How AI answers are made
Hybrid search
Also called: BM25
Hybrid search combines keyword matching, often with a formula called BM25, and meaning-based vector search, and then merges the results.
In plain words
Keyword search is good at exact matches: names, product codes and error numbers. Vector search is good at similar meaning.
Anthropic describes BM25 as a ranking function that finds precise word or phrase matches. It builds on an older idea called TF-IDF.
Example
Anthropic gives this example. Someone searches a support database for "Error code TS-999". An embedding model might find content about error codes in general, while BM25 finds the exact "TS-999" match.
Why it matters for AI search
In Anthropic’s own tests, combining embeddings with BM25 worked better than embeddings alone.
So both your exact words and your meaning count. Exact names help keyword search; clear explanations help vector search.
What you can do
- Write exact names, model numbers and codes as plain text, not only in images or PDFs.
- Explain what they are in plain words as well.
- Use one consistent name for each product.
Often confused with
- Vector search
- Vector search is one half of hybrid search: the meaning-based half.
Sources
Facts and sources last checked on 19 October 2026.
Related terms
How does your page do in AI search?
GEO-Score checks one page on 22 signals, from AI crawler access to structured data, and shows what to fix first.
Analyze a page