Keyword research from first principles
Before you buy a keyword database, do this: mine your customers' language, your search box, Google's own hints, and Search Console. The list you get is better.
Keyword research has been sold as a database problem: pay for a tool, type a seed word, export ten thousand suggestions sorted by volume. You can do useful work that way. But the highest-value keywords for a small site usually come from sources you already own — and they cost nothing.
Start with sentences customers actually say
Support emails, sales calls, chat logs, reviews of your product and your competitors’. Write down the exact phrasing: nobody emails you about “workflow optimization solutions”, they say “how do I stop copying this into a spreadsheet every week”. The gap between your category language and customer language is where rankings hide, because your competitors write in category language too.
Google will finish your sentences
Type your topic into Google slowly and note the autocomplete suggestions — those are real queries, ranked roughly by popularity. Then check the “People also ask” boxes and the related searches at the bottom. Ten minutes of this produces a map of how people actually phrase the problem, with zero subscription.
Your own search box is a keyword tool
If your site has internal search, its logs are a list of things visitors wanted and — when the results page was empty — things they wanted that you don’t have. Both halves are gold: the first tells you what to feature, the second what to create.
Search Console: the queries you’re already almost ranking for
Google Search Console shows the queries where you appear, with impressions and clicks. Filter for queries with real impressions but a position between 8 and 25: those are pages Google already half-believes in. Improving an existing page from #14 to #6 is usually far less work than ranking something new from scratch — and GSC hands you that shortlist for free.
Then organize, don’t hoard
A hundred well-understood keywords beat five thousand exported ones. Group them by intent — learning, comparing, buying — and by page: every keyword should have exactly one page whose job it is to rank for it (two pages competing for the same query is cannibalization, and it’s self-inflicted).
Volume data is the one thing this approach doesn’t give you, and honestly: for small sites, relative volume is guessable from autocomplete order and common sense, and precision volume numbers are mostly a way to feel scientific about guesses. Rank for the sentences your customers say, measure the results, and let reality re-rank your list.