Keyword Research for New Websites
A focused research approach for sites with little search history.
Build a small, coherent first backlog
A new site has limited evidence about which topics resonate. Begin with a clearly defined audience and a few foundational problems. Create a hub that helps readers navigate the subject, then publish distinct supporting pages with specific outcomes. Keep a backlog of ideas rather than publishing every discovered variation.
A concrete example
A site for apartment composting might begin with a beginner's setup guide, odor troubleshooting, and choosing a suitable container. These pages support different tasks and can link naturally. Ten pages that restate 'what is composting' would not provide the same value.
How to make the decision
Use early Search Console impressions as feedback, not as proof that every query needs a new article. Improve existing pages when the intent overlaps.
Practical checklist
- Choose a narrow audience
- Outline a foundational guide
- Identify distinct supporting problems
- Avoid overlapping page promises
- Publish a small batch
- Review Search Console evidence after launch
Record your reasoning
For every promising query, keep a short note describing the reader, the expected page type, what the current results miss, and whether an existing URL already covers it. This small habit turns a keyword list into a defendable publishing plan. Recheck a sample of results before writing because search intent and competing pages can change.
What to do next
Write down the single question this research should answer and the evidence you still need. Check the current search results, identify the page that best matches the reader's task, and make one specific improvement or publishing decision. If the answer depends on a third-party product, verify its current terms on the provider's website before acting.
Common question
Should I publish dozens of pages immediately?
A small set of complete, distinct guides is easier to maintain and evaluate. Expand when audience needs and observed queries justify it.