Every keyword research project starts with seed keywords, the 5 to 15 terms that describe what your business does in plain language. From those seeds, you expand into hundreds or thousands of candidates using a handful of predictable techniques. Good seeds make good lists. Bad seeds waste weeks. This page walks through where great seed keywords come from, how to expand them, and how to avoid the "too broad" trap that buries half of all keyword research.
If you start with the wrong seeds, the expansion step multiplies your mistake. Every tool gives you more of what you asked for. If you seeded "insurance" and you actually sell B2B insurance software, you'll get back 500 consumer life-insurance queries, spend an afternoon filtering them out, and end up with nothing useful.
Good seeds are specific enough to stay on-topic when multiplied, and broad enough to capture different ways customers might phrase a query. The art is balancing those two.
Each row is a seed type. Mix them to get coverage. One or two from each row gives you a set of 8 to 12 solid seeds that generate a well-rounded expanded list.
The most obvious seeds. What you sell, in the plain words your customer would recognize. Not your marketing copy. "Insurance CRM" beats "Modern Revenue Platform for Insurance Organizations."
This is where most seed lists go wrong. Customers don't use your internal terminology. A B2B insurance CRM might call itself that, but customers often search for "insurance agent tool" or "producer tracking system." Mine sales calls, support tickets, review sites, and Reddit threads to find the actual words.
Seeds based on what the customer is trying to accomplish. "Follow up with insurance leads," "track insurance policies," "automate insurance renewals." These often surface long-tail opportunities that pure product-name seeds miss.
The broader market you're in. "Sales software," "agency software," "CRM." Category seeds are useful because they generate "best X" and "X for Y" queries that are ideal for comparison content. They also tend to have higher volume than product-specific seeds.
Plug each seed into Ahrefs, SEMrush, or Google Keyword Planner. Pull four types of suggestions: matching terms (contains the seed), related terms (semantic relatives), questions (queries phrased as questions), and autocomplete suggestions. This generates the bulk of your raw pool.
Every Google search shows a "People Also Ask" box with related questions. Those are queries Google itself knows are related. Search each seed, copy the questions, click one to expand and reveal more. Each one is a potential content target.
At the bottom of every SERP, Google shows "related searches." Add them to the pool. Low effort, surprisingly useful.
For each of your 3 to 5 SEO competitors, export their top 200 to 500 ranking keywords from a tool. Merge into your list. This is the highest-signal expansion method because every keyword is validated by an actual ranking.
Reddit, Quora, niche forums, LinkedIn posts. Search your seeds and read how real customers phrase things. You'll find queries no tool surfaces because they're too new or too niche. This is how you catch the language shifts early.
The most underused goldmine. What do customers ask your support team? What do they type into your site's search bar? What questions come up on sales calls? This is evidence-grade intent data nobody else can see.
For a new project, 500 to 2,000 raw candidates is normal before filtering and clustering. After filtering by relevance, you'll drop 50 to 70% of them. After clustering, 500 raw keywords usually collapses to 50 to 100 clusters. That's your actual content plan.
Tools return lots of technically-related queries that are wrong for your business. "Insurance" as a seed pulls life insurance consumer queries when you sell B2B software. "Running" pulls races and training content when you sell shoes. "Coffee" pulls recipes and cafés when you sell beans.
Filter aggressively on intent and audience fit before you spend any time on volume, difficulty, or clustering. Most expansion lists have 30 to 50% noise. Remove that noise first or everything downstream is polluted.
The first pass is slow. The second time you do it on the same niche, it's twice as fast. By the fifth time you're running keyword research on a related project, the whole workflow takes a couple of days.
Write your seed list now, before you open any tool. Do it on paper if you can. Keep the list to under 15. Hit all four sources. Then, and only then, fire up the expansion methods. A thoughtful seed list saves you hours of filtering. A lazy one wastes them.
Next: keyword research tools, the stack that powers every step above.