If you plan content around buyer questions, you need to distinguish candidate searches from language that may belong in the answer. Keyword Research separates model-suggested search terms, model-suggested writing phrases, and related searches collected from search engines. The three have different uses and should not be merged into one list.
Check LLM Search Keywords first
LLM Search Keywords contains the terms models proposed for the campaign and its personas. The campaign uses selected terms when collecting search results.
That makes this the first list to inspect after a campaign is set up, before reading any other report. An irrelevant query set explains irrelevant results everywhere downstream, and it is far easier to notice here than to diagnose from a confusing opportunity report three days later.
Read the list and ask whether a buyer in your market would search these terms. If several are about an adjacent industry or a different buyer, the campaign description is usually the cause.
Search terms and writing vocabulary are different
LLM Content Phrases contains wording associated with the subject. These support a writer's vocabulary research and are not automatically submitted as searches.
The distinction matters because the two lists are used by different people for different things. A search term is a query to collect results for. A content phrase is language a writer might use because it belongs to how the field is discussed.
Related Searches contains suggestions returned with collected Google and Bing results, which is a third source of wording drawn from observed behavior rather than from a model.
Comparing the model lists against Related Searches is worthwhile: where they agree, the wording is well established; where the model proposes terms that never appear in related searches, treat them as vocabulary rather than as queries.
Interpret the columns correctly
Model lists use a position-weighted score and an occurrence count. A phrase named early by a model can rank above a phrase mentioned later more often, so the two columns can disagree and both are informative.
Related Searches uses observed counts rather than the model scoring rule, so its numbers are not comparable with the model scores.
None of these is a search-volume estimate. They describe the collected research: what models proposed and what search engines returned alongside results. A high score means a term was prominent in that research, not that many people search it.
If you need volume, that is a different type of data from a different tool.
How to use the output
Select wording because it fits the customer's question, then put it where it does work: in a content brief for a writer, or as a tracked keyword or prompt.
Choosing terms by score alone produces a list that reflects how models discuss the subject rather than how buyers ask about it.
Before preparing the brief, group the terms by the reader’s information need. Keep each group’s sources visible so the writer can check how the terms were used.
See Content FAQs, choosing prompts to track, or campaigns.