How do I do keyword research for an online store?

Three sources tell you how your buyers describe what you sell, in their own words, with no estimation involved.

Your internal site search log. Every term typed into your own search box is a person telling you what they wanted and whether you had it. Sort by volume, look at the terms returning no results, and you have found both content gaps and product gaps. This is the single most underused report in ecommerce.

Your ads search terms report. If you run any paid search or Shopping, this shows the actual queries that reached you, along with whether they converted. Terms that convert are proven commercial intent, and they should shape the pages you build.

Your support inbox and reviews. The questions people ask before and after buying are the phrasing they would type. They also reveal the concerns that a page has to address to be useful.

Start there and you will have a few hundred real terms before you open a tool. Then use a keyword tool to find volume, variations and phrasings you have not seen, rather than to generate the list from scratch.

The mistake that wastes the most effort is treating all keywords the same. A store has distinct page types and each answers a different kind of search. Sorting by intent turns a list into a plan.

Category searches name a product type, often with an attribute: a type of bag in a colour, a size of tool, a material of flooring. These belong on collection pages, and they are usually the highest value terms a store can win.

Product searches name a specific item, model or brand and model. These belong on product pages, and they are usually easier to win and convert best.

Attribute and filter searches combine a category with a qualifier: under a price, for a use, in a size, for a particular person. These often justify their own filtered collection page, if your platform can make one that is crawlable with its own title and copy.

Question searches ask how, what, which or whether. These belong in guides, and only where the question is real for your buyers.

Brand searches name you. Those need nothing new, just a clean homepage and store pages.

Everything else is probably not worth a page.

Volume is the most visible number in every tool and the least useful on its own. A term with high volume that half matches what you sell will bring visitors who leave, and it will cost months of work to reach.

Score each candidate term on four things instead.

Fit. Does the page you would build genuinely answer that search with products you actually stock. If a shopper typing it would be disappointed, skip it.

Margin. Terms leading to your better margin categories are worth more effort than terms leading to your thinnest ones, whatever the volume difference.

Commercial intent. Terms with a buying shape convert. Terms with a browsing shape mostly do not, and they need a different page and a different expectation.

Realism. Look at what currently ranks. If the results are all large marketplaces and the pages are substantial, a new collection page will not appear soon, and your effort is better spent on a more specific version of the same term where you can genuinely compete.

A short list of terms you can win, on products that make money, beats a spreadsheet of terms you cannot.

The broad category term in your niche is contested by everyone. The specific versions are contested by far fewer, they convert better, and there are more of them than most stores realise.

Build them systematically from the attributes in your own catalogue. Category plus material. Category plus size. Category plus colour. Category plus use case. Category plus who it is for. Category plus compatibility with something else.

Most of those combinations already exist as filters on your site. The question is whether the filtered view can be a real page with its own address, its own title and a paragraph of copy, or whether it is a JavaScript state that no search engine can reach.

Be selective. Creating a page for every possible filter combination produces thousands of near identical thin pages, which is worse than having none. Choose the combinations that people actually search, that you have enough products to fill, and that you are willing to write copy for.

A good rule: if you cannot write two honest paragraphs about that combination and show at least a handful of products, do not create the page.

Research that ends in a spreadsheet changes nothing. Convert it into a queue of specific pages with an owner and an order.

For each page, record the target search, the page type, whether the page already exists or needs creating, what has to change on it, and which products it will feature. That level of detail is what turns research into work.

Order the queue by expected value rather than by ease. Existing pages that already rank on the second page of results are usually the fastest gains, because they need improvement rather than creation.

Then ship in small batches and measure. Ten collection pages properly rewritten, with copy, filters, internal links and titles, will do more than a hundred pages given a new heading.

Keep the research alive. Add new terms from the site search log and the search terms report every month, because your customers invent new phrasings and your catalogue changes. A keyword list built once and filed away is stale within a season.

And record when each page changed, so that when rankings move you can tell what caused it.