E-Commerce Keyword Research

Keyword research for a store is matching search intent to a page type in your catalog, then writing titles and copy that page can honestly support. It is not a list of phrases to paste into a homepage, and it is not a density quota.

If you need the crawl/index/rank picture first, read How E-Commerce SEO Works. This article is the mapping step.

Three query families

Shoppers do not all type the same kind of query.

Product queries name a model, SKU, or a specific configuration: “Sony WH-1000XM5 navy,” “replacement gasket for AeroPress.” These belong on a product page (or a variant URL if you have chosen to index that variant). See Product Page SEO.

Category queries name a class: “noise cancelling headphones,” “AeroPress accessories.” These belong on a collection/category with a grid of relevant SKUs and unique intro copy—not on a single hero product. See Category Page SEO.

Problem or job queries describe an outcome: “headphones for commuting,” “how to pack coffee for travel.” These often belong on a guide or a curated collection. Linking from that page to in-stock SKUs is the conversion path. Do not force a problem query onto a PDP that only sells one model.

A fourth bucket—brand + product—is usually a PDP or a brand collection. Do not build a fake “brand page” that duplicates the manufacturer’s site with no stock.

Map to information architecture, not to a spreadsheet tab

Intent	Example query	Primary URL	What the page must prove
Product / model	“Model X 64GB black”	PDP	Specs, price, stock, variant truth
Category	“64GB USB drives”	Collection	Range, filters, unique category copy
Problem	“backup photos on the go”	Guide or curated collection	Advice + links to SKUs you sell
Comparison	“Model X vs Model Y”	Guide or comparison module	Honest differences, both in stock or clearly labeled

Process:

1. Dump seeds: current titles, site-search logs, marketplace search term reports, autocomplete on your target locales.
2. Cluster by intent (the table above), not by shared words.
3. Assign one primary URL per cluster. Secondary URLs can link; they should not all target the same title.
4. Note gaps: queries you cannot serve because you do not stock the item. Those are merchandising decisions, not SEO tasks.

Use the URL slug generator when you create new collections so slugs stay consistent. Preview the SERP with the SERP snippet preview.

What not to do with the list

Homepage stuffing. The homepage is navigation and trust. It cannot rank for fifty categories. Give each category its own URL.

Repeating the exact phrase in every H2. Readers bounce; you have not “optimized.” A spec table and a unique first paragraph beat a mantra.

Chasing volume for SKUs you will not fulfill. A high-volume term for a product you cannot ship in that region is a support queue.

Copying competitor title tags blindly. They may be ranking for brand, or they may be stuffed. Look at the SERP: shopping packs, marketplaces, and thin affiliates change whether a category page can win.

The keyword density checker is a lint tool. If one phrase is 8% of a 200-word description, you probably wrote a keyword, not a product. Fix the prose.

Templates vs unique copy

At catalog scale you will use templates: {{brand}} {{product_type}} {{attribute}}. That is fine for titles if the attributes are real. It is not fine if every description is the same paragraph with the SKU swapped.

Prioritize unique copy where:

• Margin or attachment rate is high.
• The SERP is not dominated by the manufacturer and Amazon.
• Variants actually differ (material, fit, compatibility).

The rest can be a tight spec-led template plus one unique sentence (compatibility, included accessories, who it is for).

Research sources that match how stores actually get queries

• On-site search. If people search “navy” and you only have “blue” in the title, that is a data problem.
• Marketplace term reports. These are commercial queries with purchase intent. They also include terms you should not copy if they violate the marketplace’s rules—use them as intent clues, not as a stuffing list. For Amazon listing structure, see Amazon Product Listing Optimization.
• Support tickets. “Does it fit X?” is a PDP FAQ and a schema candidate only if the answer is visible on the page (schema markup).
• Paid search search terms (if you run ads). Negatives in ads are often missing collection pages.

Keyword tools estimate volume. Treat estimates as relative (this cluster is bigger than that one), not as a forecast of orders. No research document ranks a URL. Publishing a crawlable, unique page that matches the intent does.

Local and language variants

If you ship to more than one country, the same English query may need different collection copy (voltage, sizing systems, duty notes) on different URLs—not one stuffed page. Do not map “colour” and “color” to two indexable URLs that duplicate the grid; pick a locale and redirect.

Misspellings of model names belong as 301s or as on-site search synonyms, not as extra thin PDPs.

A mapping session that fits on one sheet

Columns: query cluster, intent (product / category / problem), assigned URL, in stock?, unique copy needed (Y/N), notes. Walk the top 50 clusters. Anything without a URL is a merchandising or content ticket. Anything with two URLs is a cannibalization ticket.

For the full catalog SEO sequence, use the e-commerce SEO guide. After the map exists, write the product and category templates to match it.