AI Evaluation Metrics Explained: Precision, Recall, Hallucination Rate and More

Teams often pick an eval metric because it sounds rigorous, not because it measures the failure mode that actually matters. What each metric tells you, and what it hides.

Mert Mutlu·Founder & CEO, Aiporate··9 min read·Share on XLinkedIn

Key takeaways

  • Precision answers 'of what the system flagged as positive, how much was actually right'; recall answers 'of everything actually positive, how much did the system catch.' They trade off against each other and neither alone tells the full story.
  • Hallucination rate and faithfulness/groundedness are related but distinct: hallucination rate measures fabrication in general output, faithfulness specifically measures whether a RAG answer is actually supported by the retrieved source documents.
  • Exact match is the right metric for tasks with one correct answer (data extraction, classification); semantic similarity is the right metric for tasks with many valid phrasings (summarization, open-ended generation). Using the wrong one produces misleadingly bad or misleadingly good scores.
  • Latency percentiles (p50, p95, p99) matter more than averages because a mean latency can look fine while a meaningful fraction of real users experience a multi-second stall the average hides.
  • The right metric is a function of the failure mode that's expensive in your specific use case, not a general ranking of which metric is more rigorous.

Ask five teams how they evaluate their AI system and you'll often get five different metrics, chosen less because someone reasoned about the failure mode they're guarding against and more because the metric sounded credible in a planning meeting. That's a real problem, because every metric on this list is genuinely useful for some failure mode and genuinely blind to others. A model with excellent precision can still be dangerously unreliable if recall is what your use case needs. A RAG system with a low hallucination rate on easy questions can still be unsafe if its faithfulness collapses on the hard 10%. This is a plain-language guide to what each metric actually measures, with a worked example for each, and a table mapping use case to the metric that should actually be driving your ship/no-ship decision.

Precision and recall: the tradeoff underneath almost every classifier

Precision and recall are the two most fundamental metrics for any system that makes a yes/no call, flagging spam, detecting fraud, classifying support tickets, routing documents. Precision asks: of everything the system labeled positive, what fraction was actually positive? Recall asks: of everything that was actually positive, what fraction did the system catch? A worked example makes the difference concrete: imagine an AI system reviewing 100 support tickets, 20 of which are genuinely urgent. If the system flags 15 tickets as urgent and 12 of those really are urgent, precision is 12/15 (80%), the flags you got were mostly trustworthy. But recall is 12/20 (60%), the system missed 8 of the 20 truly urgent tickets entirely. These two numbers pull in different directions: a system tuned to flag everything remotely urgent-looking will catch nearly all real urgent tickets (high recall) but bury the queue in false alarms (low precision). A system tuned to only flag the certain cases will have high precision but miss real ones (low recall). F1 score is the harmonic mean of the two, useful as a single number when you genuinely don't have a reason to weight one over the other, but it can mask a bad asymmetry, an F1 of 0.75 could be 90% precision and 63% recall, or the reverse, and those describe very different systems.

Use caseCostlier failureWhat to optimize for
Fraud detectionMissing real fraud (false negative)Recall, tolerate more false positives for manual review
Spam filteringBlocking a real email (false positive)Precision, tolerate a little spam getting through
Medical/safety-critical flaggingMissing a real positive caseRecall, near-zero tolerance for false negatives
Content auto-publishing decisionsPublishing something wrong (false positive)Precision, err toward holding for review
Precision vs. recall: which failure costs more

Hallucination rate vs. faithfulness/groundedness

Hallucination rate is the broadest metric: across a sample of outputs, what fraction contain a claim that's factually wrong or fabricated, regardless of whether the system had access to source material at all. It's typically measured by having a human or a strong secondary model check claims in the output against known ground truth. Faithfulness (also called groundedness) is a narrower, more useful metric specifically for retrieval-augmented systems: it measures whether the answer is actually supported by the documents the system retrieved, independent of whether those documents themselves are correct. A RAG system can have perfect faithfulness (every claim traces back to a retrieved passage) while still giving a wrong answer, because the retrieved passage itself was outdated or wrong; that's a retrieval problem, not a faithfulness problem, and the two metrics together tell you which stage of the pipeline to fix. Conversely, a RAG system can retrieve perfectly correct documents and still produce an unfaithful answer, because the model added an unsupported inference on top of what the documents actually said, that's a generation problem, not a retrieval problem. Measuring only overall hallucination rate collapses this useful distinction; measuring faithfulness separately from retrieval accuracy tells you which half of the pipeline to fix.

Exact match vs. semantic similarity

For tasks with one clearly correct answer, extracting a specific field from a document, classifying a ticket into a fixed category, answering a factual question with a single correct value, exact match (does the output literally match the expected answer, sometimes with light normalization) is the right metric, and using semantic similarity instead tends to inflate scores by giving credit for 'close but wrong' answers that shouldn't count. For open-ended generation tasks, summarization, drafting, open-ended Q&A, there often isn't one correct phrasing, and exact match against a single reference answer will score a genuinely good output as a failure just because it used different words. Semantic similarity (comparing meaning via embeddings, or using a model-based judge to rate whether the output captures the same content as a reference) is the appropriate metric there. Using the wrong tool for the task in either direction produces a misleading score: exact match on a summarization task looks artificially terrible; semantic similarity on a data-extraction task looks artificially fine even when the extracted value is subtly, importantly wrong.

Latency percentiles: why the average lies

Average (mean) latency is one of the most commonly reported and least useful numbers in AI system monitoring, because it's dominated by the common case and hides the tail. A system with a 500ms average latency can still be delivering a multi-second stall to 5% of users, and that 5% is often disproportionately the users hitting the hardest, longest, most context-heavy requests, exactly the interactions where a slow response is most noticeable and most damaging to trust. Percentile metrics fix this: p50 (median) tells you the typical experience, p95 tells you what the slower-than-typical fraction of users see, p99 tells you the near-worst case. A team that only tracks average latency can ship a change that doubles p99 while barely moving the average, and have no idea it happened until support tickets about 'the AI is slow sometimes' start piling up.

Which metric actually matters, by use case

Use casePrimary metricWhy
Fraud/anomaly detectionRecall (with acceptable precision floor)Missed positives are far more costly than false alarms needing manual review
Document data extractionExact match (field-level)There's one correct value; near-misses are still wrong
RAG customer-facing Q&AFaithfulness/groundedness + retrieval accuracy separatelyDistinguishes a retrieval problem from a generation problem
Summarization/draftingSemantic similarity or model-judged qualityMany valid phrasings exist; exact match unfairly penalizes good output
Real-time conversational agentp95/p99 latency, not averageTail latency is what users actually notice and complain about
Content moderation/auto-publishPrecisionA wrong auto-published action costs more than a held-for-review item
Use case to primary metric mapping

How to actually pick the right metric for your system

Start from the failure mode that would actually hurt, not from the metric that sounds most rigorous. Ask what a false positive costs versus what a false negative costs, that answer alone tells you whether to weight precision or recall. Ask whether your task has one correct answer or many valid phrasings, that tells you exact match versus semantic similarity. Ask whether your system retrieves source material at all, if it does, faithfulness deserves to be tracked separately from overall accuracy, because it tells you which half of the pipeline broke. And track percentiles, not averages, for anything with a latency requirement, because the tail is where users actually notice. A single project usually needs two or three of these metrics tracked together, not one; picking just one, however rigorous it sounds, guarantees a blind spot somewhere the metric wasn't designed to see.

Frequently asked questions

What's the difference between precision and recall?

Precision measures how many of the system's positive flags were actually correct; recall measures how many of the true positives the system actually caught. They trade off against each other, and the right balance depends on whether false positives or false negatives cost you more in your specific use case.

What's the difference between hallucination rate and faithfulness?

Hallucination rate measures fabricated or wrong claims in output generally. Faithfulness (or groundedness) specifically measures whether a RAG system's answer is supported by the documents it retrieved, which separates a retrieval problem (wrong source material) from a generation problem (unsupported claims added on top of correct source material).

When should I use exact match instead of semantic similarity?

Use exact match for tasks with one correct answer, data extraction, classification, factual lookups. Use semantic similarity for open-ended tasks like summarization or drafting, where many different phrasings can all be correct and exact match would unfairly penalize good output.

Why is average latency a misleading metric?

It's dominated by the common, easy case and hides the tail. A system can have a fine average latency while a meaningful fraction of users, often on the hardest requests, experience multi-second stalls. Tracking p95 and p99 catches what averages miss.

How many eval metrics does a typical AI system need?

Usually two or three tracked together, chosen based on the specific failure mode that's costly for that use case, not one metric picked because it sounds the most rigorous. A single metric, however well-chosen, tends to leave a blind spot.

MM

Founder & CEO, Aiporate

Mert founded Aiporate to close the gap between AI adoption and AI-native capability. He writes on how organizations should reorganize around AI, and on what it actually takes to hire, vet and ship AI talent.

Need the team to make this real?

Describe your need in plain English, get the exact hire, forward-deployed talent or a fractional leader, vetted and matched in 72 hours.

Scope your need →

Keep reading

The Weekly Brief

Intelligence for building AI-native organizations.

One email a week: the sharpest thinking on AI hiring, infrastructure, teams and strategy, for the people building the future of work.

Join operators, founders and CTOs. No spam, unsubscribe anytime.