AI keyword research: separate real queries from AI ideas
AI can expand hundreds of keywords quickly, but those terms are not evidence of demand. Start with real Search Console queries to establish existing demand, then use AI expansion for new directions, labeling each term with its source (Search Console, Semrush estimate, AI expansion or client input) and date. Use estimated volumes only to prioritize, never on the page or as site traffic. Assign each group one target page and improve an existing page when it already covers the intent.
the 23SEOGEO team · Updated
A mistake we corrected
Our keyword module once mixed real queries with model-guessed volumes in the same table, making guesses look measured. They are now shown separately with a source label on every term. The lesson: mixing sources in a keyword table is riskier than picking one wrong term.
Reading zero volume and missing difficulty
A zero in a third-party tool does not mean nobody searches; small markets and long-tail terms are often underestimated. Missing difficulty means unknown, not easy. Judge these terms alongside support questions and Search Console impressions instead of discarding or labeling them low competition.
Related questions
- How to do AI keyword research?
- Are AI-generated keywords reliable?
- Should I drop zero-volume keywords?
Common questions
Can I mass-produce articles from AI terms?
Not recommended. Confirm demand and target pages first, and merge same-intent terms to avoid pages competing.
Can Semrush numbers go on the page?
No. They are estimates for internal prioritization only.
How many keywords can one article cover?
One primary term plus a few same-intent supporting terms; different intents need separate pages.
Sources
- Search Console Performance report
Source of real query data.
- Google: Creating helpful, reliable, people-first content
Warns against producing pages primarily for search engines.