Step-back prompting is a retrieval technique where the model generates a more general version of the query before retrieving, to surface broader foundational context before narrowing to the specific question. The model steps back from the specific query to retrieve the conceptual grounding the answer depends on.
Effective for complex queries where the specific answer depends on foundational concepts that may not be explicitly named in the query. For content strategy, it reinforces the value of foundational explainer content alongside specific answer content — the step-back query needs something to retrieve too.