Understanding tool limits prevents both over-trusting precise-looking numbers and dismissing terms that tools report as zero-volume but which real people demonstrably ask.
One of the most underused sources in keyword research is data you already own. Search Console shows the actual queries for which your pages are already appearing, including many you never targeted — a direct record of real demand meeting your content. Your internal site search reveals what visitors want but cannot find. Support tickets and sales conversations capture the questions customers actually ask, in their own words.
These sources have a decisive advantage over external tools: they reflect your actual audience rather than a modelled average. The practical routine is to mine Search Console for queries where you appear but rank poorly — often the fastest wins, since relevance is already established — and to harvest customer language from support and sales. Understanding the value of first-party data is why research should start with what you already know before turning to tools.
A recurring mistake is grouping keywords by textual similarity rather than by what the searcher wants. Two queries sharing most of their words can have entirely different intents, while two queries sharing almost no words can want exactly the same thing. Grouping by string similarity therefore produces incoherent clusters that lead to pages trying to serve incompatible needs.
The practical method is to group by the underlying job: what would satisfy this person? Queries that would be satisfied by the same content belong together regardless of their wording. A useful check is to examine what currently appears for each query — if the results are similar in kind, the queries likely share intent; if they differ markedly, they do not. Understanding intent-based grouping is what produces pages with a coherent purpose rather than pages hedging between several.
With grouped, classified opportunities in hand, prioritisation balances two axes: how much a win would be worth to the business, and how realistically you could achieve it. High-value, high-winnability opportunities are obvious first choices. High-value, low-winnability terms belong in a longer-term plan supported by the authority-building that makes them reachable. Low-value opportunities are worth skipping regardless of how easy they look.
Business value is not the same as volume: a low-volume query used by people about to make a decision may be worth far more than a high-volume query used by casual browsers. The practical exercise is to score opportunities on both axes and sequence accordingly, revisiting as your authority grows. Understanding prioritisation as value against winnability is what stops keyword research from producing an undifferentiated backlog with no clear starting point.
Search demand shifts: new terms emerge as products and concepts appear, phrasing changes as language evolves, seasonal patterns recur, and the competitive landscape moves as others publish. Research conducted once and never revisited gradually stops reflecting reality, leaving plans anchored to demand that has changed. This is particularly true in fast-moving fields where vocabulary can turn over quickly.
The practical routine is periodic refresh: revisiting the research on a real cadence, checking Search Console for new queries you are appearing for, monitoring for emerging terms in your field, and adjusting priorities as difficulty and demand shift. Understanding that research needs maintenance is why it should be treated as an ongoing input to planning rather than a one-time project completed at the start of a content programme.
The recurring mistakes are consistent across organisations. Chasing volume while ignoring intent produces traffic that never converts. Targeting terms far beyond your authority produces content that never surfaces. Creating a page per keyword produces thin, cannibalising content. Grouping by string rather than intent produces pages with no coherent purpose. Trusting tool numbers as precise leads to misplaced confidence. And treating research as a one-off leaves plans anchored to stale demand.
The remedy in each case follows from the fundamentals: classify intent first, weigh difficulty against your real position, group by underlying job, treat tool figures as directional, mine your own data, and refresh periodically. Because these mistakes are so common, avoiding them is itself a competitive advantage. Understanding the failure modes is often more practically useful than any single technique, since most poor keyword strategy is a matter of these errors rather than missing sophistication.
A distinction worth drawing early is between branded terms — those including your name — and non-branded terms describing what you do generically. They behave completely differently: branded terms convert far better because the searcher already knows you, while non-branded terms represent the demand you have yet to capture. Mixing them in analysis systematically flatters performance, because brand traffic can grow through entirely separate marketing while non-brand visibility stagnates.
The practical discipline is to separate them in both research and reporting. Branded demand is largely a downstream effect of brand-building rather than something keyword research addresses, while non-branded terms are where content strategy actually operates. Understanding this split is why non-brand performance is the more honest measure of whether your search work is producing results, and why an aggregate view can conceal a genuine problem for months.
Many terms fluctuate predictably across the year — seasonal products, annual events, budget cycles, weather-dependent needs — and average volume figures obscure these patterns entirely. A term with modest annual average volume may be enormously valuable in a two-month window and irrelevant the rest of the time, which changes both its worth and when content for it should exist.
The practical implication is to check demand trends over time rather than relying on averages, and to publish seasonal content well before the peak so it has time to be discovered and indexed. Planning to publish at the moment demand arrives is usually too late. Understanding seasonality is why the calendar belongs in keyword planning, and why a term’s timing can matter as much as its volume.
Question-phrased queries — those beginning with how, what, why, which, or when — deserve particular attention because they carry unusually explicit intent. A question states precisely what the searcher wants to know, which makes both the required content and the right format obvious in a way that ambiguous noun phrases do not.
They also bridge directly to AI visibility, since question-shaped queries most closely resemble the prompts people pose to assistants, and question-and-answer content is disproportionately extracted into answers. The practical approach is to harvest question queries deliberately from People Also Ask, autocomplete, and community sources, and to structure content around them. Understanding the value of question keywords is why they often deserve priority over higher-volume but vaguer terms.
One of the most informative research steps is often skipped: actually looking at what currently appears for a target query. The results reveal what the engine considers a satisfying answer — the content formats that rank, the depth expected, whether answer features appear, and what kind of sources dominate. This is direct evidence about intent that no volume figure supplies.
The practical routine is to examine the results for any query you intend to target seriously before planning the content, and to build something that fits what the engine demonstrably rewards while being better than what is there. Understanding results-page reading as a research method is why keyword tools alone produce weaker plans: they describe demand without showing what satisfying it actually requires.
Keyword research has diminishing returns, and it is possible to spend so long refining a plan that nothing gets published. Beyond a certain point, additional research produces marginal refinement while the real learning — what actually performs for your site, in your market — can only come from publishing and measuring.
The practical guidance is to research enough to identify a defensible set of priorities, then publish and use the resulting performance data to refine. Search Console will tell you within weeks what estimates could only approximate. Understanding when to stop is why research should be sized to unblock action rather than to achieve completeness, and why the strongest keyword plans are usually the ones that have been revised in light of real results.
A keyword research output that records only terms and figures loses the most valuable part of the exercise: the reasoning. Why a high-volume term was declined, why an apparently obscure query was prioritised, which competitor holds a space you decided not to contest, what assumption a priority rests on — these judgments are what make the plan defensible six months later when someone asks why a promising-looking term was ignored.
The practical addition is a short rationale column alongside each priority, and an explicit list of considered-and-rejected opportunities with reasons. This costs little at the time and prevents the same debates recurring, while making the plan reviewable when circumstances change. It also protects against a common organisational failure, where a plan built on careful reasoning is later overridden by whoever most recently looked at a volume figure, because the reasoning was never written down anywhere.
Keyword research finds the terms and questions your audience actually uses and turns them into a content plan, resting on three measures: demand, difficulty, and — most consequentially — intent, which dictates what kind of page could satisfy the query at all. Volume is the most visible and most misused of the three: match real demand and intent rather than chasing big numbers that convert into nothing.
The practical method is to expand from seeds using tools, autocomplete, People Also Ask, and your own first-party data, classify by intent, weigh difficulty against your actual authority, group queries by the underlying job rather than by wording, and prioritise by value against winnability. Concentrate on the long tail, where wins are achievable and AI citations are disproportionately earned — and pair this with prompt research, since people search in keywords but ask in questions.
“Volume is the most visible number in keyword research and the most misused. What matters is whether real demand meets real intent — and whether you can actually win it.” The Age’X Research Team
Get a free GEO audit — the same analysis behind every article here.