Keyword research from scratch: a repeatable process for a new site

Keyword research has a reputation for being a numbers exercise: export a list from a tool, sort by volume, write articles for the top rows. That approach produces sites full of pages nobody needed, competing for terms they will never win. Done well, keyword research is closer to customer research. It tells you what people want to know, in their own words, and how hard it will be to become the best answer.

This guide describes a process that works for a new site or a new section, using a hypothetical business that sells refurbished espresso machines. The output is a page plan: which pages to build, what each one targets, and in what order.

Step 1: start from customers, not tools

Tools only expand what you give them. If your seed list is the business's internal vocabulary ("pre-owned prosumer espresso units"), the tools will expand that, and miss what buyers actually type ("second hand espresso machine worth it"). Collect seeds from sources that capture the customer's language:

• Sales and support conversations: emails, chat transcripts, call notes. Every question asked more than twice is a seed.
• Reviews, both yours and competitors': what people praise and complain about.
• Forums and communities where your audience talks. Look at thread titles; they are often phrased exactly like searches.
• Your own site search logs, if you have a site already.
• Search Console for an existing site: the queries you already appear for, especially those where you rank on page two.

For the espresso example, this produces seeds like "refurbished espresso machine", "is a used espresso machine worth it", "descaling", "dual boiler vs heat exchanger", "espresso machine warranty refurbished", "best espresso machine under 500".

Step 2: expand each seed

Now expand each seed into the full set of related searches. Free sources get you far:

• Google autocomplete: type the seed and note suggestions, then add each letter of the alphabet after it.
• People Also Ask boxes and related searches at the bottom of results. Clicking a PAA question loads more.
• Search Console queries filtered by a word (use the regex filters described in the regex guide).
• Paid tools for volume estimates, keyword difficulty scores and the keywords competitors rank for.

Collect everything into one list with the source noted. Expect hundreds of rows for a small niche. Do not filter yet.

Step 3: group by meaning, not by spelling

Many rows are variations of the same need: "refurbished espresso machine", "refurbished espresso machines for sale", "reconditioned espresso machine". One page should serve all of them. Group rows that a single good answer would satisfy. A practical test: search two candidate keywords; if most of the top ten results are the same pages, they belong in the same group.

Name each group by its clearest phrase, sum the volume estimates, and note the dominant intent. You will end up with tens of groups instead of hundreds of keywords. Each group is a candidate page.

Step 4: judge intent from the results page

For each group, look at what currently ranks. The results page shows you what Google believes searchers want.

What ranks	Intent	Page you need
Product listings, shop category pages	Transactional	Category or product page
Reviews, "best X" lists, comparisons	Commercial investigation	Comparison or buying guide
How-to guides, explanations, videos	Informational	Guide or tutorial
One brand's pages	Navigational	Usually not worth targeting unless it is your brand
A mix	Mixed or unclear	A page that serves the dominant intent first

In the espresso example, "refurbished espresso machine" shows shop pages: it needs a category page. "Is a used espresso machine worth it" shows articles and forum threads: it needs a guide. Writing a blog post for the first or a category page for the second would fight the results page. The search intent guide covers the details.

Step 5: estimate difficulty honestly

Difficulty scores from tools are mostly based on the links pointing to the ranking pages. They are a reasonable start but miss things. Look at the top results yourself and ask:

• Are the ranking sites large, well-known brands, or sites similar in size to yours?
• Do forum threads or thin pages rank? That suggests a gap a good page could fill.
• How good is the best result, really? Could you produce something clearly more useful, with first-hand detail?
• Does the results page have ads, shopping blocks, AI-generated answers or video carousels that push organic results down?

Rate each group simply as easy, medium or hard for your site specifically. A new site should mostly build easy and medium pages first, earning visibility and links before attempting the hard ones.

Step 6: prioritise

Score each group on three axes, even roughly:

• Business value: how close is the searcher to buying, and how valuable is a customer? "Refurbished espresso machine warranty" is close to a purchase; "history of espresso" is not.
• Opportunity: volume and difficulty together.
• Effort: can you write it from existing knowledge, or do you need testing, photos, data?

High value, decent opportunity and low effort go first. High value and hard gets planned for later, once the site has authority. Low value groups are dropped, however tempting their volume.

Step 7: map groups to pages

The final output is a table with one row per page: URL, page type, primary keyword group, secondary phrases, intent, priority, and status. Rules for the mapping:

• One primary group per page. Two pages targeting the same group compete with each other; see keyword cannibalization.
• Existing pages first. If a page already ranks on page two for a group, improving it is usually faster than writing a new one.
• Plan the links. Each guide should link to the relevant category or product; categories should link to their buying guide.

Each new page then gets a brief before writing. The content briefs guide covers what goes into one.

What the plan looks like

For the espresso business, the first rows of the plan might read:

Page	Type	Primary group	Priority
/refurbished-espresso-machines/	Category	refurbished espresso machine(s)	1
/guides/is-a-used-espresso-machine-worth-it/	Guide	used espresso machine worth it	1
/guides/dual-boiler-vs-heat-exchanger/	Comparison	dual boiler vs heat exchanger	2
/warranty/	Policy page with buyer focus	refurbished espresso machine warranty	2
/guides/descaling-an-espresso-machine/	How-to	descale espresso machine	3

Short, specific and assigned. Anyone on the team can pick up a row, write a brief and know exactly what the page is for.

Mistakes that waste the research

• Chasing volume over fit. A high-volume term your product cannot satisfy brings visitors who leave.
• One page per keyword variation. Separate pages for "refurbished" and "reconditioned" split the same intent in two.
• Ignoring the results page. Writing a 3,000-word guide for a query that shows only product listings.
• Never revisiting. The plan made at launch is guesswork; the plan made after six months of real data is knowledge.

Keeping it alive

Keyword research is not a one-off. Revisit it every quarter with fresh Search Console data: new queries appear, pages rank for things you did not plan, and some groups turn out to be worth more than expected. Look specifically for queries with many impressions and low clicks at positions 5–15, which are often the cheapest wins. Tracking the average position of each primary keyword over time, which DeskSEO does daily once Search Console is connected, shows whether new pages are climbing or stalling.

Also watch for questions that never appear in keyword tools but come up in AI assistants and support chats. Searches are getting longer and more conversational, and the specific question a customer asks a chatbot today is often the long-tail query that keyword tools will report as zero volume for another year.