I took over SEO for a home-security retailer and the previous manager left me a parting gift: a spreadsheet with 200 keywords, sorted neatly by search volume, and absolutely nothing else. No grouping. No intent labels. No sense of which terms belonged together or which to tackle first. The client read 'we have keyword research' as 'we have a content plan,' and I had to gently explain that a list of keywords is to a content plan what a pile of lumber is to a house. Related, technically. Not the same thing.
I have clustered keywords by hand before. For 200 terms in a niche you're still learning, it eats the better part of a day, and by hour four you start putting 'wireless camera installation' in three different buckets because your brain has turned to soup. The real failure mode of manual clustering isn't slowness, it's drift: the criteria you use to group keyword number 12 are subtly different from the criteria you use for keyword number 180, because you got tired and re-defined a cluster halfway through without noticing. So I tested the Keyword Cluster Builder — Group Seed Keywords Into Topical Themes prompt instead, half-expecting to redo it manually afterward.
What 200 keywords looked like after 90 seconds
I pasted the full list, set the website topic to home-security products and the target market to US suburban homeowners, and ran it on Gemini 2.5 Pro. I chose Gemini specifically because this is a long-input synthesis task with a lot of items to hold in context and a table-shaped output, which is exactly its wheelhouse. Ninety seconds later I had seven clean clusters: installation and setup, monitoring services, smart-home integration, security cameras, alarm systems, rent-versus-own, and insurance discounts. Each cluster came as a tidy Markdown table with intent labels, a named pillar keyword, and four to six supporting articles mapped underneath it.
The priority order put smart-home integration first, and I immediately disagreed — it wasn't the highest-volume cluster, so my gut said start elsewhere. Then I read the one-sentence justification: it had the densest concentration of commercial-investigation intent and the site had zero existing coverage of it. That is the textbook definition of a gap worth attacking. The model was right and my gut was lazy.
- 7 clusters, 200 keywords, organized in under two minutes of run time
- The gap section surfaced 3 topics (DIY security assessment, insurance-claim walkthroughs, smart locks) that weren't in the keyword list at all
- 8 keywords got an [AMBIGUOUS] flag — I checked the SERP and 6 had a different intent than I'd assumed
- The whole output became the client's six-month content calendar, basically untouched
The gap section is the part everyone sleeps on
Everyone fixates on the clusters, but the most strategically valuable output is the gap flag. Knowing what you are NOT covering is as important as knowing what you are, because it tells you where your topical authority has holes a competitor can drive a truck through. Those three implied-but-missing topics became three of the first articles we commissioned, and one of them — the insurance-claim walkthrough — ended up being the highest-converting page on the site, because nobody else in the niche had bothered to write it plainly.
What makes the gap section so good is that it's reasoning about implication, not just sorting what you gave it. The keyword list mentioned 'home insurance discount cameras' but said nothing about the actual claim process — and the prompt inferred that a homeowner researching insurance discounts will, eventually, have to file a claim, so a claim walkthrough is an obvious adjacent need. That is the kind of connective leap I'd expect from a strategist who'd lived in the niche for years, and the model made it in the same ninety seconds it spent clustering. I've since learned to read the gap section first and the clusters second, because the gaps are where the un-obvious money is.
When I'd still cluster by hand
I won't pretend the prompt is always the answer. For a small keyword set — say under 40 terms — in a niche I know cold, manual clustering is fine and arguably better, because I'm encoding real domain judgment. But at 200-plus keywords in a topic I'm learning, the AI is faster and, weirdly, catches thematic connections I'd miss because I don't yet have the mental map. The trick is to treat the [AMBIGUOUS] flags as homework, not gospel — they point you at the exact keywords where a 20-second SERP check pays off.
These days, the moment a keyword-research export lands in my inbox, this is the first prompt I reach for, and I feed its winners straight into my Search-Intent Content Brief — Turn One Keyword Into a Writer-Ready Outline prompt to spin up the actual briefs. Grab the Keyword Cluster Builder — Group Seed Keywords Into Topical Themes prompt on Prompt Dock and paste your own messy keyword export into it — watch a parking lot become a blueprint.