Keyword ResearchReading demand, not lists

What keyword research is really for#

Keyword research is usually presented as list-making: export thousands of terms, sort by volume, start writing. That produces a spreadsheet, not a strategy. The real job is to read demand: what your buyers want to know, in their words, at each stage of a decision, and which of those needs your business can answer better than anyone else.

The output that matters is not a keyword list but a page plan. Each important page should own a cluster of related queries with one shared intent. If research does not end with 'this page, for these searches, in this order', it is not finished.

Volume tells you how many people are asking; intent tells you whether they are asking you.

Step 1: Start from the business, not the tool#

Before opening any tool, gather the language your customers already use. Tools only show variations of the seeds you give them, so the seeds decide the ceiling of the research.

• Sales and support conversations: the questions prospects ask before buying and the objections they raise.
• Search Console queries: what people already search when they find you, including terms you never targeted. See Search Console.
• Site search logs: what visitors look for once they arrive and do not find.
• Product and service names, including the generic names buyers use rather than your brand terms.
• Problems before solutions: what buyers search when they know they have a problem but not what fixes it.

In markets like India, also capture transliterated and mixed-language queries, regional spellings and the way people phrase searches by voice. These often have less competition precisely because generic tools under-report them.

Step 2: Expand and group#

Now use keyword tools to expand the seeds into related terms, questions and modifiers. Every major SEO tool does this; the differences between them matter less than how you use the output. Search volume figures from third-party tools are estimates, often rounded or bucketed, and should be read as relative size, not precise counts.

Then group. Queries that would be satisfied by the same page belong together. The quickest test is the results page itself: if two queries return largely the same top results, search engines treat them as one intent, and you should too. Grouping turns ten thousand rows into perhaps a few hundred page-sized topics.

Step 3: Read the results page for every priority group#

For each group you might target, look at the live search results. They tell you three things a tool cannot. Intent: are people looking to learn, compare, buy or find a specific site? Format: guides, lists, product pages, videos, local packs? Competition: are the top results from large publishers, marketplaces, government sites, or businesses like yours?

If the results are dominated by marketplaces for a product query, a single brand's product page may struggle regardless of its quality. If they are dominated by forum threads, the searcher wants lived experience, and a polished corporate page may not fit. Read search intent for the full method.

Step 4: Judge value, not just volume#

A query with modest volume and clear buying intent often outperforms a huge informational query that attracts students and the merely curious. Score each group on three questions: how valuable is a visitor from this search, how feasible is ranking, and how strongly does it fit what you actually sell?

Treat difficulty scores from tools with care. They are mostly a function of the links pointing at ranking pages and say little about content quality or intent fit. Your own judgement after reading the results is usually a better guide.

Step 5: Map groups to pages#

Every priority group now gets a home. It may be an existing page that needs improving, a new page, or two existing pages that should be merged because they compete for the same intent. That last case, known as cannibalisation, is common on older sites and is one reason content pruning belongs alongside research.

Keyword research when answers come from AI#

Searches are getting longer and more conversational, and AI answer features now resolve many simple questions on the results page. This changes research in two ways. First, short definitional queries may deliver fewer clicks than their volume suggests, so weight them lower unless being the cited source has brand value for you.

Second, the questions behind a decision matter more than the head term. People ask assistants things like 'which accounting software suits a ten-person firm that bills in two currencies'. Few tools show volume for such prompts, but the underlying needs are discoverable through sales conversations and the follow-up questions that appear in results. Pages that address those specific situations clearly are the ones answer engines can quote. See answer engine optimisation.

Documenting research so it stays useful#

Research decays quickly when it lives in one person's spreadsheet. Keep a single shared map with one row per intent group, not per keyword. Each row should record the primary query, a handful of representative variants, the intent, the format the results reward, the assigned URL, its current status and the date the results were last reviewed.

That map becomes the reference for editors, developers and anyone briefing content. It also makes cannibalisation visible: if two rows point at the same URL with different intents, or one intent has two URLs, the conflict is obvious before it costs rankings. Review the map whenever products, services or markets change, not just on a calendar.

Common mistakes#

• Chasing head terms that are broad, ambiguous and dominated by giants, while ignoring specific queries buyers actually use.
• One page per keyword variant, producing near-duplicate pages that compete with each other.
• Ignoring zero-volume queries. Tools under-report niche and new terms; if customers ask it, someone searches it.
• Treating research as a one-off. Demand shifts with seasons, products and language. Revisit priority groups at least twice a year.
• Skipping the results page. Most targeting errors would be caught by a two-minute look at what currently ranks.

Once the plan exists, it becomes the backbone of topic clusters and editorial briefs. Each brief should name the primary group, the secondary questions to answer, the format the results reward and the evidence only you can provide.

Published by Fabulous.Media, a network of specialist marketing agencies. Updated 9 October 2026. Platform features change often; check current official documentation before acting on platform-specific detail.