Ecommerce keyword research: the complete guide

The first keyword list comes out of your own product data, because those are the only terms you can honestly compete for. Export everything: product titles, product types, collection names, tags, variant options, materials, sizes, and the brands you stock. That export is a map of the demand you are allowed to chase. A store that sells three kinds of dog bed has no business writing a page about pet furniture in general.

Read the export as a stranger would. Most catalogues are named internally. A supplier calls something a Model 4 recirculating unit; a shopper calls it a countertop water filter. A collection is called New In; a shopper searches by the room, the material or the size. Put a second column beside every internal name and write the word a customer would use. That column is the seed list, and it is usually more accurate than anything a tool will hand you, because it comes from the products you can actually ship today.

Then sort the catalogue the way a buyer narrows down. Most people move from a broad type, to a constraint, to a specific product. Someone buying a chair goes from office chair, to office chair for a small room, to a named model. Each of those three steps is a different page on your store, and each needs different words. Writing the ladder out per category tells you how many pages you are missing before you have looked up a single search volume.

Finish this step with an honest count. How many product types do you carry, how many attributes matter inside each one, and how many of those combinations already have a page on the site. Most stores find that the top of the ladder has a page, the bottom has thousands, and the middle rung, the one that captures buyers who know what they want but not which product, is missing entirely.

Your store already records what people ask it, and that data beats any estimate because it comes from people who reached you. Start with internal site search. Every query typed into your own search box is a shopper telling you the exact words they use, and the ones that return no results are a list of either missing products or missing pages. Pull a year of it, strip the obvious typos, and count.

Next, open Search Console. The queries report shows terms where your pages already appear, including the many where you rank on page two and get almost no clicks. Those are the cheapest wins in the whole project, because the page exists and search engines have already decided it is relevant. Sort by impressions with low click share and you have your first work order.

If you run Shopping or search ads, the search terms report is the highest quality keyword source you will ever have, because every term there is attached to spend and, sometimes, to an order. Terms that convert in ads are terms worth a page. Terms that spend and never convert tell you which words look relevant and are not, which saves you writing a page nobody wanted.

Then go outside the analytics. Read the last few hundred customer service messages and pre-sale questions. People write to you in the same language they search with, and they ask about fit, compatibility, delivery and comparison in ways a tool will never surface. Sales team notes, review text and the questions on your product pages all work the same way.

The point of this step is that you finish it with a list built from real demand at your own store, in your own category, in your customers' vocabulary. Everything after this is expansion and sorting, not invention.

Expansion should follow the seed list, not replace it. The goal is to find the variations of terms you already know convert, not to collect a wider pile of words. Autocomplete is the first stop. Type your category word into a search box and read what gets suggested, then repeat with each constraint your buyers care about. Those suggestions are drawn from what people actually type, which makes them more useful than a raw volume number.

Related searches at the bottom of a results page do the same job from the other direction. They show what people searched next when the first attempt did not finish the job, which usually reveals the constraint that matters most in your category. If every related search adds a size, size is the attribute your collection pages should be split by.

Keyword tools are worth using at this stage, once you have a seed list to feed them. Use them for two things: finding modifier patterns you had not thought of, and getting a rough sense of relative demand between two options you are already considering. Treat the volume numbers as a comparison, not a forecast. They are modelled, they are often grouped across close variants, and they do not know your margin.

A few sources are specific to ecommerce and worth the time. The filter names on your own category pages tell you which attributes shoppers already sort by. Product review text shows the words people use after they own the thing, which is where problem language comes from. The questions people ask an AI assistant about your category tend to be longer and more conditional, and they hint at the comparison pages worth building.

Stop expanding when new sources start returning terms you have already seen. That saturation point usually arrives sooner than people expect, and going past it produces a list too large to act on.

Intent decides what you build, so sort before you do anything else with the list. In a store catalogue almost every useful term falls into one of four groups, and each group belongs on a different kind of page.

Category terms name a product type with no constraint. Running shoes, desk lamps, protein powder. These are broad, competitive and usually the hardest to win, but they belong to a collection page and they carry real buying intent, so they stay on the list even when they take a year.

Constrained terms are the middle rung and the most valuable part of the project. A type plus an attribute, a use case, a room, a size, a compatibility, a price position. Waterproof hiking boots for wide feet. Desk lamp with a clamp. These convert better than the broad term, they are winnable, and they map cleanly onto a filtered collection page you can build in an afternoon.

Product terms name a brand, a model or a specific item. These are already served by product pages, and the work here is technical rather than editorial: make sure the right page ranks, the title matches how people write the model, and duplicate or discontinued pages are not competing with the live one.

Question and problem terms are people describing a situation rather than a product. How to stop a lamp glaring on a screen. What size boot to buy for wide feet. These do not convert on the first visit, and they are where a guide or a buying advice page earns its place, but only if the page routes cleanly to the collection that solves the problem.

Tag every term with one of the four labels. Anything that does not fit a label is usually informational traffic with no path to an order, and it should be dropped rather than saved for later. A list that is honestly sorted is already most of a content plan.

One page, one group of terms, no exceptions. The most common ecommerce keyword mistake is not picking the wrong words, it is pointing several pages at the same words and letting them split the signal. A collection, a tag page, a filtered view and an old blog post can all target waterproof hiking boots, and none of them wins.

Group by what the searcher expects to see, not by word similarity. If two terms should show the same set of products in the same order, they are one group and they get one page. If one term expects a grid of options and the other expects a single answer, they are two groups even when the words look close. Boots for wide feet expects a filtered grid. Which boot fits wide feet expects advice, then a grid.

Give every group a primary term and a set of variations. The primary term shapes the page title, the H1 and the URL. The variations shape the copy, the internal links and the filter names. You are not writing the variations into the page repeatedly, you are checking that the page reads naturally to someone who used any of them.

Then assign a page type to each group. Existing collection, new collection, filtered view that needs to be made indexable, product page, or guide. Be strict about the last one. A guide is the right answer only when the search is genuinely a question, and it should always link into the collection that sells the answer.

Finally, resolve the overlaps you find. Where two existing pages target one group, pick the stronger one, redirect the other or consolidate its content, and change the internal links so they all point at the survivor. This consolidation step usually moves rankings faster than writing anything new, because you are giving one page the strength that three pages were sharing.

Difficulty is not a number in a tool, it is a judgement about what already ranks. Open a results page for the term and look at what is there. If the first page is filled with marketplaces, big retailers and comparison sites, a small store is not going to displace them with a better paragraph. If the first page holds collection pages from stores roughly your size, that term is available.

Ask three questions for each group. Do the results look like the page type you planned to build, does anything on that page answer the search better than the pages already ranking, and do you sell enough products to fill the grid the searcher expects. That last one matters more than most people admit. A collection page for a term that returns four products will lose to one that returns forty, no matter how well written it is.

Then weigh the term against your business rather than the search results. A term with modest demand for a product with healthy margin and a natural reorder cycle is worth more than a high demand term for a thin margin item you stock reluctantly. Keyword lists sorted by volume tend to send stores after the products they make the least money on.

Sequence what survives. Start with pages that already exist and already have impressions, because the fix is a rewrite rather than a launch. Then build the constrained pages where you have deep stock. Leave the broad category terms in the plan, since the constrained pages you build feed them with internal links and relevance, but do not measure the project on them in the first months.

Write the discards down and say why. Half the value of this step is being able to answer the question in six months about why you are not targeting an obvious word. Usually the answer is that you cannot fill the page, and that is a stock decision rather than a search decision.

A keyword group is not a deliverable. The deliverable is a short brief that tells whoever builds the page exactly what it must contain, and it should fit on one screen. Without it, keyword research ends in a spreadsheet nobody opens.

Every brief carries the same fields. The primary term and its variations. The page type and the URL. The products that must appear, and the rule that keeps them appearing as stock changes. The questions the page has to answer, taken from the support inbox rather than invented. The filters that should be visible. The internal links in and out, naming the specific pages. And one line describing what a shopper should be able to do within five seconds of landing.

The product selection rule is the part people skip. A collection built by hand looks great on launch day and is empty of relevant items three seasons later. Build it from a tag, an attribute or a smart rule so the grid maintains itself, and note in the brief what should happen when the rule returns too few items.

For a collection page, the brief should also say what copy is needed above and below the grid. Above the grid, a short paragraph that confirms the shopper is in the right place and names the attribute they were searching for. Below the grid, the buying detail: how to choose, what the sizes mean, what is worth paying more for. Neither section is there for a search engine. Both exist because a shopper who arrives from a constrained search has one specific question, and answering it is what stops the back button.

Hand the brief over with the search results page attached. The person writing needs to see what they are being asked to beat, and it is faster to show them than to describe it.

Some of your best terms only have demand for a few weeks, and they need to be planned backwards from that window. Gift terms, back to school terms, weather driven terms and event terms all follow the same pattern: nothing for months, a steep climb, a short peak, then nothing again. A page published during the peak has almost no chance. A page published well before it, with links pointing at it, does.

Build the seasonal list separately from the evergreen one, and put a date beside every term for when demand starts to move. You already have that data in your own order history and in Search Console year over year, so you do not need to guess. Work back from the start of the climb and set the build date early enough that the page has time to be found and to gather internal links.

Keep the same URL every year. A gift guide at a dated URL throws away everything it earned the previous season and starts from nothing. Use one permanent page per seasonal group and update the products, the copy and the dates inside it. This is the single biggest difference between stores that get seasonal search and stores that reinvent it annually.

Decide what the page does out of season. Leaving it live and unlinked is usually right, since it holds its history and can be brought forward again. Emptying it is not, because a page with no products looks abandoned and can lose the position it took a year to build. Where the products genuinely do not exist out of season, point the page at the nearest evergreen collection until stock returns.

Finally, look for the terms that are seasonal in your category but not obviously so. Replacement and consumable searches often cluster around the same weeks each year because the first purchase clustered there too, and that pattern is visible in your own reorder data.

Judge the project on pages, not on rankings. A ranking report tells you a position moved. It does not tell you whether the page brought people who bought, and in a store that is the only question that matters. Set the measurement up before the first page ships so you have a clean before.

For each page in the plan, track four things: impressions and clicks from search, the number of distinct queries the page appears for, add to cart rate from search traffic specifically, and orders assisted or completed. The query count is the underrated one. A collection page that starts appearing for many related searches, rather than only the one you targeted, is a page that search engines have understood, and that usually happens before the position on the primary term moves.

Give pages a realistic window. New pages in a competitive category take months, not weeks, and rewrites of pages that already had impressions move fastest. Judging either at four weeks produces the wrong decision, which is normally to abandon a page that was about to work.

Revisit the research on a schedule rather than when someone remembers. Twice a year is enough for most catalogues, plus a check whenever you add a product line, drop a supplier or change how you describe a category. Each revisit does the same three things: pull the new site search and Search Console queries, look for terms your pages now appear for that you never targeted, and retire groups where the products have gone.

Those accidental terms are the most useful output of any revisit. When a page ranks for something you did not plan, the search engine is telling you there is demand you did not know about, and it is usually cheaper to build the page it wants than to keep pushing the one you planned.