Keyword Research in 2026

Why is search volume less useful than it was?

Volume tells you how many people search a phrase. It says nothing about what they want, whether you can serve them, or whether the ranking even produces revenue. Two keywords with identical volume can have opposite commercial value and wildly different difficulty.

Three shifts have made this worse. Search volume reporting is increasingly bucketed and imprecise, so small differences are not measurable. AI answer engines intercept queries before the click, so some informational volume now produces no click at all. And a large share of buying research happens in assistants rather than on results pages, where traditional volume tools see nothing.

The practical shift is that you now research intent and topic, and treat volume as a secondary filter rather than the starting point.

What are the four intent stages?

Intent determines what content a query needs and where it sits in your funnel. Get this wrong and you produce excellent content that nobody converts from.

The mapping is not optional decoration. A problem-aware query served a service page converts poorly; a provider-aware query served a blog post converts badly. Same keyword volume, opposite required action.

How do I research a topic rather than a keyword?

Topic research is the higher-leverage activity, because topics generate clusters of queries rather than single terms, and clusters are what build authority.

1. Start from what you sell, not what you might sell. List the problems you genuinely solve. Keywords disconnected from your offer are traffic, not revenue.
2. Map the topic as a set of questions. What does someone need to know to do this properly? Each question is a candidate page, and the questions you cannot answer well are the ones to skip.
3. Check what currently ranks for each question. The pages that rank are the pages Google considers sufficient. If you cannot see what they do better, you are guessing.
4. Look for the weak pages. Ranking pages that are thin, dated, off-topic or thin on evidence are your opportunities. A strong competitive set means pick a different question.
5. Decide coverage before you start writing. Which questions will you answer, and which will you deliberately not? Half-covering a cluster dilutes authority across it.
6. Assign intent and content type to each question up front. This prevents the common failure of ten thin posts where three strong pages would have worked.

How do I find queries AI assistants actually receive?

AI assistants receive full natural-language questions rather than keyword strings, and those questions are a distinct and largely invisible set. They are found by asking rather than by typing into a keyword tool.

• Collect the questions your sales team is actually asked on calls. This is the highest-quality source available and nobody else has it.
• Look at autocomplete and People Also Ask for the problem-aware questions, then rewrite each as a full sentence.
• Ask several assistants the same set of questions and note which produce different answers — those are the contested ones.
• Check your own support inbox and review feeds for recurring phrasings.
• Read the questions in forums and industry communities where your buyers talk to each other rather than to vendors.

Because assistants answer in prose rather than listing links, a page that fully answers one complete question has a better chance of being cited than a page that covers a topic broadly. This is where the intent mapping above does double duty — each well-mapped question becomes a citable passage.

How do I find keywords I can realistically rank for?

Difficulty is only half the question — the other half is whether you could produce something genuinely better than what already ranks. Work through four filters in order.

1. Is there real demand? Non-zero, and ideally with commercial intent attached to it.
2. Does the intent match something you sell? If the answer is 'not really', skip it regardless of how easy it looks.
3. Can you see what would beat the current top result? Not 'it looks weak' but specifically what you would add that is missing.
4. Is the cluster worth building? One isolated term is rarely worth a page; a related group of five or six is.

Where you cannot answer the third question, the keyword is not an opportunity regardless of what a difficulty score says. Difficulty tools measure competition, not the possibility of outranking it.

What tools do I actually need?

The most useful free tool is looking at what ranks and asking why it is not good enough — because if the current results are strong, the difficulty is real regardless of what a tool reports.