Keyword Research and Search-Demand Guide

What does a demand estimate actually measure?

A demand estimate measures how often a query and its close variants were searched, in a chosen location and network, averaged over a chosen period. Google's figure is the average monthly searches column in Keyword Planner, which Google Ads Help defines as the average number of times people searched for a keyword and its close variants, based on the month range, location and Search Network settings you selected; by default it is averaged over twelve months, and Google states that the statistics are rounded.

A twelve-month average smooths seasonality: a query that peaks before Hari Raya shows as a steady figure. The count already includes close variants, so adding two rows that are variants of each other double counts. At the volumes typical of Malaysian queries, a rounded 10 or 20 a month is an approximation, and the gap between two such figures is noise.

Keyword Planner is an advertising tool, and its other columns describe advertisers, not searchers: competition counts advertisers relative to all keywords across Google, and the top-of-page bid columns approximate what advertisers paid. A high bid says the query is worth money to someone; it says nothing about how hard the query is to rank for. Forecasts are a separate product adjusted for bid, budget and seasonality; for organic planning use the historical metrics only.

Where the queries come from: seeds, your own data, and expansion

Start with seeds, not with a tool. A seed list is the ten to thirty words a customer uses before they know your vocabulary: the problem, the service, the place, the price question, the comparison. Write it from sales calls, WhatsApp enquiries and front-desk questions, in the languages customers use; for many Malaysian businesses that means a Bahasa Malaysia set and an English set, each its own audience rather than a translation. A ready-made list, the thing a search for web design keywords hopes to find, skips the seeds and the decisions, which are where the value is.

Keyword Planner's discover function takes seed phrases, or a website address from which Google extracts related terms, and returns ideas with the historical metrics above. Your Search Console performance report is the only first-party source: grouped by query, it lists the searches for which your site appeared, with impressions, clicks and the average position of your topmost result, for the past three months by default. Google notes that a query in that list does not mean you would see your site if you searched it yourself, and Planner ideas are generated, not vetted. No query is qualified demand until a person has decided what task it represents.

How should competitor query evidence be retained?

Retain competitor query evidence as dated observations of what ranked, never as demand. A competitor-ranked query is one for which you saw a competitor's page in the results, from a stated location and device, on a stated date, at a stated position. Record exactly that, plus the kind of page it is (service page, guide, price list, directory listing or map listing). It is a fact about one results page on one day: Google judged that page a good answer to that task. It does not say how many people searched, and it does not tell you to build the same page.

Keep the observation lane separate from the demand lane even when they agree, because the disagreements are the information. No Planner figure but a competitor ranking a dedicated page suggests a real, small demand rounded to nothing or folded under a variant. A healthy figure with no competitor page suggests a gap worth taking, or a task answered by a result type a business page cannot win, such as a forum thread or a government portal. Third-party tools that list a competitor's ranking keywords estimate from their own index; in our experience they miss and mis-rank queries in a market the size of Malaysia often enough that no entry should be acted on before it is checked on a real results page. Your page still has to be a better answer built from your own information, not a rewrite of theirs.

The two-lane worksheet

The worksheet joins the two lanes on the query and forces a decision per row. The sample is for a hypothetical Johor Bahru physiotherapy clinic; every figure is invented to show the method.

The workflow, step by step

1. 1

   Collect seeds

   Customer words from enquiries, calls and reviews, in the language used.

2. 2

   Expand and label the source

   Keyword Planner ideas, three months of Search Console queries, competitor observations; every row carries its source and date.

3. 3

   Fill both lanes

   The Planner figure for Malaysia and your own impressions and position; then what ranked from a Malaysian location on a phone. Leave cells empty rather than guessing.

4. 4

   Write the intent and cluster

   One line on what the searcher is trying to do; rows with the same task and result types form one cluster.

5. 5

   Decide and assign one owner

   Own, fold, cover as a section, or leave; exactly one accountable page per cluster. Unassigned rows stay as a log.

When should close variants share a page?

Close variants share a page when they describe the same task and Google shows the same kind of result for them. Google's SEO Starter Guide makes the point from the reader's side: people with different expertise use different words for the same thing (its example is charcuterie against cheese board), and you do not need to anticipate every variation because Google's language systems can relate a page to a query without an exact match. The same guide warns that repeating words excessively, even in variations, tires readers and that keyword stuffing is against Google's spam policies.

The test is the results page, not the spelling. If the same pages, or the same types of page, appear for physiotherapy johor bahru and physio jb, they are one task. If physiotherapy cost returns price guides while physiotherapy johor bahru returns map results, they are different tasks even though they share a word. In our experience word order, plurals, abbreviations and a city spelt out or shortened are almost always one task; adding cost, near me, versus, how to or a condition name usually changes it.

Language is the Malaysian exception. A Bahasa Malaysia query is not a close variant of the English one; it is the same task for a different audience, and the honest answer is a page in that language. Whether that is a separate language version of the site or a bilingual page is a decision for your technical SEO set-up; in the worksheet it is a second row with its own owning page.

When is a zero-volume concept still necessary?

A zero-volume concept is necessary when the owning page cannot complete its task without it. Demand tools show no figure, or a small rounded one, for most of the specific questions customers ask: the figure is rounded, close variants are grouped under a head term, and Malaysian volumes for specific phrases tend to be small. None of that means nobody asks. The front-desk question, the objection on every sales call, the step every ranking page covers: these belong on the page as sections, chosen by semantic necessity rather than volume. Would a reader have to leave and search again without the section? Then it is necessary.

Google's guidance on helpful content points the same way. Its warning signs for search-engine-first content include producing lots of content on many topics hoping some performs, chasing topics because they seem to be trending, writing to a word count, and mainly summarising what others have said; its positive signs include first-hand expertise, original information, and readers leaving satisfied that they learned enough. A section answering the question your business hears daily, written by the person who answers it, is what that guidance describes; a page created because a tool showed a number is what it warns against.

SERP validation and page ownership

Validate every cluster you intend to own by looking at the real results page from a Malaysian location, on a phone, before the page is planned. Record what is above the fold (map results, an AI overview, shopping results, videos or plain organic results), then the types of page that fill the organic results. The result type is the strongest evidence of the task Google thinks the query represents, and it often overrules the demand figure: a large figure that returns only map results is a Business Profile job, covered in local SEO and the Business Profile foundations guide; a small figure that returns thin directory listings may be the easiest genuine gap on the list.

Then assign ownership: one page, existing or planned, is accountable for each cluster. Check existing pages first; the commonest finding is that the home page or a service page already half-answers a cluster and needs reworking rather than a sibling. Where the decision is to build, the on-page checklist turns the row into a brief and the internal linking guide connects the new page to the steps around it. Once the page has been live long enough to appear, the Search Console queries it shows for are compared with the cluster it was meant to own, as described in the Search Console measurement guide. Google's starter guide is candid that changes can take hours or months to show, so compare after weeks, not days.

What to do next

1. 1Write your seed list from the last twenty customer enquiries, in the languages they arrived in, before opening any tool.
2. 2Pull Keyword Planner historical metrics for Malaysia for those seeds and export three months of Search Console queries into the demand lane, dated.
3. 3Search your ten most important candidates from a Malaysian location on a phone and record what ranks in the observation lane.
4. 4Fill the decision and owning-page columns for every row. If most rows point at new pages, the clustering is too fine; if one existing page owns most rows, start there.