Distrify

Keyword Research for an Online Store: Buying Intent First

August 21, 2026·14 min read
A man in his early thirties in a linen shirt writes customer phrases in a notebook next to his laptop and two parcels, in the wooden back room of a small shop lit by the storefront window

A query with twenty thousand monthly searches where you sit in ninetieth place brings you nothing. A query with forty monthly searches where you come out on top, typed by someone looking for exactly what you stock, brings you orders.

Keyword research for an online store is almost always done backwards: open a tool, sort by volume, collect a list of generic category terms that marketplaces have owned for a decade, then write pages that will never surface. The method below starts from the other end, the buyer's end. What does this person actually want, can you answer better than anyone else, and which single page is going to receive them.

TL;DR

  • Volume measures curiosity, not purchase. On a young store, a precise low demand query beats a generic term you cannot reach.
  • Four intentions hide behind queries: discover, compare, buy, return. Each one calls for a different type of page.
  • Your best keyword source is not a tool. It is your catalogue, your on site search box and your customer inbox.
  • One intent, one page. Three pages aimed at the same intent compete with each other instead of bringing you buyers.
  • Collecting and grouping can be automated. Deciding that a query is not worth it cannot: that needs your margins.

Table of contents

Keyword research for an online store: why volume misleads

Volume is the one number a tool can hand you instantly, so it becomes the default criterion. For a store that is still building an audience, it is a poor one, for three reasons.

First, high volume terms are category terms. Marketplaces, comparison sites and established brands hold them, with years of head start and thousands of inbound links behind them. A better product page does not take that spot.

Second, volume says nothing about intent. Behind a generic term you find browsers, students, competitors, people looking for a manual, and occasionally a buyer. Behind a long specific query there is nearly always someone who has already decided to buy and is now looking for the right version of the thing.

Third, the number you see is a rounded estimate, usually averaged over a year, and it ignores your seasonality entirely. An outdoor product averaging forty searches a month can be at three hundred in June and zero in December.

So the useful question is not how many people type this phrase. It is what does this person want, and do I sell it.

The four intentions behind a query

Every query falls into one of four families. This is the most profitable lens in the whole exercise, because it immediately tells you which page should answer.

Intention What the person wants Typical phrasing The page that answers What it is worth to you
Discover Understand a subject, form an opinion how to choose, why, what is the difference between A guide or an article A buyer several weeks from now
Compare Decide between two or three options which one for, best for, pros and limits A collection page or a buying guide A buyer within days
Buy Order now with a size, a material, a colour, next day delivery A product page or a filtered collection An order today
Return Come back to a site already known your brand name, your name plus a product The home page or the relevant page A customer you already earned

Two practical consequences. A discovery query is not served by a product page: the person is not ready and leaves. And a buying query is not served by a blog article: they want a button, not eight paragraphs.

The most common mistake on a small store is to treat everything as a category page, with pages that try to serve all four intentions at once and satisfy none of them.

Start with your catalogue, not with a tool

Open your catalogue and take your twenty best sellers. For each one, write five lines in plain language.

  1. What it is, said the way a customer would say it, not the way your supplier writes it.
  2. Who it is for, with age, skill level and context of use.
  3. The problem it solves, phrased as a complaint.
  4. The constraints that cause hesitation: size, material, compatibility, budget, delivery time.
  5. The occasion that triggers the purchase: a gift, a move, a new school year, a breakdown.

That is a hundred lines in an hour, with no tool involved, and every line is a candidate query. These phrases have a quality no purchased list will ever have: they describe your actual stock, with your variants and your constraints. They are exactly the queries nobody else bothered to serve, because they never show up in a volume ranking.

This is also when the gaps appear. If one constraint comes back across five products and has no page anywhere on your site, you have just found your next collection page.

Your customers do not use your words

After six months in a trade you adopt the language of the trade. Your customer never did. They are not looking for an ultrasonic diffuser, they are looking for something that makes the living room smell good without a flame. They are not looking for a part reference, they are looking for the piece that fits their model.

That vocabulary gap costs you sales quietly, because your pages are written in a language your buyers do not type. Four places give you their exact wording, for free.

  • Your on site search box. These are people already on your site who cannot find what they came for. Every zero result query is a missing page, a missing name or a missing product.
  • Your inbox and your customer service. Pre purchase questions are queries written out in full, with the doubt attached.
  • Customer reviews, on your products and on comparable ones. People explain there why they bought, in their own words.
  • Questions asked over the phone or in person if you also have a physical outlet. It is the richest source and the least used.

Reuse those phrases as they are. Polishing your customers' language to make it sound more professional is the surest way to disappear from their searches.

Three free sources when you own no tool

You do not need a subscription to start. Three sources take you from a hundred lines to two hundred.

Autocomplete and related searches. Type the beginning of your phrases and note what completes. Those are queries humans actually typed, which beats an estimate. Then scroll to the bottom of the results page and collect the related searches.

Search Console, as soon as you have impressions. It is the most honest source in existence, because it only ever reports queries you already appear on. Queries where you are shown but never clicked tell you a need exists and your page answers the wrong intent. On a new site it takes a few weeks before this becomes readable, and until then the other two sources carry the work. How to read that data without fooling yourself is covered in our monthly AI visibility tracking routine.

Your own sales data. Order line names, the filters people use most, the products often bought together. A frequent pairing with no dedicated page is a query waiting for you.

A paid volume tool becomes genuinely useful later, when your list is long and you have to arbitrate between serious candidates. It will still never tell you whether a buyer sits behind the phrase. That part stays yours.

Sorting two hundred queries in an hour

A two hundred line list is worthless until it is sorted. Ask four questions of every line, in this order, and stop at the first no.

  1. Can someone typing this buy from me today? If not, it is not a bad subject, it is an audience subject: second pile, not the first.
  2. Do I already have a page that answers this exact intent? If yes, the job is not to write, it is to improve that page and link to it.
  3. Can I answer better than what currently ranks? Better means more precise, more honest, better illustrated, with your photos and your real constraints. If your only argument is writing longer, skip it.
  4. Will I be able to tell in a month whether it worked? If you do not know which number you would look at, the line is not ready.

You end with three piles: do now, keep for later, never do. The third pile is where the time is saved, provided you accept it.

Twenty to thirty well chosen queries keep a small store busy for six months. The rest is noise that feels like progress.

You need no tool and no budget to run that sort. Join the waitlist to get the automated version of this loop when it opens, but the grid above is filled in by hand once and serves you all year.

One intent, one page: the map that prevents cannibalization

Take the queries you kept and build a three column table: the query, its intention, the exact address of the page that receives it. An intention may appear only once in the right hand column.

That rule is what protects you from cannibalization. When three pages on your site aim at the same intent, the engine picks one, often not the one you would have picked, and the three of them split signals that should have gone to a single page. You worked three times for a third of the result.

The split is simple to hold on a store:

  • Product page for the precise buying intent, with the variant, size or compatibility in the title.
  • Collection page for the broad commercial intent, the one that compares several options for the same need.
  • Guide or article for the discovery intent, which then points to the collection and to two product pages.

Link the three levels together. The guide sends to the collection, the collection to the products, and the products back to the guide for the undecided. That internal linking is what keeps deep pages alive, and it follows the same logic as our content distribution strategy.

Keeping the map current takes ten minutes per new page. Not keeping it costs months of scattered work, and publishing faster does not help: shipping more pages against a wrong map only scatters faster, which is why we argue for a publishing rhythm you can hold.

What assistants change about this list

More buyers now ask an assistant before they open a search engine. The list you just built does not become obsolete, but its shape shifts.

Questions asked to an assistant are longer, written as full sentences, and often comparative. Nobody types three words at an assistant: they describe a situation, a budget and a constraint, then ask what to get. So your comparison queries and your constraint queries gain value, while bare category terms lose some.

In practice, keep the same list and change how each page opens. A page that answers in two sentences before expanding gets picked up by an assistant. A page that takes three paragraphs to reach the point does not. The full reasoning sits in our guide on AI search optimization for ecommerce, and the underlying mechanics in the generative engine optimization guide.

Two honest caveats. Nobody can guarantee that an assistant will cite your store: you work on what makes a brand citable, you never decree a citation, and what actually works to get mentioned comes down to a few fundamentals. And the split between assistants and the classic engine depends on your market, so it is measured in your own data rather than in a general study, as we set out in ChatGPT vs Google Search.

What can be automated here, and what cannot

Worth saying plainly, because the opposite promise sells very well.

Automates without trouble: collecting suggestions, removing duplicates, grouping the variants of one intention, tracking positions over time, drafting a first version, publishing and distributing what you publish. On those tasks a machine is steadier and cheaper than any person.

Does not automate: knowing that a word means something different in your shop than elsewhere, deciding that a query profitable for a competitor is not profitable for you because your margin is thinner, recognising the exact phrasing of a loyal customer, and choosing between two pages when you only have time for one. Those calls require knowing your numbers, and your numbers are nowhere on the internet.

Where the line falls exactly is detailed in our piece on SEO automation, and the same divide applies to measurement: tracking a position is mechanical, interpreting a share of voice is not.

Where to start this week

Three actions, in this order, with no budget and no new tool.

  1. Write the five lines for your twenty best sellers. One hour of work, and you hold material nobody else has.
  2. Export the last three months of your on site search queries and highlight the ones that returned nothing. Each is either a missing page, a missing product or a wrong word.
  3. Fill the three column map for your first ten queries and check that no intention appears twice. If one does, you just found your first cannibalization, and fixing it beats publishing one more page.

Everything after that, the writing, the publishing and the distribution to other channels, is repetitive work that can be handed off.

FAQ

How many keywords does a small store need? Twenty to thirty well chosen ones cover six months. A thousand line list is not a strategy, it is a spreadsheet nobody will ever open.

Do I need a paid tool to start? No. Your catalogue, your on site search, your customer messages and autocomplete cover the first months comfortably. A paid tool earns its place when you have to arbitrate between dozens of serious candidates.

What if my queries show no volume at all? Serve them anyway when the buying intent is clear. A zero often means too little data to measure, not absence of demand. On a niche store, those are the queries that convert.

Should I target competing stores' keywords? Look at what they cover to spot intentions you forgot, never to copy their list. Their catalogue, their margins and their age are not yours, so their priorities are not either.

How long before this shows an effect? On precise, lightly contested queries, a few weeks to appear and a few months to settle. On category terms, count in years, or drop them and spend the time elsewhere.

Conclusion

Keyword research for an online store is not about finding heavily searched terms. It is about recognising the phrases with a buyer behind them, then giving each one a page of its own. It is shopkeeper work more than SEO work: you know your stock and your customers better than any tool ever will.

Once the map is set, what remains is repetitive: write, publish, adapt for the other channels, measure, repeat. Distrify is built to run that loop for you across all three channels at once. The product is not open yet, so Join the waitlist is currently the only way to hear about it first.