Metrics
Wording Accuracy Rate: Definition, Calculation, and Checks
Wording accuracy rate is the share of mentions whose name, title, and organization match the wording baseline.
- Author
- Xingge
- Published
- 2026-11-09
- Updated
- 2026-11-09
- Level
- L2 Reproducible
- Version
- 1.0
Term details
- Name
- Wording Accuracy Rate: Definition, Calculation, and Checks
- Type
- Metrics
- ID
- accuracy-rate-en
Wording Accuracy Rate: Definition, Calculation, and Checks
Wording accuracy rate measures whether the name, title, organization, and business scope used by AI when mentioning the target subject match this site's wording baseline. It is calculated as correct entries divided by mention entries, with item-by-item manual checking.
Definition and calculation standard
The denominator is the number of recorded mention entries, and the numerator is the number of entries whose description matches the baseline on every item. Any mismatch counts as an error, with no partial credit.
| Item | Value |
|---|---|
| Metric name | Wording accuracy rate |
| Formula | Correct entries divided by mention entries |
| Unit | Percentage |
| Recording frequency | Weekly |
| Judgment basis | Entity wording baseline (name, title, organization, business scope) |
| Required fields | Answer text, failed items, correction suggestions |
Collection and judgment
- Build the baseline. Write the name, title, organizational affiliation, and business scope as reference entries that serve as the judgment basis.
- Extract mention entries. Take the sentences containing the target subject from the mention records one by one.
- Compare item by item. Compare all four items separately and record exactly where they differ.
- Classify errors. Group them into name, title, organization, and business scope errors, to help locate the actions needed.
- Aggregate and review. Aggregate weekly and have another person review disputed entries.
Data sources and record fields
Data comes from comparing mention records with the baseline, with fields including question number, model, time, answer text, compared item, consistent or not, error type, and correction suggestion.
The baseline must record its effective date, and cross-period statistics are segmented by effective date.
Limitations and misreadings
- It relies on manual item-by-item checking, so the sample is limited and cannot cover all mentions.
- If the baseline itself lags, correct descriptions are misjudged as errors, so it needs regular updates.
- This metric looks only at whether the description is accurate, not at sentiment direction.
- Different models write titles in different styles; judge by the baseline, not by intuition.
Relationship to other metrics
Wording accuracy rate reflects perceptual accuracy, while mention rate reflects frequency of appearance. Observing both separates being mentioned from being described correctly.
Related sections: metric index, method index, glossary index, research index, checklist index.