The Keyword Research Process

Keyword research is how you decide what to build. Done well, it produces a prioritized map of topics with realistic traffic expectations. Done poorly (or skipped), it produces content nobody searches for. This is the repeatable five-step workflow, updated for how search behaves in 2026.

Step 1: Generate seed keywords#

Seeds are the raw, obvious terms in your space. Gather them from:

• Your own knowledge: what would you type to find your product or content?
• Customers and support tickets: the exact language real users use
• Competitors' navigation: their menu structure is their keyword strategy (more in Competitor Analysis)
• Existing Search Console data: queries you already appear for (Performance report)

Ten to thirty seeds are plenty. The quality of seeds matters less than you'd think; expansion does the heavy lifting.

Step 2: Expand#

Feed each seed into expansion sources and collect everything:

• Keyword tools (Google Keyword Planner, Ahrefs, Semrush): "matching terms" and "related terms" reports
• Google autocomplete: type the seed plus each letter a to z
• People Also Ask: a recursive goldmine, since expanding one question loads more
• Related searches at the bottom of the SERP
• Google Trends: check whether demand is growing, seasonal or fading before you commit
• Forums, Reddit and YouTube comments: how people phrase problems when they aren't talking to a search box (see Reddit, Forums & Communities)

crawl budget                        → seed
what is crawl budget                → PAA
crawl budget optimization           → autocomplete
crawl budget for small sites        → PAA
googlebot crawl rate                → related
crawl stats report search console   → tool suggestion
how to check crawl budget           → autocomplete

Expect hundreds to thousands of candidates. That's correct; pruning comes next.

Think in fan-out, not just in strings#

Google describes query fan-out as "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results". AI Overviews and AI Mode may use it. Google's own example: for "how to fix a lawn that's full of weeds", fan-out queries might include "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn" (Google).

Three consequences for your research:

1. Sub-questions matter as much as the head query. Your page can be retrieved for a fan-out sub-query even if it doesn't rank for the user's original question. Ahrefs found only 38% of AI Overview citations came from pages ranking in the top 10 for the main query (March 2026, 863K SERPs), down from about 76% in July 2025, and attributed much of the gap to fan-out (Ahrefs). Treat it as a correlation study, not a rule.
2. Expansion sources already approximate fan-out. PAA, related searches and "vs" queries are the natural sub-questions around a topic. Capture them and map each one to a section or a page.
3. Don't build a page per variant. Google's guide warns that creating separate content for every variation or fan-out query "primarily to manipulate rankings or generative AI responses in Google Search violates Google's scaled content abuse spam policy". It also says you don't need to chase every long-tail phrasing, because its systems understand synonyms. Cluster instead (Step 4).

Step 3: Attach metrics#

For each candidate, gather three numbers:

Reading Search Console data correctly in 2026#

Search Console is the only first-party keyword data you get, and it changed a lot between 2025 and 2026:

• Branded vs non-branded filter. In Performance → Search results, a branded queries filter splits data into Branded and Non-branded across web, image, video and news search. It is AI-assisted rather than regex-based, so it catches misspellings and brand products (Google's example: "Gmail" counts as branded for google.com). It is only available for top-level (domain) properties with enough query volume, and was available to all eligible sites from 2026-03-11 (Google). Do opportunity analysis on non-branded queries: branded clicks come from people who already know you.
• Query groups. The Insights report clusters similar queries (typos, rephrasings, languages) and shows Top, Trending up and Trending down groups. It's a free first pass at clustering your existing demand, available only to properties with large query volume (Google).
• AI features are blended in. AI Overview and AI Mode impressions and clicks count in the normal "Web" totals. The separate Generative AI performance report (global since 2026-08-31) shows impressions only, with no queries. See Measuring AI Visibility.

Step 4: Cluster#

Many keywords are the same topic in different clothes. "how to make sourdough", "sourdough recipe for beginners" and "easy sourdough bread" can likely be served by one page. Group candidates into clusters where a single page can satisfy all queries in the group.

The reliable test: do the same URLs rank for both queries? If the top-10 overlap is heavy, one page can win both. Keyword tools automate this ("keyword clustering" or "parent topic"), Search Console query groups show it for your existing traffic, or you can spot-check SERPs manually.

Each cluster becomes one planned page with:

• A primary keyword (usually the highest-volume phrasing)
• Secondary keywords (variants and fan-out sub-questions to cover within the page as sections)
• An intent label deciding the page type (previous guide)

Step 5: Prioritize#

Score each cluster. A simple model that works:

priority = (volume potential × business value) / difficulty

Then sequence pragmatically:

1. Quick wins: non-branded queries where you already rank around #5 to #20 in Search Console; improving an existing page is cheaper than building one (see Content Refresh & Decay)
2. Low-difficulty, high-value clusters: winnable now
3. Strategic heads: long-term targets that need authority you'll build via #1 and #2

The output is your keyword map: a sheet with one row per planned page (cluster, primary keyword, intent, page type, priority). This document drives your content roadmap for the next topical authority phase.

Next: Topical Authority, turning the keyword map into clusters that build subject-level trust.