LLM-PREDICTED (Query Expansion Tag)
Knowledge GraphsLLM-PREDICTED is a tag in the query expansion system marking sub-queries generated purely by a language model (LLM) without confirmation from Google search results (SERP). LLM-PREDICTED phrases are hypotheses — the LLM predicts users might search for such topics, but there's no evidence from real data. In the SERP Grounding system, LLM-PREDICTED has lower priority than CONFIRMED (validated by SERP) and SERP-ONLY (exclusively from Google data).
LLM-PREDICTED phrases are valuable as inspiration for content gaps — they may point to topics nobody's covered yet, but they can also be model hallucinations. That's why they require validation: checking in Google Search Console, Ahrefs, or manual searches.
For example, an LLM might generate the sub-query 'probate a konkubinat' (inheritance rights and cohabitation), which isn't actually searched but after verification proves to be a valuable long-tail. In practice, treat LLM-PREDICTED as a list of hypotheses to test, not as certain phrases to target.