Keyword research: intent before volume
Keyword research is deciding which questions your site should be the answer to. Done badly, it is a spreadsheet sorted by search volume. Done well, it starts from intent, weights opportunities by how close you already are to winning them, and treats every volume number with the skepticism it deserves.
Start with intent, not numbers
Every query encodes a purpose, and the purpose decides what kind of page can rank for it. The standard classes:
- Informational. The searcher wants to understand something. Guides, explainers and reference pages compete here.
- Navigational. The searcher wants a specific site or brand. You either are the destination or you are not.
- Commercial investigation. The searcher is comparing options before a decision. Comparisons, reviews and buying guides fit.
- Transactional. The searcher is ready to act. Product, pricing and signup pages belong here.
Mismatched intent is the most common silent failure in content plans: a product page will not rank for an informational query however well it is optimized, because the results Google shows for that query are all explainers. Look at what currently ranks before writing anything; the results page tells you what intent Google has concluded the query carries.
The cheapest wins sit at positions 11 to 25
Queries where you already rank just off the first page, roughly positions 11 to 25, are the highest-yield targets in almost every plan. Google has already judged the page relevant enough to nearly rank; what is missing is usually depth, freshness, internal links or a clearer answer, and each of those is a revision, not a new page. Improving an existing near-ranking page is typically far less work than lifting a new page from nowhere, and the result is measurable within your own measurement setup.
Be honest about volume
Search volume numbers are estimates, and different tools produce different estimates for the same query because they sample and model differently. Treat volumes as rough tiers, useful for telling a big topic from a tiny one, and not as forecasts. A common trap is dismissing queries with low reported volume: tools underreport long and specific queries, and a query a tool shows as near zero can still bring a steady trickle of exactly the right visitors.
Long tail versus head terms
Head terms are short, broad and brutally competitive, and their intent is often ambiguous. Long-tail queries are longer, more specific, individually small and collectively large, and their specificity is precisely what makes them convert: someone who typed the whole question wants the whole answer. A realistic plan for a site without dominant authority is a cluster strategy, covering a topic's long tail thoroughly with interlinked pages so that authority accumulates toward the head term over time, rather than attacking the head term first.
How AI answers change what a keyword is worth
AI Overviews, AI Mode and standalone answer engines change the arithmetic of keyword value. When an AI answer sits above the classic results, being one of its cited sources can matter more than holding position four below it, because the citation is where the reader's attention actually is. Google's AI features also use query fan-out, issuing several related sub-queries and assembling the answer from pages that rank for them, so a page can be cited for a question nobody ever typed into the box. The practical consequence: judge a keyword not only by its volume tier but by whether you can say something citable about it, in the sense described in how to get cited.
Build the plan from data you own
Google Search Console shows the queries your site actually appears for, with impressions, clicks and average positions, and it is the one keyword dataset that is measured rather than estimated. A simple, durable planning loop: export your queries, flag those sitting at positions 11 to 25 for revision, mine the long-tail queries you rank weakly for as topics for new pages, and group everything by intent before scheduling. Repeat quarterly. Vupie runs this same loop automatically for the sites it writes for, and it works just as well as a manual habit. Third-party tools add competitor context on top; your own Search Console data is the ground truth the plan should stand on.