Search-Intent Topic Clusters: A Keyword Research Playbook

Let's pretend your team has a spreadsheet with 200 keywords in it. A few have decent volume. Some are phrases your sales team hears all the time. Others are variations of articles you've already published.
You could sort that sheet and start assigning posts. But you still wouldn't know whether you're planning 20 different answers or five answers with four slightly different titles.
That's the decision I'd make before building a content calendar. A search-intent topic cluster groups related questions around a shared subject, while giving each page a distinct job. Keyword research helps you understand how people look for those answers. The cluster map helps your team decide which answers belong together, which deserve their own page and what you already have.
Start with the question your buyer is trying to answer
A broad topic doesn't give a writer much direction. Take "customer onboarding." Someone searching that phrase could be a customer success manager building a process, a founder choosing software or a new employee trying to understand their responsibilities.
Those people may use similar words. They don't necessarily need the same article.
For each proposed page, write down the reader, the situation they're in and the decision they need to make. A working note might be: "A customer success lead at a small SaaS company needs to decide what belongs in the first 30 days of onboarding, before choosing a tool."
Now you have something to test against the keyword list. A checklist could help that person. A comparison of onboarding platforms might be useful later, but it won't replace the checklist.
I wouldn't throw search volume away. I'd use it alongside what buyers ask, what the business can credibly explain and what already exists on the site. A popular query can be a poor use of your team's time when the answer has little connection to the people you serve.
Use search results to test intent, not dictate the whole article
Look at the current results for a small set of representative queries. Record the date, market and device where relevant. Note whether the results mostly contain explanations, comparisons, product pages, templates or something else.
Then open a few results. A title that says "guide" doesn't tell you whether the page actually helps someone make a decision.
Ask what the pages assume the reader already knows, how specific they are and what they leave unresolved. If the results are mixed, keep that uncertainty in the brief. Don't force a single intent label because your spreadsheet requires one.
The familiar labels are still useful:
- Informational: the reader wants to understand something or complete a task.
- Commercial investigation: the reader is comparing approaches, products or providers.
- Transactional: the reader is trying to take an action, such as starting a trial or buying.
- Navigational: the reader is trying to reach a particular company, product or page.
A query can have more than one plausible intent. Search results are evidence about the current result set, not a command to copy the same format, length or headings as everyone else.
Google's people-first content guidance asks whether a page serves its intended audience and adds something useful. That's a better brief than "write a longer version of the first result."
Build the cluster around distinct page jobs
Let's keep the onboarding example. An overview could explain how to build a SaaS customer onboarding process. Supporting pages could help a reader create a checklist, choose milestones or compare software.
Before adding each page, finish this sentence: "Someone would read this instead of, or after, the overview because..."
If the answer is just "it targets another keyword," I'd look at the page again.
A simple cluster record needs six things:
- The buyer question. What does the reader need to work out?
- The page's job. Explain, compare, help implement or help choose.
- The existing URL. Is there already a page that should answer it?
- The evidence. Which product facts, examples or public sources can support the answer?
- The next useful page. Where might the reader need to go after this?
- The decision and owner. Refresh an existing page, prepare a genuinely different answer or leave it out.
This doesn't require a special platform. A shared spreadsheet, a well-maintained document or your existing planning tool can work. The important part is that another person can understand why the page exists without reconstructing the entire research process.
Also, the main page in a cluster is not automatically the canonical URL for every supporting page. A hub organizes related content. Canonicalization addresses duplicate or very similar versions of a page. Distinct supporting articles shouldn't all point their canonical tags at the hub just because they share a subject. Google's canonicalization guidance explains that distinction in technical terms.
Check the existing pages before assigning new ones
Before assigning the next post, put the existing pages next to each other.
Read the body, not just the title. Two titles can sound similar while answering different questions. And two completely different titles can lead to almost the same article.
Compare the intended reader, the decision, the depth of the answer and the evidence. Then look at the available page-level search data. In Search Console, use a defined date window and inspect the queries associated with each page where they're available. Keep the property-level query export separate: a query appearing somewhere on your site doesn't establish that a particular article received it.
If two pages appear for a related query, that alone doesn't prove harmful cannibalization. Check whether they satisfy different needs and whether the overlap is actually causing a problem. A sparse or empty query table also doesn't prove that nobody wants the answer.
For a refresh, preserve what is still useful. Correct stale claims, improve the missing explanation and make the page's purpose clearer. Consolidation, redirects and deleting pages are separate decisions with their own risks. Don't make them an automatic side effect of a clustering spreadsheet.
Prioritize the answer you can support
A scoring formula can help a team explain its choices, but it can also hide a lot of guesses behind a decimal point.
I'd start with four questions:
- Does this help a buyer we actually serve?
- Is the question important enough to answer now?
- Can we provide a credible, useful answer with the evidence we have?
- Is an existing page close enough that a refresh would be more useful than another post?
Then use search estimates and first-party performance to inform the order. Label where those numbers came from. A keyword tool's monthly estimate is not the same as impressions recorded for your page. Neither tells you how much pipeline a new article will create.
If you use a score, make the assumptions visible. "High buyer relevance, limited supporting evidence" tells an editor more than a score of 8.4. It also tells them what needs to happen before the article is ready.
For the hypothetical onboarding cluster, the checklist might come first because the team has a tested process it can explain. A software comparison may need more research and a clear basis for evaluating alternatives. That's a reasonable reason to sequence the work differently, even if the comparison keyword has more estimated searches.
Turn the research into a brief someone can use
Once you hand the plan to a writer, the brief needs to carry the decisions you just made.
Carry forward the reader's question, the page's specific job, its relationship to existing pages and the facts it needs to get right. Include the claims the writer must not make, especially where product capabilities have changed.
For the onboarding checklist, that might mean specifying who owns each step, what the customer needs to supply and what happens when an implementation slips. Those details make the answer useful. Repeating "customer onboarding checklist" in every heading doesn't.
Use headings that describe the actual sections. Explain the answer in text, even when a diagram helps. Add links where another page genuinely helps the reader continue. Google's link guidance recommends crawlable links and descriptive, relevant anchor text, rather than stuffing every keyword variation into a link.
This also gives you a sensible starting point for AI search. Google says its ordinary SEO practices remain relevant to AI Overviews and AI Mode, with no special optimization or schema required. A clear cluster structure doesn't guarantee inclusion, ranking or citations. It helps you organize the answers you're responsible for publishing.
Keep the business context attached to the work
The keyword list can stay the same while the company changes. A product capability gets retired. A new audience becomes a priority. The team decides it won't make a particular claim anymore.
If those decisions live only in the last editor's head, the next brief can bring the old story back.
This is where Oleno fits. Oleno is marketing governance software: a place to maintain positioning, product knowledge, buyer context and reusable Plays so people and connected AI tools can work from the same marketing context. The AI tool still needs to request and apply the relevant context. Oleno isn't automatically researching search results, scoring clusters or publishing the plan described here.
A shared document or AI project can also support this work when someone maintains it and the team knows which version to use. The question is whether the current decisions reach the person or tool doing the next task. You can read more about that in our guide to using knowledge bases and RAG for on-brand content.
When you review the cluster later, keep the original planning assumptions alongside the results. Use comparable date windows. Look at which questions bring people to each page, whether the answer is still accurate and whether readers take a relevant next step. Record missing data as missing data.
Then make the next specific decision. Maybe one paragraph needs a better example. Maybe the checklist answers the question well and doesn't need another rewrite. Maybe the comparison still needs evidence before it should be published. You don't have to rebuild the whole cluster every time you learn something.
About Daniel Hebert
I'm the founder of Oleno, SalesMVP Lab, and yourLumira. Been working in B2B SaaS in both sales and marketing leadership for 13+ years. I specialize in building revenue engines from the ground up. Over the years, I've codified writing frameworks, which are now powering Oleno.
State of AI in Marketing 2026
Free report · 2026
Your campaign might be the wrong answer.
Read State of AI in Marketing 2026 before writing another campaign.
Get the reportFrequently Asked Questions