Skip to content

Chapter 2: Keyword Research

Entries #46–#95 · 50 entries · 3 sections · Chinese edition
Why this chapter exists Keyword research is demand discovery, not word collection. Here you learn to separate curiosity from intent, size a market from the SERP itself, and build a query map that tells you what to publish next.

2.1 Keyword basics and commercial metrics

11 entries

2.2 Intent types and the language behind them

11 entries

2.3 Finding, clustering, mapping and scheduling

28 entries

Chapter SOP and interview questions

Keyword research SOP for a cold-start site

  1. Harvest seeds from humans, not tools. Sit with product and sales for thirty minutes and write down the ten phrases customers actually use on calls. Tools give you vocabulary; customers give you intent.
  2. Pull the competitive vocabulary. Put each seed into Ahrefs Keywords Explorer or Semrush Keyword Magic, set the market to your primary country, and export everything. Use the “having same terms” and “questions” filters rather than the default list — the default list is padded with the same phrase in different word order.
  3. Clean before you grade. In a sheet, delete anything irrelevant to what you sell, then tag every remaining row with one intent: informational, commercial investigation, transactional, or navigational. A keyword you cannot assign to one of those four is not a keyword, it is a category.
  4. Apply the cold-start filter. Volume above a floor you set deliberately (50 is a common starting point in small markets), difficulty below what your domain can realistically beat. Write the floor and ceiling down in the sheet header so nobody renegotiates them later.
  5. Check the SERP by hand. For every shortlisted keyword, look at the actual top ten. You are looking for evidence the engine is still undecided: forums, Quora or Reddit threads, thin listicles, or pages from small domains. If the top ten is entirely major brands, the difficulty score is telling the truth no matter how low it looks.
  6. Cluster by what the SERP shows, not by stem. Two keywords with the same stem can need different pages; two with different stems can share one page. Merge when the top ten overlap heavily, split when they do not.
  7. Map one keyword cluster to one URL. Every cluster gets exactly one destination. If a keyword has no destination, it is a content brief. If a URL has two clusters, you have built future cannibalisation.
  8. Sequence by commercial value, not by volume. On a new site, publish commercial-investigation pages before high-volume informational ones: they convert, they face weaker competition, and they tell you within weeks whether the positioning works.

What to keep after the exercise

A research file that nobody opens again is a cost, not an asset. Keep three artefacts: the query map (cluster → URL → owner → status), the intent tags, and a short note per priority cluster explaining why it matters commercially. Everything else can be regenerated in an hour.

Interview questions and answer frameworks

Q1: How would you do keyword research for a brand-new SaaS product?

Framework: Four steps, and be explicit about the sequencing. First, discovery interviews — what the product replaces, who signs the cheque, and what those people currently type into Google when they feel the pain. Second, competitive mining — three closest competitors, their top pages and organic keywords, so you start from evidence rather than imagination. Third, intent sorting — commercial-investigation terms (“best X software”, “X alternatives”, “X vs Y”) convert first; informational terms build the topical base but pay later. Fourth, delivery — a query map with one URL per cluster and a three-month publishing calendar, so the research turns into a schedule instead of a spreadsheet.

Q2: A keyword has huge volume but very high difficulty. Do you abandon it?

Framework: No — you reframe it. Read the SERP before you read the difficulty score. If the top ten is all category-defining brands with dedicated teams, you cannot win head-on today, but you can win the modifiers: “for small teams”, “for agencies”, “pricing”, “alternatives”, and the year-qualified variants where freshness matters. If the top ten contains forums, thin pages or clearly outdated content, the engine is telling you nobody has answered well — that is a legitimate entry point, and it requires a genuinely better page, not a longer one. Either way the answer is a decision about sequencing, not a verdict about the keyword.

Q3: What is keyword cannibalisation, and how do you find and fix it?

Framework: Define it precisely: two or more URLs on the same site competing for the same query, so the engine alternates between them and neither accumulates authority. Find it in Search Console — filter the query, open the Pages tab, and look for several URLs trading impressions with unstable positions. Then decide by intent, not by traffic: if the pages serve the same intent, keep the stronger one, merge the useful parts, and 301 the rest. If they genuinely serve different intents, differentiate them properly — separate titles, separate headings, distinct angle — and link from the supporting page to the primary one with descriptive anchor text. The mistake to name out loud is fixing cannibalisation by adding more pages.

Q4: How do you decide the order in which to publish?

Framework: Three inputs: commercial value per visit, realistic time to rank, and production cost. Commercial-investigation pages usually win on the first two and lose on the third, which is why they should not be the only thing you publish. High-volume informational pages compound — they earn links and establish the topical base that makes commercial pages rank — but they take longer to pay. The honest answer is a portfolio: a few high-intent pages early to prove the model, a steady stream of informational clusters to build authority, and a review point at ninety days where the data, not the plan, sets the next quarter.