Skip to content
BENIBIX Learn

Grounding vs Faithfulness

People think grounding and faithfulness are the same thing. Grounding is how a system is built: it hands the model source text and asks it to answer from that. Faithfulness is a score. It checks whether one answer really stuck to the source, claim by claim. A grounded system can still give an unfaithful answer, and faithfulness is how you catch it.

AI🔎 Retrieval & Search

Grounding (AI)

Tying a model's answer to trusted sources, so each claim can be checked.

Grounding is tying a model's answer to a trusted source, so each claim can be traced back to it. The source can be documents found by search, a database or a web page. A grounded answer sticks to what the source says and shows where each fact came from. If the source does not cover the question, a well-grounded system says so instead of guessing. It cuts made-up answers, but it does not end them.

Example
Jack's legal tool answers with a quote and a page number for each point. When the contract says nothing about a topic, it says it found nothing.
Why it matters
It turns 'the AI said so' into 'the document says so, here', which a person can check.
Full entry
AI📏 Evaluation & Benchmarks

Faithfulness (AI)

Whether an AI answer only says what its source text backs up. It matters most when the AI answers from your files.

Faithfulness checks whether an AI's answer sticks to the source text it was given. The source can be a document, search results or a set of notes. A faithful answer makes only claims that the source backs up. To test it, the answer is split into claims, and each claim is checked against the source. The score is the share of claims that are backed. An answer can be true in the real world and still fail, if the source never said it.

Example
Lucy's help bot answers from the company handbook. A judge model finds that 46 of 50 claims in its answers are in the handbook, so faithfulness is 92%.
Why it matters
It catches made-up claims in apps that answer from your own files. Those claims sound just as sure as true ones.
Full entry

In one line

Grounding (AI)
Tying a model's answer to trusted sources, so each claim can be checked.
Faithfulness (AI)
Whether an AI answer only says what its source text backs up. It matters most when the AI answers from your files.