How to do keyword research for a Shopify store

The short answer

Keyword research for a Shopify store is territory mapping: establish what the store sells and which categories that maps to, gather the real queries buyers use at each stage of the purchase from five inputs, separate transactional intent from informational, and assign every cluster a destination, a collection page for the buying queries, editorial for the research queries. The output is an architecture plan, and the spreadsheet is only the working file behind it.

The distinction matters because Shopify's commercial power sits in its collection architecture. The collection pages that rank are the ones built around a specific, commercially defined query cluster, and those clusters come out of territory mapping.

Keyword research and territory mapping are different jobs

Traditional keyword research starts with a seed keyword, expands it into a list, filters by volume and difficulty, and produces a spreadsheet. That is a reasonable process for informational content, but it is the wrong process for Shopify category pages, because it answers "what could we write about" when the commercial question is "which pages should exist".

Territory mapping runs the other way. It starts from the catalogue and the buyer: what does the store sell, which categories do those products belong to, and what are the specific queries used at each stage of the purchase journey in each of them? The keyword list comes from mapping that territory, and the map ends as decisions, this collection exists, this one consolidates, this content gets written, and never as research that sits in a file untouched by the store.

The five inputs

1. Competitor rankings. The top-ranking collection pages of two or three direct competitors show which territories are rankable, what depth the competing pages carry, and which queries the market has already validated. This input short-circuits the hypothesis stage entirely, and it comes first.
2. Search Console. The queries the store already ranks for, even weakly, and the pages earning impressions without clicks, are the territories closest to paying off. Filter the performance report by query cluster and read what Google already believes the store is about.
3. A keyword tool. Keyword Planner, Ahrefs or similar, used to size and expand the clusters the first two inputs found. The tool is the third input, never the first, because tools expand what you ask them about and cannot tell you which questions matter.
4. The store's own language data. On-site search logs, support questions and the Knowledge Base's unanswered-questions list carry buyer vocabulary no external tool sees, including the misspellings and colloquialisms real buyers type.
5. Autocomplete surfaces. Google and Amazon suggestions for the category's head terms surface long-tail phrasings and buying qualifiers, size, occasion, price bracket, that structure the sub-clusters.

Reading a competitor's territory, step by step

The competitor input deserves its own method, because it is the input most teams do superficially. Choose the competitor closest to your positioning rather than the biggest name in the category, since the market leader ranks on authority you cannot copy, while the peer brand ranks on decisions you can.

1. Read their navigation and collection structure as a customer, noting which categories they chose to make pages for, and which they left as filters.
2. Pull their top organic pages from a tool's top-pages report, and keep only the collection URLs, which is their commercial territory ranked by value.
3. For each of their winning collections, read the page itself: content above and below the grid, metadata, and which sibling collections it links to. Depth tells you the cost of competing, and thinness tells you where they are beatable.
4. Note the queries where they rank with a product page or a blog post, because a query the market serves with the wrong page type is an opening for the right one.

Two competitors read this way, plus your own Search Console, produces most of the map before a keyword tool opens. The tool's job is then confirmation and expansion, which is what tools are good at.

The commercial intent filter

Not all search volume is equal. A query with 8,000 monthly searches from people who are researching is worth less than a query with 800 monthly searches from people who are ready to buy.

The commercial intent filter separates transactional queries (buy, shop, near me, delivery, specific product names) from informational ones (what is, how to, best, guide, review). Collection pages get built around the transactional clusters, and editorial content gets built around the informational ones. Where a pattern is ambiguous, search the query yourself and read what Google chose to rank, because the result mix is the intent verdict in public.

The most common mistake we see is the architecture running the other way round, with collection pages chasing informational queries like 'how to choose walking boots' and blog posts chasing transactional ones like 'waterproof walking boots'. Swapping those assignments is often the single most valuable change an audit produces.

A worked slice of the map

For an illustrative premium sleepwear brand, one category's slice of the finished map looks like this. Volume bands are deliberately coarse, for reasons the next section covers.

Every row ends in a destination, which is the difference between a map and a list. Reading down the destination column gives you the collection architecture and the content calendar in one pass, and any query the map cannot place is either noise or a gap in the architecture worth noticing.

Search volumes lie, in both directions

Tool volumes are estimates, averaged over twelve months, rounded into bands, and blind to seasonality until you look at the trend line. Treat them as ordering information, this cluster is bigger than that one, rather than as forecasts, and never let a 20 percent volume difference decide an architecture question that buyer logic answers better.

They also miss demand in both directions. Zero-volume long-tail queries convert startlingly well in aggregate, because specificity is intent, and a store's collection pages collect that long tail automatically when the architecture is specific. Meanwhile the informational volumes are increasingly answered before the click: with AI summaries present, clicks on conventional results drop to roughly half their usual rate on Pew's data, so an informational cluster's tool volume overstates the visits it can deliver. That does not make informational content pointless, it changes its job, from traffic capture to evidence for AI citation and category authority.

Multi-market territory is separate territory

A brand selling into several markets maps each one separately, because vocabulary is market behaviour. The same product is a jumper in the UK and a sweater in the US, gumboots in Australia and wellies in England, and volume tools respect those borders only when you filter by country before reading anything.

Language multiplies the effect. A Hong Kong brand maps English, Traditional Chinese and often Simplified Chinese as three territories with their own clusters, since the two Chinese scripts diverge in vocabulary and search behaviour, and a translated keyword list is not a keyword list. The practical rule for any market: the map is built from that market's queries, validated by someone who shops in that market's language, and the architecture serves each market's clusters on its own locale instead of assuming demand translates.

From map to work plan

1. Architecture decisions first: which collections exist, which consolidate, which get created, per the collection structure method.
2. Metadata and on-page for the surviving collections, cluster by cluster, highest commercial value first.
3. Content calendar from the informational clusters, each piece assigned its cluster and its linking targets before it is written.
4. Quarterly re-read of Search Console against the map, because the map is a hypothesis the data keeps correcting.

The full architecture method, consolidation runbook included, is in our collections structure guide, and the map is its required input.

Questions we hear about keyword research