Search Demand Analysis: A Qualification Workflow for Ranking Search Opportunities
A step-by-step method for turning a raw query universe into a scored, go/no-go shortlist, so you stop treating a volume sort as demand analysis.

Why a Volume Sort Is Not Search Demand Analysis
Export a keyword list, sort by volume, work top-down. That is a qualification-free prioritisation, and it fails for one structural reason: volume measures how often a query is typed, not whether the searcher is a viable customer for a page you can build and rank. Search demand analysis is the qualification gate between raw query data and content budget. Its output is not a list. It is a scored, sequenced shortlist with a documented reason each cluster earned a yes or a no.
Rule of thumb: if your analysis cannot produce a NO-GO decision, it is keyword research wearing a demand-analysis hat.
The Three Demand Layers: Total, Addressable and Obtainable
Before scoring anything, split the universe into three layers. Total demand is every query in the space. Addressable demand is what your inventory, services or expertise can actually satisfy. Obtainable demand is what your site can realistically rank for given authority and SERP composition. A category with 40,000 monthly searches where 30,000 sit behind a marketplace and 8,000 require stock you do not carry has an obtainable figure far smaller than the headline.
Prerequisites: What You Need Before Scoring Any Opportunity
- A directory or service list you can map queries against (your own taxonomy, not a third-party one).
- Margin data per product line, or average lead value per service, so commercial value is a number.
- Search Console query export for existing coverage, to spot clusters you already partially serve.
- A notepad for SERP notes: which features occupy the first screen for each candidate cluster.
Step 1: Build the Query Universe from Your Own Taxonomy First
Start inside out. Walk your category tree and write the modifier patterns a buyer would attach to each node: best [item], [item] under R1 500, [item] for [use case], [item] vs [item], [spec/material] [item]. Then expand outward with tools. This order matters: tool-first discovery produces a universe shaped like the tool's index, full of queries you cannot serve.
Step 2: Cluster by Intent, Not by String Similarity
String-similar clustering groups "cheap running shoes" with "best running shoes" because they share tokens. They are different demands: one is price-led and comparison-shallow, the other is evaluation-led. Cluster by what the searcher is trying to decide. A cluster earns its own page only if a distinct searcher job maps to one satisfying answer.
Step 3: Score Each Cluster on Value, Servability and Confidence
Score five dimensions, then combine as a weighted score or plot on a value-versus-servability 2x2. A single opaque number hides the decision.
| Dimension | What you score | Scale |
|---|---|---|
| Intent stage | Research / compare / transact | 1-3 |
| Commercial value | Margin or lead value in ZAR | 1-5 |
| Servability | Live SERP composition, stock and page fit | 1-5 |
| Content cost | Effort to build a genuinely useful page | 1-5 (inverse) |
| Confidence | Data reliability across sources | 1-3 |
The One High-Volume Head Term That Hides Three Different Demands
Take "mountain bike". Under that phrase sit a researcher comparing frame types, a size-checker, and a buyer searching a specific store. A single head-term sort averages three intents into one meaningless score. Split the phrase into three clusters, score each on the table above, and the transact-weighted cluster usually wins despite a fraction of the volume. Query ambiguity, not low volume, is the most common reason a priority list misfires.

