Most eCommerce keyword research goes wrong before a tool ever opens and it is because it starts from the store owner’s words instead of the buyer’s.
Someone names a collection “The Aurora Edit” in their Shopify admin, feels clever about it, and at that exact moment the customer who’d happily buy three of those lamps is typing “warm dimmable bedside lamp” into Google and clicking a competitor who simply called the page what the thing is.
Keyword research is the unglamorous work of closing the gap between how you talk about your catalog and how the world searches for it.
Get it right and search gets a chance to turn into your most profitable channel.
Get it wrong and you’ve built a store that sells to the handful of people who already know your name but stays invisible to everybody else.
And the trickiest part is… there’s no error message for “ranked for words nobody types,”. You only see a flat traffic line and a creeping suspicion that “SEO is dead” (for you).
You’ll, however, see that most guides around keyword research for eCommerce stores explain what search volume is, compare tool dashboards, then wave at “map your keywords to pages” as though that step were the obvious bit.
The mapping is the biggest part of the job. It decides which pages exist on your store, what each one targets, and how the whole site is organized, and it’s where this guide will spend most of its time.
This one’s for the people who have to live with these decisions so all of the store owners, and the marketers who need to create and rank these pages.
We’ll go the whole distance, from a blank spreadsheet to a keyword map you could hand straight to a developer to implement so you’ll know where real keyword ideas come from, how to read the metrics without getting hypnotized by the biggest number on the screen, how to decide which kind of page each keyword deserves, and how to tell months later whether any of it sold a single thing.
One idea runs under everything here: match what people want to the pages that can sell it to them.
If you’d rather watch than read, I recorded this entire process as part of my free Shopify SEO course on YouTube with theory first and then the live tool walkthrough on a real store.
The video is below; the article goes deeper in places, the video shows every click.
Why keyword research for a store isn’t the same job as for a blog
Start with what the work actually buys you.
Most marketing creates demand. For example, a TikTok video shows your product to people who didn’t know it existed.
Search runs the other way. The person typing “linen dress” already wants one, and keyword research is how you find out what they’re typing so your pages are standing there when they arrive.
That’s also why the research stage carries so much weight. It uncovers the demand, it tells you what to build and in what order, and it ends up defining your site’s entire architecture.
Everything else in SEO builds on proper keyword research.
For an eCommerce store, the keyword research guides all of the architecture and taxonomy decisions, as well.
When you settle that “linen shirts for men” is a keyword worth having, you’ve also decided a page should exist, what type of page it is, where it sits in your navigation, and what links to it.
This means a keyword choice becomes a structural commitment to your whole site. That’s a much heavier decision, and it’s why doing the research badly is so expensive to undo later.
Then there’s the language gap I mentioned earlier.
Owners most often describe inventory the way they bought it: by season, by supplier, by “the Henderson collection.”
Buyers organize everything by the problem they’re solving and the plain words they’d say out loud, and if you build your store around the owner’s mental model instead, every page ends up answering a question nobody asked.
And finally, the asymmetry that should not be missed – collection pages and product pages are playing two different games.
Collection pages (“running shoes,” “linen dresses”) target the category-level terms where most of the search volume lives, typed by buyers who know what they want but not yet which one.
Product pages convert the specific demand once the buyer has narrowed down.
Inside the collection game sits a second asymmetry, and it’s where most of the money is. Any store only has a handful of true head terms.
Sell linen clothing and your head terms will be “linen dress,” “linen pants,” “linen skirt,”etc. Maybe fifteen or twenty terms total, every one of them brutally competitive, because anyone selling linen clothing is trying to rank for these.
The modifiers, however, are where those few head terms expand nearly infinitely: “linen dress” becomes red linen dresses, long linen dresses, affordable linen dresses, linen dresses with pockets.
Ten or twenty sub-collection pages per head term, depending on what you actually stock, and you get to have hundreds of collection pages instead of dozens, which immediately gives you 10x to 20x odds of securing search traffic with high revenue potential.
Each modifier term also carries less volume and far less competition, which is exactly why a newer store can win them while the head term stays out of reach.
When SEOs say the money in eCommerce is in the modifiers, this is what they mean, and it’s why so much of this guide points at collection pages.
What you’re actually optimizing for (intent and money, not just volume)
Search volume is only a filter and should be viewed like that. It helps you by telling you a door exists, but it tells you nothing about whether anyone walks through it with a wallet.
A ranking that doesn’t sell is just a nicer screenshot you can brag about on LinkedIn, but it doesn’t do anything for your business.
That’s why rankings and revenue are two separate problems, and the gap between them is where a lot of SEO projects go to die.
So every keyword needs to be judged on three things at once.
- How much demand is behind it.
- How commercial the intent is, meaning how likely the searcher is to buy rather than browse.
- And how winnable it is for a store your size.
We’ll build that into a real prioritization system later, but you need the lens on from the very first seed you write down, because it changes which keywords you even bother to collect.
One underrated signal for the “will they buy” question is cost per click.
CPC is what advertisers are willing to pay Google for a click on that term, and advertisers are not really sentimental people.
When “linen dresses” shows a CPC around six dollars, that’s the market telling you the click has commercial value.
High CPC on a product term is an exact confirmation that money changes hands behind that search. It’s not a perfect proxy, of course, but it’s a free one, and it’s sitting right there in almost every KW research tool worth using.
And don’t write off the small and the weird, because the long tail is enormous and permanently renewing.
Google has said for years that 15% of the searches it sees every day are brand new, never typed by anyone before.
For a store, a huge share of real purchase intent lives in these specific, low-volume, four-and-five-word phrases that no keyword tool will ever show you a clean number for.
A search like “waterproof hiking boots for wide feet” might report ten searches a month on Ahrefs or none at all, and still be the exact thing someone types when they decide to buy.
Optimizing only for the fat head means ignoring most of the actual demand for what you sell.
Before we collect anything, though, you need raw material. Most of the good sources sit outside your keyword tool.
Building your KW seed list
You should always start in a spreadsheet. If your store already exists, your own data outranks everything else on this list, so start with sources two and six and work outward.
A brand-new store starts at source one and builds head terms.
Your own catalog is seed source number one, used at the right altitude.
A seed is a head term: the broadest honest name for a product line you can actually sell.
If you sell red linen dresses and nothing else, the seed is still “linen dress,” because the process needs to start broad and let the modifiers surface from data instead of from your assumptions.
Go one level broader than that, though, and the list turns to sludge, so seeding bare “linen” as a keyword pulls in around 22,000 keywords, including linen blazers and linen curtains you’ll never stock.
And you should always write seeds in the singular. Tools like Ahrefs match “linen dress” inside “linen dresses” automatically, since English plurals mostly just add an s, so the singular captures both while the plural drops half the list by accident.
From each head term, you should generate variations the way buyers modify it.
Across most stores the phrasings fall into a handful of predictable patterns:
- by type (“linen shirts”)
- by audience (“men’s linen shirts”)
- by attribute (“white linen shirts,” “oversized linen shirts”)
- by use case (“linen shirts for a wedding”)
- by problem (“non-see-through linen shirts”)
- by price (“affordable linen shirts,”)
- by brand
The second source is your buyers themselves (if you have an existing store, of course), and most stores never open the reports where they’re talking.
Support tickets, live chat logs, product reviews, and above all your on-site search box are pure, unfiltered buyer language.
When somebody lands on your store and types something into the search bar, they’re telling you exactly what they came to buy, in their own words, and sometimes telling you that you don’t have a page for it.
Seed source three is the marketplaces. A lot of product discovery doesn’t even start on Google anymore.
By Jungle Scout’s consumer research, around 56% of US shoppers start their product search on Amazon, ahead of the roughly 42% who start on a search engine.
So the Amazon search bar’s autocomplete, and Etsy’s if you’re in that world, is a firehose of exactly how buyers phrase their way toward products, ranked by what actually sells.
Type your category in and write down every suggestion.
Seed source four is Google’s own free surfaces, which are often better than people give them credit for.
Autocomplete, the “People also ask” boxes, the “related searches” at the bottom, and the little filter or refinement chips Google now generates at the top of many shopping-style results all tell you, for free, which variations and sub-categories Google itself believes searchers care about.
Those filter chips in particular are Google handing you its own opinion on how a category subdivides.
Seed source five is your competitors, treated as a shortcut worth making.
This is my favorite part of the whole process, honestly because a store that already ranks well in your niche has done most of your research for you and got it validated with Google.
Drop a few competitor domains into Ahrefs or Semrush and pull the keywords their category and product pages rank for.
You’re not copying their list blindly though, you’re harvesting the terms a store in your space has already proven, then keeping the ones that fit your catalog.
It’s the fastest way to find the gaps where they rank and you don’t.
And seed source six, if you already have any traffic at all, is your own Google Search Console.
The queries where you’re sitting on page two, positions 11 through 20, are gold, because they’re searches Google already associates with you and that you’re one good push away from winning.
We’ll come back to this in the measurement section, but it doubles as one of the richest seed sources you own.
A pile of seeds with no numbers attached is just a brainstorm though, so we want to move further from that.
To turn it into research you need data, which means tools, and the metrics that actually mean something on a store.
The tools, and the metrics that actually matter
I’m going to disappoint the people who came here for a tool shootout.
The tool matters far less than the method, and the endless “Ahrefs vs Semrush” debate is mostly two very good tools doing nearly identical jobs while you avoid the actual work.
Here’s the honest tiering. The free layer (Google Keyword Planner, the Google surfaces we just covered, and Search Console for your own data) is enough to start a small store.
The paid layer (Ahrefs and Semrush, primarily) buys you bigger keyword databases, competitor data, and the ability to do this at scale without losing your mind.
I use Ahrefs; I’m not affiliated with them, I just like the product, and everything in this guide works the same in Semrush.
What I’d steer you away from is the bargain tool somebody vibe-coded last quarter: the entire value of a keyword tool is the size and reliability of its database, and you have no idea where a cheap tool’s numbers came from.
If the subscription stings, pay for one month, run the whole keyword research process in this guide, export everything, and cancel.
The heavy research is front-loaded enough that a couple hundred dollars covers it.
What deserves more of your attention is how to read the numbers, because they mean different things on a store than they do on a blog.
Search volume first, with three cautions.
Google Keyword Planner reports volume in vague buckets (“1K-10K”) unless you’re spending on ads, which is why more precise numbers come from third-party tools.
Check whether you’re looking at local or global volume; a term doing 800 searches in the US might do 1,200 globally, and if you only ship to the US, the global figure is a fantasy.
And treat every volume as an estimate that runs low.
One client of mine ranks for about 300 keywords according to Ahrefs, while their Search Console shows over 50,000.
Not all 50,000 are worth acting on, but the gap tells you how much real search activity never reaches any tool’s database.
The tools are the best we have though, but they’re still guessing in many ways.
While you’re on the volume screen, glance at the trend graph too: “linen dress” swells every summer and sags by October, and an annual average hides exactly the kind of seasonality a store needs to plan around.
Keyword difficulty is the metric I’d tell you to mostly ignore, which surprises people, because it looks like precisely the thing you need.
Here’s what the score measures: an estimate of how hard the top ten is to crack, calculated mainly from how many backlinks the currently ranking pages have.
That logic breaks on eCommerce terms especially, because collection pages almost never earn backlinks naturally.
So the pages ranking for “linen dress” show few or zero referring links, the tool reads that as weakness, and it prints a difficulty of 10 out of 100 for a term every established linen brand in the country is fighting over.
A low KD on a category term doesn’t mean easy though.
It means the real difficulty lives somewhere the metric doesn’t look: domain strength, brand recognition, catalog depth, internal architecture etc.
What I do instead is read the live SERP like a scouting report, and I’d call this the single most useful habit in the whole keyword research process.
Search the term and look at who ranks.
You’ll recognize your actual competitors immediately, which is itself information.
Check their backlink profiles against yours (if you have one).
Check how many backlinks the individual ranking pages have (Ahrefs lists this per result in its SERP overview), and on collection terms you’ll mostly find zeros, which confirms those pages rank on their domain’s overall strength and internal linking rather than on links to the page itself.
A SERP of DR 70 brands is a different fight from a SERP with two DR 25 Shopify stores sitting in the top five, and no 0-to-100 score captures that difference.
Ten minutes of looking beats the number every time.
Two Ahrefs metrics that do earn their keep are Traffic Potential and Parent Topic, and the reason connects to a misconception I hear from business owners constantly: that one page ranks for one keyword.
One page ranks for hundreds.
In the linen niche, the number-one page for “linen dress” ranks for 290 different keywords by Ahrefs’s count, and given how much search activity the tools miss, the true number is probably several times that: the plural, “linen dress women,” “best linen dresses,” dozens of phrasings you’d never think to target separately.
Traffic Potential reports how much total traffic that number-one page pulls from everything it ranks for, which is a far better read on a keyword’s worth than its solo volume.
Parent Topic names the broadest term you can rank for while still capturing the smaller one.
Between them, they stop you from building three separate pages for “linen shirt,” “linen shirts,” and “mens linen shirt” when one page would take all three.
They’re clustering hints disguised as metrics.
Last, read the SERP features, because they tell you what the page even looks like before you commit.
For a product term you want to see shopping results, image packs, and refinement chips: Google confirming it treats the query as commercial.
A first page of how-to articles and videos is Google telling you the intent is informational, and no product page will rank there no matter how badly you want it to.
We’ll put that to work in the intent section. But first, the mechanical part. Turning seed terms into an actual list.
From seed terms to a raw list: the workflow
This is the part that’s faster to show than to describe, so if you skipped the video earlier, it runs every click of this on a live store.
Here’s the prose version.
Take your first head term into the tool’s matching terms report, which returns every keyword in the database containing your words.
The raw count will look unhinged: “linen dress” matches about 17,000 keywords.
The instinct is to start reading them one by one, deciding relevance keyword by keyword.
Don’t. You’d burn a week and quit. At this stage you’re collecting, not deciding, and the skill is entirely in the filters.
Four filters do most of the cutting.
Language: English, or whatever your market speaks.
Intent: commercial and transactional only, because right now you’re hunting collection pages, not blog ideas; the informational keywords get their turn at the mapping stage, where they’re routed to guides.
Branded: exclude, because “lulus linen dress” belongs to Lulus and you will never outrank a brand for its own name.
Local: exclude, because “linen dress shop near me” is a different game.
Those four filters take 17,000 keywords down to roughly a thousand honest candidates.
Export the CSV, move to the next head term, repeat.
Fifteen or twenty head terms later you’ll have something like nineteen files and twenty-thousand-plus keywords, which sounds like chaos and is exactly the goal.
Two filters to resist. Don’t set a volume floor because tools round small numbers down to nothing, a keyword showing zero volume usually still collects a trickle of monthly searches, and those specifics add up to real money across a catalog.
And don’t filter by keyword difficulty, for all the reasons above.
More data wins at this stage. Every extra phrasing you capture now is a page or title decision you get to make on evidence later.
Matching terms has a blind spot, though. It only returns keywords that contain your literal words, and buyers don’t search like that.
The person who wants a linen jumpsuit might type “linen romper” or “linen overalls,” and the matching report for “linen jumpsuit” will never surface either one.
That’s what the related terms report is for.
It finds keywords that share a SERP with yours rather than sharing words with yours, which is exactly where the synonyms live.
Run it for each head term and export again. You’ll collect plenty of duplicates across all these files, and that’s fine; they’re about to stop mattering.
Related terms is also where you catch benefit searches, my favorite category of keyword in eCommerce.
Some of your best future customers don’t know your product category exists.
The person searching “sheets for hot sleepers” needs linen sheets and has never once typed the word linen.
“Cooling sheets,” same story.
These queries describe the problem or the payoff instead of the product, their SERPs are usually a mixed bag of Reddit threads, listicles, and collection pages, and stores that catch them reach buyers their keyword-matching competitors never see.
This part is manual: browse the related and question reports for each product line with its benefits in mind. I
t’s some of the highest-yield browsing you’ll do.
Then round the list out from your competitors.
Take the five or six stores you kept recognizing in the SERPs, drop each into Site Explorer, and export the keywords they rank for; one mid-size linen brand yields about 7,000.
Apply the same four filters.
Most of it will duplicate what you already have, some will be irrelevant, and a slice will be phrasings and product angles you missed entirely.
The slice is the point.
Now the cleanup, and the tooling advice.
You’re holding twenty-odd CSVs full of overlap, and the wrong move is merging them by hand in Google Sheets.
The other wrong move is pasting them into a chatbot and asking it to merge them directly, because a language model freestyling your data will drop rows and invent others.
Ask the LLM to write you a Python script instead. A script is deterministic, you can rerun it on every future export, and the same conversation can bolt on your own rules: dedupe everything, drop keywords below whatever floor you care about, drop keywords mentioning products you don’t stock.
Tell it what you sell and it’ll figure out that “linen blazer” doesn’t belong in your list. Five minutes of prompting, one master sheet of unique keywords.
And keep in mind that the list is never finished.
Once the store is live, real query data starts arriving in the Search Console and shows you what you missed, and every new product line reopens the research. You’ll revisit this map, trust me.
What you have now is a raw list. 20,000 keywords, deduplicated, zero decisions made though.
Everything from here is deciding, and the first decision is reading what each searcher actually wants.
Search intent and the eCommerce buyer journey
Intent is the question behind the keyword. Get it wrong and you can rank perfectly while selling nothing, because you answered a question the searcher wasn’t asking.
The standard model splits search intent into four buckets, and they map onto a shopper’s journey cleanly.
Informational (“how to clean cast iron”) is someone learning, usually early and not ready to buy.
Commercial (“best cast iron skillet,” “lodge vs le creuset”) is someone comparing, circling the purchase.
Transactional (“buy cast iron skillet,” “lodge skillet free shipping”) is someone with a card out.
Navigational (“lodge cookware”) is someone looking for a specific brand.
Marketers sometimes relabel these as top, middle, and bottom of the funnel, which is the same idea wearing a sales hat.
The words people add to a query give the intent away most of the time.
“Best,” “top,” “review,” “vs,” and “alternative” are commercial tells, the language of comparison.
“Buy,” “for sale,” “price,” “discount,” “coupon,” and “free shipping” are transactional tells, the language of someone about to spend.
Notice these as you build your list, because they sort your keywords into page types almost automatically.
The intent of a keyword is whatever Google is already rewarding for it, so make sure you remember that.
Google has billions of behavioral data points and has spent twenty years tuning results to what searchers actually want, so the live results page is the closest thing to ground truth you’ll get.
Search your term, ideally in an incognito window or through a VPN so your own history doesn’t color it, and look at what ranks.
If page one is full of category pages from other stores, the intent is commercial and you need a collection page.
If it’s all blog posts and guides, the intent is informational and your product page is never going to rank there, no matter how badly you want it to.
This is also why Google’s own guidance is so insistent that you create content that satisfies the searcher’s intent rather than just stuffing in keywords.
The algorithm’s whole job is matching a query to the page that resolves it. Your job in keyword research is to read that match correctly before you build anything.
A lot of valuable terms are mixed-intent, and the SERP will show it by mixing page types.
“Standing desk” might return a few collection pages and a couple of buying guides, which is Google telling you it’s not sure whether you want to buy or learn.
Those terms usually mean you want both a strong collection page and a supporting guide that links to it.
Knowing which intent dominates, and whether it’s split, is exactly what tells you the page type.
Time to make that decision explicit.
Mapping each keyword to the right page type
Mapping is the heart of eCommerce keyword research.
Every keyword you keep gets assigned a home, and the home is a page type, not just a page.
The decision logic is more mechanical than people expect.
A broad, commercial, category-level term (“running shoes,” “linen dresses”) becomes a collection page, because that’s the page type built to capture browse-and-compare demand.
A specific, transactional term tied to one item or model (“Brooks Ghost 16 men’s”) becomes a product page.
An informational query (“how to choose running shoes,” “are linen dresses good for summer”) becomes a blog post or a buying guide.
A high-demand attribute combination (“waterproof running shoes,” “black linen dresses”) becomes a subcategory, if and only if there’s real search demand behind it.
And a huge number of your seeds become no page at all, because they’re duplicates of an intent you’ve already covered, or merchandising ideas with no search behind them.
Here’s the mapping in one view:
| Keyword pattern | Dominant intent | Page type it becomes | Example |
| Broad category term | Commercial | Collection / category page | “linen dresses” |
| Category + attribute (with demand) | Commercial | Subcategory or indexed filter | “black linen dresses” |
| Specific product / model | Transactional | Product page | “Reformation Gwyneth linen dress” |
| “Best / top / review / vs” | Commercial investigation | Buying guide / comparison post | “best linen dresses for summer” |
| “How / what / why / can” | Informational | Blog post / guide | “how to wash a linen dress” |
| Brand term | Navigational | Homepage / brand collection | “Reformation linen” |
| No volume, no distinct intent | — | No page (merchandise only, or fold in) | “Spring ’26 drop” |
Sit with that last row.
The instinct on a store is always to make another page, and it’s often wrong.
A page earns its URL by owning a distinct, searched-for intent, and a collection with no search demand behind it isn’t an SEO asset, it’s a maintenance liability with a URL.
The sooner you can say “this is a lovely merchandising page but it is not a keyword target,” the healthier your store stays.
Collections deserve special attention because they’re where the volume and the durability both concentrate, and they’re their own discipline once you get into titles, descriptions, URL structure, and the Shopify-specific traps.
I’ve written a separate, equally obsessive walkthrough of collection page SEO that picks up exactly where this mapping leaves off, so I won’t re-run all of it here.
For our purposes, hold the principle that the keyword research decides a collection should exist and what it targets, and the collection setup executes on that decision.
So mapping tells you which page each keyword belongs to.
The next question is which keywords share a page.
From a list to a map: clustering, SERP similarity, and killing cannibalization
A keyword list becomes a keyword map the moment you group variants that mean the same thing onto a single page.
Skip this and you build five thin pages that split your own strength five ways.
Most of your list is near-duplicates.
“Ripped jeans,” “distressed jeans,” and “torn jeans” are very likely one intent that shares the same search results and therefore one page and not three.
The test for whether two keywords belong together is simple. Search both, in incognito, and compare the results.
If Google returns substantially the same pages for both queries, Google considers them the same intent, and you should build one page and let it rank for the whole cluster.
If the results differ meaningfully, you’ve found two distinct intents and you need two pages.
SEOs call the overlap a SERP similarity score, and many tools do this process very well.
Clustering is also the one part of this process I’d automate completely.
The failure mode all of this prevents is keyword cannibalization: two or more of your own pages competing for the same query, so Google can’t decide which to rank and tends to rank neither of them well, or flip-flops between them and rotates you out of the results entirely.
So you’ve now got clusters, each mapped to one page of a known type.
The last question before you build anything is the most important one for your time and sanity: which of these do you do first?
Prioritizing by money and not volume
Prioritization is where “money over volume” turns into a formula.
You rank each mapped cluster by three things multiplied together: search demand, commercial intent, and winnability.
Demand is the volume, the size of the door.
Intent is how likely that traffic is to buy, which you already estimated from the keyword’s wording, its CPC, and the page types ranking for it.
Winnability is how realistic a top-three finish is for a site your size, and you read it from the SERP scouting habit: who ranks, their authority and their pages’ backlinks against yours, plus where you currently rank, if you rank at all.
Multiply, sort, and you get a list ordered by likely revenue rather than by vanity.
Run two keywords through that and the point makes itself.
A category sitting at position 6 for a 900-searches-a-month term that converts at 3% will, in almost every case, beat a category sitting at position 40 for a 10,000-searches term you have no realistic shot at this year.
This is why the keywords I get most excited about for a client are usually the ones already sitting in positions 4 through 15.
They’ve got demonstrated demand, Google already considers the page relevant, and they’re close enough to the top to taste it but not yet earning what they should.
And based on experience, I can tell you that the realistic path for most stores runs through the specific, the long-tail, and the commercially intense modifier terms the giants are too big and too bored to optimize for individually.
Picking those battles deliberately is most of the strategy.
Once you’ve sorted by money, you write it all down somewhere you’ll actually use.
The keyword map: the thing you actually ship
Everything up to now produces one deliverable, and it’s boring on purpose: a single spreadsheet, one row per keyword cluster, that turns research into a to-do list anyone on your team can execute.
A workable keyword map has a column for the cluster’s primary keyword, its volume, its intent, the page type it maps to, the target URL (existing or to-be-created), the current ranking position if any, and a priority rank.
There’s no difficulty-score column, on purpose. The winnability judgment came from the SERP check, and a 0-to-100 number pasted next to it would only invite sorting by the wrong thing.
Here’s a small slice of what one looks like for, say, a cookware store:
| Primary keyword | Vol | Intent | Page type | Target URL | Now | Priority |
| carbon steel pans | 2,400 | Commercial | Collection | /collections/carbon-steel (exists) | 7 | 1 |
| cast iron skillet | 6,600 | Commercial | Collection | /collections/cast-iron (exists) | 12 | 2 |
| best pans for gas stove | 900 | Commercial inv. | Buying guide | /blogs/guide-gas-stove-pans (new) | — | 3 |
| how to season carbon steel | 1,300 | Informational | Guide → links to collection | /blogs/season-carbon-steel (new) | — | 4 |
| Lodge 12 inch skillet | 480 | Transactional | Product | /products/lodge-12-skillet (exists) | 5 | 5 |
| Spring ’26 cookware drop | — | None | No SEO page (merch only) | — | — | — |
Keyword research in the age of AI search
AI is changing how results get clicked but not whether people search for products.
The cleanest public read we have is Pew Research Center’s 2025 analysis of real browsing data, and it found that when an AI summary appears, the click-through rate to regular links falls from about 15% to about 8%, with users clicking a link inside the AI summary only around 1% of the time.
That’s a meaningful dent in clicks for informational queries where an AI Overview can just answer the question outright.
But AI Overviews dominate informational queries (the “how do I season cast iron” stuff), and they’re far less present and far less satisfying on transactional and commercial queries, because nobody wants an essay when they’re trying to buy a skillet. They want the skillet.
The commercial core of your catalog, the collection and product terms that actually drive revenue, is the part of search least eaten by AI summaries.
The research process doesn’t change either because you map demand to pages exactly as before.
What changes is that you weigh your informational content with clearer eyes about what it’ll return and you might deprioritize them in your SEO strategy.
One thing that’s changing is that queries are getting longer and more conversational as people talk to AI search the way they’d talk to a person.
“Good waterproof hiking boots for someone with wide feet who hikes in Scotland” is a real shape of query now, and it lands squarely on the intent-rich, long-tail, benefit-flavored terms this whole guide has been steering you toward.
So what should you actually do?
If you skimmed all of this (no judgment, it was a lot), the process compresses to six moves:
- Build a seed list from your buyers, your catalog, the marketplaces, and Google’s free surfaces.
- Pull the raw list with the matching-and-related-terms workflow and real numbers from a real tool.
- Read each keyword’s intent off the live results page.
- Map every keyword you keep to one page type, and refuse to build pages for intents you’ve already covered.
- Cluster the variants, kill the cannibalization, and prioritize by money instead of volume.
If you’d rather have this done than read about it, or you’ve built a keyword map and want a second set of eyes before you commit six months of work to it, grab a free call and we’ll look at your store together.
I’ll have the live results and your Search Console open while we talk, which beats any slide deck I could make you.
And if you’re weighing whether to do this in-house or bring someone in, I’ve written separately and honestly about what eCommerce SEO actually costs so you can tell whether a quote you’ve been given is fair.
Either way, do the mapping. Every other SEO decision on your store inherits from it.