Keyword Research: Why Intent Beats Volume

Keyword research is the work of finding the words people search for and choosing one page that matches the intent behind each of them. Volume is an input to that decision, not the decision. The most-searched keyword is often the hardest one, and sometimes it is full of people looking for something you don't sell.

We saw this in our own market. In September 2026 we looked up 14 keywords from our sector in Semrush's Türkiye database. "Seo ajansı" ("SEO agency"), searched 8,100 times a month, had a difficulty score of 16. "Web tasarım ajansı" ("web design agency"), searched 5,400 times a month, scored 68. The less-searched keyword sat in Semrush's "difficult" band, not its "easy" one.

This post uses that data as a worked example of a method you can apply step by step. For the wider picture of SEO, see our SEO guide.

Why shouldn't keyword research start with volume?

Short answer: because people looking for the same thing use different words, and people using the same word look for different things. Volume tells you how often a keyword is searched; it doesn't tell you who searches it or why.

Google's SEO Starter Guide makes the point with a simple example. People who know a topic well search differently from newcomers: one types "charcuterie", another "cheese board". According to the guide, anticipating these differences and writing for your readers could help how your site performs in search (Google Search Central, SEO Starter Guide).

The rest of that section is quoted less often. Google's language matching systems can relate a page to many queries even when the exact terms don't appear on it. So the job is not collecting variations. It is understanding which word maps to which need.

What do volume and difficulty figures tell you, and what don't they?

Short answer: both are estimates, useful as an order of magnitude and not as exact counts. Knowing where a number comes from tells you how much weight it can bear.

Search volume. According to Semrush, average monthly volume is the total searches over the past 12 months divided by 12. The figure is modelled, not measured: Semrush combines third-party data with historical clickstream data and calculates volume with a machine learning model (Semrush Knowledge Base, search volume).

Google Keyword Planner. Google's own tool also presents its figures as estimates. Average monthly searches include a keyword's close variants, are averaged over 12 months by default and are rounded (Google Ads Help, Keyword Planner forecasts). The same page holds an easy-to-miss detail. Keyword Planner's "competition" column measures how many advertisers showed ads on a keyword relative to all keywords, not organic difficulty.

Keyword difficulty (KD). Semrush's KD is an estimate from 0 to 100 of how much SEO effort it would take to rank in Google's top 10 for a keyword. The factors include the median number of referring domains pointing to the ranking pages and the authority score of those domains (Semrush Knowledge Base, KD).

Semrush splits the score into six bands:

Two tools can also give the same keyword different numbers. We pulled Google Ads data through DataForSEO in July 2026 and Semrush data in September 2026 (both Caliency's own queries). For "kurumsal web tasarım" ("corporate web design") Google Ads said 2,900 and Semrush 2,400; for "web sitesi fiyatları" ("website prices") the order flipped: 1,900 against 2,400. Two months separate the queries, so part of the gap may be timing.

Our rule is simple: read volume as an order of magnitude and write the source and date next to it. The gap between 2,400 and 2,900 won't change a decision; the gap between 260 and 8,100 will.

What did we find in our own market?

Short answer: in our service keywords there is almost no relationship between volume and difficulty; in tool keywords the two rise together. These two pictures show why a single rule such as "pick the easy keywords" or "pick the big keywords" doesn't hold.

Data: In Semrush's Türkiye database, we looked at 14 keywords from web, software and SEO services in September 2026. On 4 October 2026 we added 10 keywords people use when looking for an SEO tool, and recorded monthly volume and KD for all of them. We chose the keywords ourselves; this is a sample close to our own business, not a market survey.

The chart orders keywords by volume, largest first, and the bar length shows the KD score. Three things stand out (Semrush, Türkiye database, September 2026; Caliency's own query):

1. Volume doesn't predict difficulty. "Seo ajansı" is searched 1.5 times as often as "web tasarım ajansı" (8,100 ÷ 5,400), yet its KD is 16 against 68. Across the 14 keywords, the rank correlation (Spearman) between volume and difficulty is 0.05, which is effectively zero.
2. Near-synonyms are not the same keyword. "Yazılım şirketi" ("software company") is searched 4,400 times a month, "yazılım firması" ("software firm") 880 and "yazılım ajansı" ("software agency") 260. The first is searched 16.9 times as often as the third (4,400 ÷ 260), and its KD is 20 against 62. Calling yourself an "agency" doesn't mean buyers use that word.
3. Easy keywords aren't small. The 11 keywords with a KD under 30 add up to 28,860 searches a month; the 3 with a KD of 50 or more add up to 23,760. That total counts searches, not people, so one person may be counted twice. Still, postponing the big keyword doesn't mean giving up the demand.

Tool keywords tell the opposite story.

The four highest-volume tool keywords are "seo tool", "site hız testi" ("site speed test"), "seo checker" and "seo test". Their KD ranges from 51 to 83, and together they get 5,970 searches a month. The six keywords with a KD under 30 add up to 1,060, which is 5.6 times less. The rank correlation here is 0.87: as volume rises, difficulty rises with it (Semrush, Türkiye database, 4 October 2026; Caliency's own query).

We didn't check which sites rank for those keywords, and we didn't measure the cause. The practical lesson: an easy-keyword strategy reaches real demand in service keywords but only a thin slice in tool keywords. You only know which picture you are in by looking at your own keywords.

How do you do keyword research? Seven steps

Short answer: start from the real data you already have, expand it with a tool, group by intent and give each intent one page. The order matters; research that starts in a tool skips the queries your site already appears for.

1. Start in Search Console

Search Console is the first-party source for the queries your site actually appears for. Google offers it as a free service (Search Console Help, About Search Console). The Performance report gives clicks, impressions, click-through rate and average position by query. Google recommends it for learning which queries bring traffic to your site (Search Console Help, Performance report).

Know its two limits. Some very rarely searched queries are left out of the table for privacy but still count in the totals. The table also shows at most 1,000 rows, so rare and long-tail rows can drop out (Search Console Help, data discrepancies).

We look at two things in this report: queries with high impressions but few clicks, and queries you get impressions for without ever targeting them. The second shows intents Google associates with you but that no page of yours answers. On a new site the report will be nearly empty; if so, go straight to step two.

2. List the customer's own words

Before opening a tool, collect the words customers use when they write or call: quote requests, sales emails, recurring questions. The test from our small business SEO post works here too: write down, in their words, the three questions your best customers ask before they buy.

This list gives you what no tool can: the buyer's language. It is where you will first notice the gap between "software agency" and "software company".

3. Expand with a tool and add volume and difficulty

Feed the list into Keyword Planner or a tool such as Semrush to get related keywords and estimated volume. Keyword Planner lives inside Google Ads. According to Google, its keyword ideas feature requires billing information and a completed account setup (Google Ads Help, Use Keyword Planner). For an organic difficulty score you need a paid tool such as Semrush.

Write the source, country database and date next to every keyword; three months later you will need to know where each number came from.

4. Group keywords by intent

Classify each keyword with one question: what does the person searching it want to do? We use four groups:

• Informational: wants to learn something ("landing page nedir", "what is a landing page").
• Commercial: looking for a service or provider, comparing prices ("seo ajansı", "seo fiyatları" or "SEO prices").
• Tool: wants to do something right now ("site hız testi", "ücretsiz seo analizi" or "free SEO analysis").
• Local: looking for a business nearby ("istanbul seo ajansı").

There are also brand (navigational) searches, but those come from awareness, not research. The classification below is our interpretation, not Semrush's label. Some keywords belong to more than one group: someone typing "seo analizi" ("SEO analysis") may want a tool, a service or a guide.

Volume and KD: Semrush, Türkiye database, September 2026 and 4 October 2026 (Caliency's own query).

5. Give each intent one page

Once the groups are clear, assign each group one target page. Commercial intent goes to a service page, informational intent to a guide, tool intent to a working tool. Trying to answer "what is" with a service page, or "I'm looking for an agency" with a blog post, leaves both half-served.

Don't open separate pages for variations of the same intent. Google's generative AI search guide, updated in July 2026, is explicit here. Separate content for every variation of a search violates its scaled content abuse policy if the main aim is manipulating rankings (Google Search Central, generative AI search guide). We collected what still makes a difference on the page itself in our on-page SEO post.

6. Weed out pages competing for the same query

This is called keyword cannibalisation: two of your pages compete for the same query and it isn't clear which one is the target. You can check it in Search Console. In the Performance report, filter by query, then switch to the Pages tab. If more than one of your pages gets impressions for the same query, at least one of them has an unclear job.

Our fix is one of two routes. If the pages serve the same intent, we merge them and redirect the old one to the new. If they serve different intents, we separate their titles and content by intent and link them to each other.

7. Order by difficulty, then measure

Among keywords with the right intent, order the work by difficulty. On a new site or one with few links, starting with commercial keywords in Semrush's 0–29 band makes sense. In our table, "seo ajansı" (8,100 / 16) and "yazılım şirketi" (4,400 / 20) fall in that band.

Then measure: track in Search Console whether the target page earns impressions and clicks for that query group. How long to expect before results show is covered, with Google's current documentation, in how long does SEO take.

How does AI search change keyword research?

Short answer: the value of a keyword list falls and the value of an intent map rises. Google's guide describes how its AI model generates related queries at the same time to answer a user's question, and calls this "query fan-out". In its example, a question about fixing a lawn full of weeds fans out into sub-queries about herbicides and chemical-free weed removal (Google Search Central).

According to the same guide, AI systems understand synonyms, so you don't need to worry about lacking enough "long-tail" keywords. The practical upshot: answering a topic's sub-questions on one strong page is sturdier than opening a page for each. We cover how to measure visibility in these answers in our GEO guide.

How did we use this table in our own plan?

At Caliency we organised our blog plan around this table. This is a record of decisions, not a results report. Most of these posts are a few weeks old, too early to claim ranking or traffic results.

• "Software company", not "software agency". We chose the target keyword for our post on choosing a software company from this comparison: 4,400 / 20 against 260 / 62.
• Low-difficulty commercial keywords. We have a post each for "seo ajansı" (8,100 / 16) and "seo fiyatları" (1,000 / 13): how to choose an SEO agency and SEO pricing.
• An easy keyword we deliberately skip. "İstanbul seo ajansı" sits at KD 13 with 2,900 searches, on paper one of the best openings in the table. But its intent is local. Google says local results are based mainly on relevance, distance and prominence (Google Business Profile Help). We are a remote team with no office in Istanbul; the person searching that is most likely not looking for us.
• Big keywords we are postponing. "Web tasarım" (18,100 / 51) and "web tasarım ajansı" (5,400 / 68) sit in the "difficult" band and need quality backlinks on top of content.

Where doesn't keyword research help?

An answer you may not expect from an SEO agency: if your pages aren't in Google's index, even the best keyword list is useless. On a site with indexing problems we don't start with keyword research; we check the technical foundation first. Our free SEO checker is enough for that first step.

The second limit: no tool tells you whether a keyword will bring you customers. Volume and KD estimate demand and competition; your own sales data shows conversion. Third: a low KD doesn't guarantee the first page. Nobody can honestly promise that.

In our SEO services, keyword and intent work is part of the job. We target the queries that convert, not just the ones with big volume, and tie each keyword group to a single page.

Frequently asked questions

How do you do keyword research?

Start with the queries your site already appears for in Search Console's Performance report. Then write down the words customers actually use and expand that list with a tool such as Keyword Planner or Semrush. Group the keywords by intent, give each intent one page, weed out pages competing for the same query and order the work by difficulty. The last step is measuring whether the target page earns impressions for those queries.

What is keyword difficulty (KD)?

It is Semrush's estimate, from 0 to 100, of how much SEO effort it would take to rank in Google's top 10 for a keyword. The calculation draws on factors such as the median number of referring domains pointing to the ranking pages and the authority score of those domains. Semrush calls 0–14 very easy, 15–29 easy, 50–69 difficult and 85 and above very hard. It is a prioritisation signal, not a guarantee.

Can you trust the search volumes in keyword tools?

As an order of magnitude, yes; as an exact count, no. Semrush volume is a 12-month average calculated with a model built on third-party data and clickstream data. Google Keyword Planner also presents its figures as estimates and rounds them. In our own queries, two sources put the same keyword at 2,400 and 2,900. Always record the source and the date next to the number.

Can you do keyword research for free?

Largely, yes. Search Console shows the real queries your site appears for in Google at no cost, and that is the most valuable data you have. Google Keyword Planner gives keyword ideas and volume estimates; according to Google, that feature requires billing information to be entered in the Google Ads account. What paid tools mainly add is a difficulty score and competitor data.

Should I create a separate page for every keyword?

No. Keywords with the same intent belong on one page. Google writes that creating separate content for every possible variation of a search violates its scaled content abuse policy when the main purpose is to manipulate rankings. With two pages for the same query, it is also unclear which one is the target. Filter by query in Search Console and open the Pages tab to see it.

In short

Keyword research isn't collecting keywords; it is finding the intent behind each search and giving it one page. Volume and difficulty scores are inputs to that decision. Both are estimates, and two tools can give the same keyword different numbers.

Across 14 service keywords in our own market, volume did not predict difficulty. "Seo ajansı", searched 8,100 times a month, sits at KD 16; "web tasarım ajansı", searched 5,400 times, sits at KD 68. Across 10 tool keywords, the two rose together (Semrush, Türkiye database, September–October 2026; Caliency's own query). There is no single rule; you can't know which picture you are in without looking at your own keywords.

To see which queries your site appears for today, start with Search Console; to check the technical foundation, try our free SEO checker. If you'd like to build the keyword and intent map together, take a look at our SEO services or simply get in touch.