AI Glossary Term

What does “Factuality” mean?

Definition

How accurately a model's claims match verifiable real-world information.

Browse the full AI Glossary Explore all free AI guides

How Factuality works

Example

A checker splits a model's long answer into individual claims and compares each with reference sources, counting how many are supported, contradicted, or unverifiable.

Why it matters

People rely on AI for facts, and fluent errors are easy to miss. Factuality measures help compare models and catch regressions.

Common questions

Factuality vs groundedness?

Factuality compares claims with the real world; groundedness compares them with provided sources.

How is it measured?

Question-answering benchmarks, human fact-checking, and automated claim checking against references.

How can it be improved?

Retrieval from trusted sources, citations, and training methods that reward accuracy and uncertainty.

CalibrationHow well a model's confidence scores match actual correctness probabilities.Data DriftA shift in real-world input data over time that can degrade model performance.Model DriftPerformance degradation over time as real-world conditions diverge from training assumptions.Frequency PenaltyA decoding setting that reduces the probability of tokens proportionally to how often they've appeared so far.AI GovernancePolicies, standards, and oversight mechanisms that guide how AI is developed and used in society.Algorithmic TransparencyHow clearly an AI system's logic, data sources, and limitations are documented and understandable.