Start studying
Module 13 · AI Evaluation Module contents
Module 13  ·  Card 9 / 18intermediate

Groundedness

Groundedness measures how firmly output is anchored to supplied evidence.

It is closely related to faithfulness and often used interchangeably.

Every factual claim should be traceable to a specific source passage.

Ungrounded claims are exactly where hallucination lives.

Requiring inline citations makes groundedness directly checkable.

Learn in depth ↓
Module 13 · AI Evaluation · 18 concepts
In depth  /  15 lines

Groundedness, in more detail

Groundedness measures how firmly output is anchored to supplied evidence.

It is close enough to faithfulness that the terms are often swapped.

Where they are distinguished, groundedness is about the presence of evidence.

Faithfulness is about whether the evidence actually supports the claim.

The working rule is that every factual statement should be traceable to a passage.

Ungrounded claims are exactly where hallucination lives.

They are also the claims a user cannot check, which is what makes them dangerous.

Requiring inline citations turns groundedness into something you can test automatically.

For each sentence, look for the cited passage and check it says that.

Asking the model to quote the supporting text before answering improves grounding.

Quoting first forces it to look rather than reconstruct from impression.

Not every sentence needs a source, since connective and structural text carries no claim.

Over-strict grounding checks flag those and create noise.

Grounded on a wrong document still produces a wrong answer, well cited.

Groundedness is about provenance, and source quality remains a separate problem.

Test yourself

One question on this card. Every option explains itself once you pick, right or wrong.

Question 1
What makes groundedness directly checkable rather than a matter of opinion?
AI Evaluation

18 concepts in this module, then a graded test with every answer explained.

Study this module →
↑↓ move↵ openesc close