Find keyword gaps with AI without inventing search volume
Models are excellent at generating candidates and completely unable to tell you if anyone searches for them.
Before you start
- A keyword tool with real volume data
- Your current ranking pages
What you will be able to do
- Generate candidate terms in the language your buyers use
- Validate every candidate against real volume data
- Cluster by intent instead of by string similarity
There is one hard rule in AI keyword research: the model does not know how many people search for anything, and it will answer anyway, with two decimal places.
Keep that boundary and the rest of the workflow is genuinely faster than doing it manually.
Generate candidates from the problem, not the product
People search for their symptom before they know your category name.
Describe the problem your product solves and ask for the ways someone would describe that problem before knowing any solution exists — including wrong assumptions and the words of an adjacent industry.
This is where models genuinely outperform a team brainstorm: you are too close to the category vocabulary to produce the phrasings a newcomer would use.
Warning Never accept a volume figure from the model
It has no data. The numbers are fluent and invented.
Asked for monthly search volume, a model produces plausible round numbers with no relationship to reality. Same for difficulty scores and CPC.
The output shape is identical to a real tool's, which is exactly what makes it dangerous — it slots straight into a spreadsheet and nothing downstream flags it.
- Pasting model-generated volumes into a forecast. Nobody reviewing the spreadsheet later can tell which column came from where.
Validate the whole list in a real tool
Bulk-check every candidate. Most will have no volume, and that is the point.
Run all of them through a keyword tool at once. Expect a large majority to return nothing — that is a normal and healthy hit rate for this method.
The survivors are terms with real demand that you generated from the problem rather than from a competitor's existing list, which is why they are often the ones nobody is targeting.
Cluster by intent, which is the part AI does well
Different words, same question, one page.
Hand the validated list back and ask it to group terms by what the searcher actually wants. This is a semantic judgement and it is a genuine strength — string-similarity clustering splits "how much does X cost" from "X pricing" and a model does not.
One page per intent cluster, not per keyword. Then check the live results for each cluster: if the top pages are all one format, that format is the answer for that intent.
Generate with the model, validate with the tool, and never let a number that came out of a chat window into a forecast.
Common questions
Was this guide useful?
92% of readers found this useful
Read next
Write ad variants and test them properly
AI removes the cost of writing variants, which makes it very easy to run tests that cannot teach you anything. The discipline is i…
Build a content brief with AI that a writer can use
AI is genuinely good at the research half of a brief and bad at the judgement half. Knowing which half you are asking for makes th…
Edit AI drafts so they stop sounding like drafts
Readers cannot always name what makes text feel machine-written, but they notice. These are the specific tics, and most of them ar…
Estimate what an AI feature will cost you per month
Per-token pricing looks trivially cheap and routinely surprises people at the invoice. The gap is almost always retries, context a…
