Tecof • September 15, 2026
How to Do Keyword Research (E-commerce Focused)
In Brief
Keyword research is the work of finding the words people actually type when they look for a product, and matching those words to the right page on your site. In e-commerce it is not simply producing a list; every word on that list has to declare the intent behind it, which page type satisfies that intent, and whether such a page already exists. Sorting keywords by volume from highest to lowest is easy; the hard part is separating the ones that will bring you orders. As of 2026 the question is no longer “how many people search this term a month” but “what does the person searching this term want to do, and on which page”.
Tuesday morning, 09.20. At a home textiles store with 1,400 products, a keyword file prepared six months ago is opened. It holds 812 keywords, all sorted by search volume, with “duvet cover set” at 74,000 monthly searches on top. Search Console shows 2,640 queries that brought clicks to the site in the last three months; 2,180 of those queries do not appear in the file at all. Of the 812 keywords in the file, 511 have no target page on the site, and 96 have two separate pages competing for them at once.
The problem is not that research was never done, but that it was done in the wrong unit. The output of keyword research is not a keyword list, it is a keyword-to-URL map. A list leads you to count words; a map leads you to create pages, merge them and delete them. The sections below explain how that map is built.
Search Intent: the One Distinction That Makes a List Useful
No list works until each keyword is classified by the intent behind it rather than by its volume. Within the same product family there are four different intents, and each one calls for a different page type. Read the intent wrong and even a perfectly written page will not rank, because Google prefers to show a different kind of page for that query.
Four intents, four page types
- Informational intent: the user wants to learn. “what is percale”, “what temperature to wash a duvet cover”. This calls for a blog post or a guide; putting a product page against this query almost never works.
- Comparison intent: the user is stuck between two options. “percale or sateen”, “difference between cotton and bamboo bedding”. This calls for comparison content or a rich category description.
- Commercial intent: the user is close to buying but still choosing. “double duvet cover set”, “cotton duvet cover prices”. This calls for a category or filtered listing page.
- Transactional intent: the user wants one specific thing. “Brand X 200x220 percale duvet set”, searching by product code. This calls for a single product page.
In Turkey there is a practical shortcut for reading this distinction: the word appended to the query gives the intent away. “fiyat” (price), “kaç TL” (how many lira), “indirim” (discount) and “ucuz” (cheap) point to the commercial-transactional side; “nedir” (what is), “nasıl” (how) and “ne işe yarar” (what is it for) point to the informational side; while “yorum” (review), “şikayet” (complaint) and “gerçekten işe yarıyor mu” (does it really work) mark the last hesitation before purchase. Review queries are ignored entirely at most stores, yet they carry the traffic closest to conversion.
The cheapest way to verify intent
Do not guess, look. Search the term in a private window and note the page type of the first ten results. If eight of the ten are blog posts, that keyword is informational and trying to enter it with a category page will cost you months. If eight are category pages, you enter with your category page, not your product page. This check takes forty seconds per keyword and sets your priorities for the next three months.
| Intent | Example query | Target page type | Expected conversion | Priority |
|---|---|---|---|---|
| Informational | what is percale | Blog / guide | Very low | Brand awareness |
| Comparison | percale or sateen | Blog / category copy | Low to medium | Medium |
| Commercial | double duvet cover set prices | Category / filtered list | Medium to high | Highest |
| Transactional | 200x220 percale duvet cover set | Product page | High | High |
What Volume and Difficulty Metrics Actually Measure
The two numbers on the tool's screen are the two most misread numbers in the field. Both are useful, and neither measures what you think it measures.
Search volume is an estimate, not a promise
The monthly volume Google Keyword Planner reports is a twelve-month average, and it pools similar terms into a single bucket. “nevresim takımı” and “nevresim takimi” land in the same bucket, even though user behaviour behind the two may differ. Worse, if your ad account has no active campaign, Planner shows you a range rather than a figure: 10K-100K, for instance. You cannot prioritise on a range like that. The way to narrow it is to run a small-budget search campaign for a few weeks; that spend costs less than building a content plan of several thousand lira on the wrong assumption.
The second trap is treating volume as clicks. Ranking first for a term with 74,000 monthly searches does not bring you 74,000 visits. If the results page carries four ads above, a shopping carousel below and marketplace listings alongside, the click share of the first organic position can fall under 10 percent. Do not look at volume without taking a screenshot of the result page.
Difficulty is calculated for an average site, not for yours
Tools' keyword difficulty scores mostly look at the number of links pointing at the top ten results. They do not account for your site's existing authority on that topic, the size of your brand search, or the product depth of your category page. In practice this correction helps: compare the score against the scores of keywords you already rank for on the same topic. If you are on page one for keywords at difficulty 25 today, a keyword at 45 is reachable; one at 70 is not for this year. We covered how the link side actually works in our piece on backlinks and earning links.
The long tail: searched rarely, sold often
The long tail is queries longer than three words with low monthly volume. Individually they look trivial; together they carry more than half of most stores' organic traffic. In one store's Search Console records, 2,100 of 2,640 queries over three months receive fewer than five searches a month, yet those 2,100 queries deliver 58 percent of total organic clicks. The second advantage of the long tail is conversion: someone searching “duvet cover set” is still browsing, while someone searching “200x220 white cotton duvet cover set” has already picked the size, the colour and the material.
What Makes Turkish Search Different
Keyword methods imported from English-language sources work only partly in Turkish, because the structure of the language differs. The three points below are where stores operating in Turkey lose the most keywords.
Suffixes: ten forms of the same word
Turkish is an agglutinative language. “Nevresim”, “nevresimler”, “nevresimi”, “nevresimlik”, “nevresim takımı” and “nevresim takımları” are separate queries and tools usually show them on separate rows. Google largely associates these suffixes with the same root, so opening a separate page for each variant produces cannibalisation. The right approach is to target the root term and let the suffixed variants appear naturally inside the same page. Only split into a separate page when the suffix changes the meaning: “nevresim” and “nevresimlik kumaş” (duvet fabric) are different products, “nevresim” and “nevresimler” are not.
ı/i and ş/s: the user who types without Turkish characters
A share of users search without Turkish characters: “nevresim takimi”, “çamasir makinesi”, “islik” instead of “ıslık”. Google usually corrects these variants, but two places do not: your own site search box and your URL structure. If on-site search has no Turkish character normalisation, a customer typing “çamasir” sees zero results and leaves. On the URL side the rule is clear: always write slugs without Turkish characters and with hyphens, “camasir-makinesi” rather than “çamaşır-makinesi”.
Brand, model and attribute combinations
A significant share of shopping searches in Turkey begins with a brand name and ends with a product attribute: “brand + product + size”, “brand + model + review”, “brand + product + how many lira”. These combinations look low-volume but carry the highest purchase intent of any query. For combinations carrying your own brand, your product page is already the candidate; for those carrying someone else's brand, compete only if you actually sell that brand. Content targeting a brand name you do not stock neither ranks nor earns trust.
Where Keywords Come From: Four Sources
Use the data you already hold before you buy a tool. The first two of the four sources below are free and at most stores have never been opened.
Mining Search Console queries
The most valuable source is the queries your site already appears for. Open the Performance report in Search Console, set the range to the last 3 months, and apply these three filters in turn. First: queries with over 100 impressions and under 5 clicks. These are queries where Google shows you and the user does not click; the problem is usually in the title and description, and the fix is editing the title tag and meta description. Second: queries with an average position between 8 and 20. These sit on the edge of page two and can be lifted to page one by deepening the content, which is far cheaper than creating a page from scratch. Third: queries that get impressions but have no matching page on your site. These are your new-page ideas, and they arrive from real demand rather than guesswork.
Google Keyword Planner and autocomplete
Use Planner for discovery, not for volume. Put your own category page URL into the “start with a website” field; Google tells you which terms it associates with that page. This is the most direct way to see how your own page is understood. The second free discovery channel is the search box itself: typing the term letter by letter, noting the “related searches” block and the “people also ask” box, produces a more current list than most paid tools give you.
On-site search and customer messages
What visitors type into your site search box is their own vocabulary and appears in no tool. Export the searches returning zero results every week: that list gives you both your missing products and your missing keywords. The same holds for questions reaching your support agents and your WhatsApp line; “how much weight does this hold” is not an FAQ item, it is a long-tail query.
Competitor gap analysis
Gap analysis means finding the keywords your competitors rank for and you do not appear for at all. Pick three to five competitors, and pick them carefully: sites with the same product depth, the same price band and the same country. Putting Trendyol and Hepsiburada on the competitor list breaks the analysis, because those sites rank for hundreds of thousands of terms and your gap list becomes unusable. Use marketplaces as a benchmark rather than as competitors: the more of the top ten results for your target term are marketplace pages, the more limited your organic share on that term is. Whether you then defend that keyword with ads or with content is a separate decision, and we worked through that balance in our SEO versus advertising comparison.
| Source | What it gives | Strength | Blind spot |
|---|---|---|---|
| Search Console | Real queries, position, clicks | Specific to your site, free | Shows nothing you never appear for |
| Keyword Planner | Volume ranges, related terms | Google's own associations | Volumes are pooled and banded |
| On-site search | The visitor's own vocabulary | Zero-result rows are direct actions | Low volume, narrow sample |
| Competitor gap | Terms you do not have | Fast idea generation | Wrong competitor choice ruins the list |
The Keyword-to-URL Map, Cannibalisation and Priority
This is the output of the research. The map is a table writing a single URL against every target keyword, and the reverse holds too: every URL has exactly one primary keyword.
How category, product and blog divide the work
A simple division rule works. Plural, general and commercial terms belong to the category page: “cotton duvet cover sets”. Singular, specific and transactional terms belong to the product page: combinations carrying brand, model, size and colour. Question-shaped, informational terms belong to the blog. Holding that distinction while writing product descriptions is hard; we explained how to hold it at scale in the piece on writing product descriptions with AI.
Spotting cannibalisation
Cannibalisation is two of your pages competing for the same keyword. The symptom is this: when you filter a query in Search Console, the “Pages” tab shows two or more URLs and their positions swap places over the following weeks. The result is that neither reaches page one. In Turkish the most common causes are opening separate pages for suffix variants and placing a blog post on the same topic underneath the category page.
The fix, in order: declare whichever of the two pages gets more clicks the primary one, move the original parts of the other page's content into it, 301-redirect the second to the first, and consolidate internal links onto a single target. Before redirecting, check the second page's conversion data; the page with fewer clicks may be making more sales.
| Page type | Keyword pattern | Example | Keywords per page |
|---|---|---|---|
| Category | Plural + attribute | cotton duvet cover sets | 1 primary + 3-5 secondary |
| Filtered list | Attribute + size/colour | 200x220 white duvet cover | 1 primary |
| Product | Brand + model + attribute | brand percale duvet cover 200x220 | 1 primary + variant names |
| Blog | Question / comparison | percale or sateen | 1 primary + 5-8 long tail |
Tie seasonality to a calendar
When you target a keyword matters no less than which keyword you target. Content takes eight to twelve weeks on average to rank, so the calendar is built backwards.
Retail demand in Turkey clusters on particular dates: back to school in mid-August, November discount searches starting in late October, new year gift searches peaking in the first week of December, and Ramadan and holiday dates shifting each year. Publish your November campaign content in November and you have missed the race; it has to go out in August. Keyword Planner's monthly distribution chart confirms these dates for your own category and removes the need to guess.
A priority score in four columns
Instead of sorting keywords by volume, score them on four criteria: size of demand, how close the intent is to purchase, your current position, and whether the page exists. A commercial-intent keyword where you already sit at position 11 with a page in place comes before a keyword with three times the volume where you do not rank at all. That ordering gives you a first result in three weeks rather than three months.
| Criterion | Scores high | Scores low | Weight |
|---|---|---|---|
| Intent | Commercial / transactional | Informational | High |
| Current position | Between 8 and 20 | Not ranking at all | High |
| Page status | Page exists but is weak | No page | Medium |
| Demand | Above the category average | Under 10 a month | Medium |
What AI Search Changed and What It Did Not
AI overviews and chat-based search did not end keyword work, but they moved its centre of gravity.
Queries are getting longer, not shorter
Writing to an assistant, users type full sentences rather than two-word queries: “can you suggest bedding for a double bed that does not make you sweat in summer”. That means the long tail is getting longer still. The practical consequence: your pages need sections that ask the question plainly and answer it clearly in two or three sentences. When a model quotes a page it looks for the definition and the figure, not the marketing sentence. We went into how this field works in our article on AEO and GEO.
Three things that have not changed
First, the accuracy of product data: if stock, price, size and variant information are not correct in your structured data, no assistant will recommend you. Second, category architecture: a deep, consistent category tree is still the strongest signal for both users and crawlers. Third, page speed and technical accessibility. Keyword work is built on top of those three; if they are missing, no list however good will produce results. On a setup where catalog, content and structured data sit on the same platform, applying the map is markedly faster than reconciling three separate systems by hand.
Building a Keyword Map in Thirty Days
The timeline below amounts to five or six hours a week for a store at the thousand-product scale.
Days 1-7: extract your own data
Pull the last three months of queries from Search Console, the zero-result searches from on-site search, and your existing URL list into a single table. No new keywords are researched this week; real existing demand is counted. By the end of the week you should hold three columns: query, impressions, current average position.
Days 8-14: tag by intent and group
Write one of the four intents against every query, and collect queries that share an intent and produce the same results page into a single group. Suffix variants are members of a group, not separate rows. This week 2,000 queries typically fall to 150-250 groups and the job becomes manageable.
Days 15-21: build the map and clear the collisions
Assign exactly one URL to each group. If two groups want the same URL, merge the groups; if one group falls across two URLs, you have cannibalisation and must pick the primary page. Groups with no match are your new-page list. The output of the week is three lists: pages to fix, pages to merge, pages to create.
Days 22-30: ship the first ten pages and set up measurement
Start with the ten pages carrying the highest priority score; do not change them all at once, because you will not be able to tell what worked. Record the starting position, impressions and clicks for each page. Read the result after eight weeks, not after four; looking earlier leads you to mistake noise for signal.
Here is the job for tomorrow morning: open the last three months in Search Console, filter for queries with over 100 impressions and an average position between 8 and 20, and put the top twenty into a table. Those twenty rows are the list where your site gains the most for the least effort today; write which URL appears for each one and the first page of your keyword map is built.
Frequently Asked Questions
Can this be done without buying a paid keyword tool?
Yes. Search Console, Keyword Planner, on-site search records and manual inspection of the results page are enough to build the first map. A paid tool speeds up competitor gap analysis and rank tracking; those save time once you pass a hundred pages, not at the start.
How many keywords should I start with?
Think in proportion to your page count. Since each page gets one primary keyword, a fifty-page site needs fifty primary keywords and perhaps two hundred long-tail terms around them. A list of a thousand keywords is an unusable file unless you have a thousand pages.
How many keywords should a product page target?
One primary keyword plus the product's own variant names. Trying to optimise a product page for two different head terms weakens both. If the second term matters enough, a filtered listing page is the right answer for it.
Can I target the same keyword on both a category page and a blog post?
The same keyword, no; the same topic, yes. The category page should target the commercial query and the blog post the question queries on that topic. Link from the blog post to the category page; the reverse helps too, as long as it does not dilute the category page's focus.
What should I do with zero-result on-site searches?
Split them three ways: define synonym mappings for the misspellings, add the customer's wording to the product name for items you sell but call something else, and for things you do not stock at all, measure the demand and feed it into your range decisions. At most stores this list is the fastest-returning file you have.
Does it make sense to target a keyword showing zero search volume?
If the intent is commercial, yes. Tools usually report fewer than ten searches a month as zero, yet a term ten people search and three buy from is worth more than one a thousand search and nobody buys from. The criterion is not volume, it is the value of the order that query brings you.
Should I open separate pages for variants with and without Turkish characters?
No. Google treats those variants as the same query and separate pages produce cannibalisation. What you should do instead is switch on character normalisation in on-site search and keep URLs free of Turkish characters.
Why is it wrong to treat marketplaces as competitors?
Because product depth and domain authority are not comparable, so the gap list you get is unusable. Use marketplaces as a keyword source instead: their category trees and filter names are the structured lists closest to real user vocabulary. As competitors, pick stores at your own scale.
How often should I update the map?
Monthly if your range moves quickly, quarterly if it is stable. Updating does not mean hunting new keywords; it means checking three things: do newly created pages have a place on the map, has any query started showing cannibalisation, and is content for seasonal terms being prepared on time.
Which metric tells me whether the keyword work paid off?
No single metric is enough, and rankings alone are the weakest of them. Read these three together: total impressions across your target groups, the number of commercial-intent keywords where you sit on page one, and the number of orders from organic traffic. If the first two rise while the third does not, you are targeting the wrong intent.