Factual Recall is the ratio of correctly generated factual statements to the total number of factual statements present in the ground-truth source, measuring the completeness of information extraction. It answers the question: "Of all the facts that should have been included, how many did the model actually capture?" A high recall score indicates the model successfully retrieved and reproduced the majority of the source's verifiable claims without omission.
Glossary
Factual Recall
The ratio of correctly generated factual statements to the total number of factual statements present in the ground-truth source, measuring the completeness of information extracted by a language model.
COMPLETENESS METRIC
What is Factual Recall?
Factual Recall measures the completeness of information extraction by quantifying the proportion of ground-truth facts successfully captured in a generated output.
This metric is the counterbalance to Factual Precision. While precision penalizes the inclusion of extraneous or incorrect facts, recall penalizes omission. In high-stakes domains like medical summarization or legal document review, low recall represents a critical failure mode where vital information is silently dropped. Recall is often calculated alongside precision to compute the Knowledge F1 score, providing a balanced view of a model's extraction fidelity.
METRICS
Key Characteristics of Factual Recall
Factual Recall measures the completeness of information extraction—how many of the ground-truth facts a model successfully surfaces. It is the counterbalance to Factual Precision, forming the foundation of the Knowledge F1 score.
01
The Core Formula
Factual Recall is calculated as the ratio of correctly generated atomic facts to the total atomic facts in the source. A recall of 1.0 means every verifiable fact from the ground truth appears in the output.
- Formula:
Correct Facts / Total Ground-Truth Facts - Unit: Ratio between 0.0 and 1.0
- Contrast: Precision penalizes extra facts; Recall penalizes missing facts
02
Atomic Fact Decomposition
To compute Recall, both the source and generated text must be broken into atomic facts—single, verifiable triples (subject, predicate, object).
- A sentence like "Paris, founded in the 3rd century BC, is the capital of France" decomposes into three atomic facts
- Tools like FActScore automate this decomposition using LLMs
- Incomplete decomposition leads to inflated or deflated Recall scores
03
Recall vs. Precision Trade-off
Factual Recall and Factual Precision exist in tension. A model that lists every possible fact achieves perfect Recall but poor Precision. A model that only states one certain fact achieves perfect Precision but poor Recall.
- Knowledge F1 is the harmonic mean of the two
- High Recall, Low Precision: Verbose, includes hallucinations
- Low Recall, High Precision: Overly conservative, misses key information
04
Entity-Level Recall
A specialized variant measuring whether all named entities (people, locations, dates, organizations) from the source appear in the output.
- Critical for biomedical NER and legal document review
- Missing a drug interaction entity is a high-risk Recall failure
- Evaluated separately from relation-level correctness
05
NLI-Based Verification
Modern Recall evaluation uses Natural Language Inference (NLI) models to automatically determine if a generated statement is entailed by the source.
- Each atomic fact from the source is checked against the output
- An entailment classification counts as a correct recall
- Contradiction or neutral classifications indicate a missed fact
- Benchmarks like SummaC and AlignScore use this approach
06
Recall in RAG Systems
In Retrieval-Augmented Generation, Factual Recall has two layers:
- Retrieval Recall: The proportion of relevant documents the retriever surfaces from the vector store
- Generation Recall: The proportion of facts from retrieved documents that appear in the final output
- A failure at either layer produces a factually incomplete response
Enabling Efficiency, Speed & Accuracy
Intelligent Analysis, Decision & Execution
We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.
Talk to Us
Search across company data
Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.
Useful when people spend too long searching or get different answers from different systems.
Enterprise searchRAGPermissions
Read more
Automate internal workflows
Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.
Useful when repetitive work moves across multiple tools and teams.
AI agentsWorkflow automationGovernance
Read more
Add AI to products and internal tools
Build assistants, guided actions, or decision support into the software your team or customers already use.
Useful when AI needs to be part of the product, not a separate tool.
AI integrationDecision supportModel routing
Read moreFACTUAL RECALL
Frequently Asked Questions
Clear, technically precise answers to the most common questions about measuring and improving the completeness of information extraction in language models.
Factual recall is the ratio of correctly generated factual statements to the total number of factual statements present in the ground-truth source, measuring the completeness of information extraction. It answers the question: 'Of all the facts that should have been mentioned, how many did the model actually retrieve?' The calculation is straightforward: Recall = (Number of Correctly Generated Facts) / (Total Number of Facts in the Source). A recall of 1.0 indicates the model extracted every verifiable fact from the source document without omission. This metric is critical in summarization and question-answering tasks where missing a key detail—such as a drug contraindication in a medical summary—constitutes a high-risk failure. Factual recall is the complement to Factual Precision, which measures exactness rather than completeness. Together, they form the Knowledge F1 score, the harmonic mean that balances both dimensions.
About the author
Prasad Kumkar
CEO & MD, Inference Systems
Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.
His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.
LinkedIn
Limited slotsGet a Free AI Consultation
Partnered with leading AI, data, and software stack.
How We Work
Custom AI workflows for your Business
One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.
01
Review the use case
We understand the task, the users, and where AI can actually help.
Read more02
Pick the right approach
We define what needs search, automation, or product integration.
Read more03
Build the first useful version
We implement the part that proves the value first.
Read more04
Improve from there
We add the checks and visibility needed to keep it useful.
Read moreThe first call is a practical review of your use case and the right next step.
Talk to Us