Keyword research has a reputation as a dark art, and the tooling industry likes it that way. In reality it is a repeatable craft with six steps, and every one of them can be done badly fast or done well fast. This guide walks the whole route, tool-agnostic first, and notes where AI genuinely shortens a leg.
Step 1: Collect seeds, not keywords
A seed is a phrase that names your territory: "cold email software", "gluten free baking", "fleet telematics". You need five to ten of them, and they come from talking, not tooling: how do customers describe the problem in emails and sales calls? What words do they use in support tickets, in Reddit threads, in the reviews of your competitors? Write those phrases down verbatim. The gap between how you talk ("outbound automation platform") and how buyers talk ("cold email tool") is where rankings are lost.
Step 2: Expand each seed into the real query space
Searchers phrase one need dozens of ways: questions ("how to warm up an email domain"), comparisons ("mailshake vs lemlist"), qualifiers ("cold email software for agencies"), problems ("emails going to spam"). Your job is to surface that variety. Autocomplete, the "People also ask" box and any keyword tool's suggestion list all work; an AI expansion does the same thing in one pass and tends to find the phrasings you would not think to type. Aim for a few hundred candidates per seed. Do not filter yet; premature pruning is how good long-tail queries die.
Step 3: Tag intent before you look at any number
Every query encodes what the searcher wants: to learn (informational), to compare before buying (commercial), to buy right now (transactional) or to reach a specific place (navigational). Intent decides what kind of page can rank, so tagging it first stops you from planning a product page for a learning query. This deserves its own article, and has one: keyword search intent: the four types and how to read them.
Step 4: Score volume and difficulty, then read them together
Volume tells you how many people search a phrase monthly; difficulty estimates how hard page one is. Neither matters alone. A 50,000-volume keyword with brutal difficulty is worthless to a young site, and a zero-volume keyword is worthless to everyone. You are hunting the quadrant of meaningful demand and winnable competition. Two caveats: volume tools disagree with each other, and difficulty scores are all modeled, not measured. Treat both as bands, not truths; keyword difficulty explained covers what the score actually contains.
Step 5: Cluster keywords into topics
This is the step most people skip and the one that decides whether your site architecture makes sense. Group keywords that one page could satisfy: "best crm for startups" and "startup crm" belong together; "crm pricing" does not. One cluster becomes one article, always. Skip clustering and you will write three articles competing for the same query, split your authority and rank with none. Manual clustering is a spreadsheet weekend; it is also the single step AI does best. The method, both by hand and by machine, is in how to build topic clusters that actually rank, and a keyword clustering tool automates it.
Step 6: Turn the clusters into an ordered plan
Research that ends in a spreadsheet is procrastination with extra steps. The deliverable is a publishing order: which article first, which cluster it serves, what angle, what it links to. Sequence by winnability: early articles target Easy and Moderate difficulty with real intent, because early wins build the authority that later unlocks harder terrain. Write a one-line brief per article while the research is fresh; future-you will not remember why "deliverability checklist" mattered. What a complete plan looks like is on the SEO content plan page.
The honest role of AI in all this
Steps 2, 3 and 5 (expansion, intent tagging, clustering) are pattern-recognition work where AI is genuinely strong and fast. Steps 1 and 6 still deserve a human: you know your customers and your business goals. Step 4 needs real data sources; an AI can estimate directional bands honestly, but it cannot know live volumes, and any tool that pretends otherwise is bluffing (more in AI tools for SEO). That division of labor, machine synthesis plus human judgment, is exactly how Keywordpilot works.
A 60-minute version to start today
- 10 min: write down 5 seeds in your customers' words.
- 20 min: expand each one; collect everything, filter nothing.
- 15 min: tag intent and rough difficulty bands.
- 15 min: cluster, pick the 5 most winnable clusters, order them, write one-line briefs.
That produces a five-article plan better than what most sites publish from. Or hand the hour to the machine and spend it writing.
Put it into practice
The fastest way to apply this article: run your own niche through the free Keywordpilot demo. One seed keyword, twenty seconds, a clustered mini content plan. No account.