Category: AI Search & Retrieval
Definition
A Relevance Score is a numerical or ranked value representing how relevant a document, webpage, passage, or other piece of information is to a particular query.
Search and AI retrieval systems can use relevance scores to compare potential results and determine which information should be prioritized.
Why It Matters
When thousands or millions of possible sources could relate to a query, a retrieval system needs a way to distinguish stronger matches from weaker ones.
A relevance score provides a signal that can help the system:
- Identify the most useful results
- Rank retrieved content
- Filter weak matches
- Select passages for AI-generated answers
- Compare competing sources
Example
A user asks:
“What are the benefits of Product X for enterprise teams?”
A retrieval system might assign different relevance scores to available documents:
- Enterprise product guide — 0.92
- Product overview — 0.78
- Customer case study — 0.71
- General company history — 0.24
The exact scoring scale varies between systems, so a score of 0.92 does not universally mean the same thing across different platforms.
How Relevance Scores Are Generated
The method depends on the retrieval system.
Traditional search systems may use signals such as keyword frequency, term importance, and document structure.
Modern AI retrieval systems can also use:
- Semantic similarity
- Embeddings
- Query intent
- Entity relationships
- Context
- Metadata
- Freshness
- Other ranking signals
Some systems combine multiple signals into a single relevance score.
Relevance Score in Vector Search
In vector-based retrieval, a relevance score may be derived from the similarity between the query embedding and the embedding of a document or passage.
Depending on the system, this may involve measures such as cosine similarity, do
