Start With the Buyer, Not the Search Bar
Keyword research, in the framing of one SEO practitioner with a decade of client work behind him, is less a technical exercise than an act of empathy: understanding what target customers search for at each stage of their buying journey so content can meet them there. He calls it the "North Star, the compass, the blueprint" that dictates every SEO action and can decide whether a campaign succeeds or fails.
The practical unit of that empathy is keyword intent — the underlying goal behind a query, whether learning, comparing, purchasing, or navigating to a site. Searchers typically move through informational queries ("best way to sharpen a knife"), commercial ones ("King wetstone reviews"), and finally transactional or navigational ones ("buy King wetstone"), and keywords map precisely onto that funnel. Notably, he stresses the end goal is a "conversion," not necessarily a sale, because most queries simply aren't transactional in nature.
That has a counterintuitive consequence for strategy. In his demonstration, building a keyword list for a hypothetical property management software company, he observes that the bulk of qualified traffic and leads will come from informational keywords — there may be only a handful of commercial or transactional terms worth targeting. Many keywords can also be converted into local variants with modifiers like "near me" or a city name, expanding the map further.
AI as an Unlimited Brainstorming Partner
The first step of his three-step process is gathering a large pool of keywords, and this is where AI has changed the economics of the work. He feeds ChatGPT a long prompt — casting it as an SEO specialist assigned to generate informational, commercial, and transactional keywords for the software product, in a mix of short-tail and long-tail forms — along with a product description. Claude receives the same prompt. What once took "at least an hour" of manual seed-keyword work now takes minutes.
The AI output is raw material, not a finished list. Each batch gets pasted into a keyword tool — he uses Ahrefs' Keywords Explorer, though he notes Keysearch is a far more affordable alternative with a 7-day free trial — to reveal metrics like search volume, difficulty, and intent. One early catch illustrates the need for judgment: "property management software for small landlords" excited him with a difficulty of just 12 out of 100 and a search volume of 1.7K, while a term like "rental tools" turned out, on inspection of Google's results, to mean people renting equipment, not software.
The trick, he says, is to keep prompting: "give me more," directed at both chatbots, produces what he calls an essentially unlimited supply of keywords, though relevance tapers as you repeat the request. He frames it like writing — no editing while gathering, just getting every idea on the page with zero judgment before the filtering stage.
Where AI Alone Falls Short: Reddit, Books, and Sitemaps
AI generates ideas, but some of the best ones live in places chatbots don't naturally reach. His next tactics are deliberately analog-adjacent: AnswerThePublic for question variants around a seed term (free, a couple of searches per day, and "always worth doing" even when results are thin), and a subreddit analysis — searching Google for "property managers" plus "site:reddit.com" — to find the communities where real customers complain and ask.
Those community sources pay off in surprising ways. The property management subreddit ranked for 7,096 keywords, and filtering to top-10 positions surfaced targets he'd missed entirely, like "best property management software for small landlords" and even oddball content ideas such as how landlords can deal with fleas in the house. A Quora analysis works the same way.
His most distinctive trick is mining books. He screenshots a table of contents from a property management book on Amazon, pastes the image into Claude, and asks it to extract keyword ideas — sometimes uploading an entire free PDF from Google Books. The advantage, he argues, is that much of this knowledge "is stuck in books" and absent from the search results, offering a genuine competitive edge. Competitor sitemaps, run through a free extractor and fed to Claude, add still more angles.
From a 225-Keyword Pool to a Validated Shortlist
The demonstration's keyword pool grew in distinct waves — 46 keywords after the initial AI-and-tool pass, 145 after the Reddit and book mining, and 225 after competitor analysis — before the second step, filtering, even began. Ahrefs' "matching terms" feature, built from the ten highest-volume keywords in the list, expanded the raw universe to 19,376 related keywords, tamed with presets like high-volume/low-competition and a user-lifecycle filter for top-of-funnel terms.
Filtering starts deliberately narrow. For a brand-new site, he sets keyword difficulty to 1 to build early wins inside a coherent topical cluster, then loosens it to 3 for slightly more attainable targets. A term with only 150 searches per month can still make the shortlist if it's hyper-relevant and barely contested — like "best property management software for small business." The real test, though, happens outside the tool: he Googles each candidate, studies the top five results, and asks two questions — can he make something better, and can he mention his product naturally in the copy? Only a yes to both earns a spot on the shortlist.
Clusters, Merged Intent, and Playing the Long Game
The final step turns a filtered list into a content plan. A topical cluster is a group of interrelated keywords with separate intents — "squatter's rights New York" versus "squatter's rights Ohio" — linked together around a pillar page, and building such clusters signals topical authority to search engines, which he links to higher rankings across related keywords. A keyword cluster, by contrast, groups near-identical intents onto a single page.
Deciding whether two similar keywords truly share intent is settled empirically: Google both terms in side-by-side windows and compare the results. If the top rankings are identical, one page serves both — and you target whichever version carries the higher search volume, ideally using the tool's "parent topic" feature to find the term driving the most traffic to the number-one result.
He recommends refreshing the research every three months or so, depending on publishing pace, and closes with a tempering note: SEO is a long-term process, unlike ads, which deliver results the moment money flows in and stop the moment it stops. But organic traffic, in his view, has "massive staying power" — and as a site's authority grows, the more competitive keywords collected early become reachable later.
Charts & Visual Insights
Growth of the Keyword Pool During the Demonstration
The demo workflow's candidate keyword list grew from 46 to 145 to 225 keywords across three collection phases.
| Collection phase | Keywords in pool | Source |
|---|---|---|
| Initial AI + tool pass | 46 | |
| After Reddit, books, Quora | 145 | |
| After competitor analysis | 225 |
Note: Values are self-reported by the video presenter and not independently verified.