Skip to content

Metrics

Citation Rate: Definition, Calculation, and Limitations

Citation rate is the share of questions whose answers link clickably to this site; cited questions over total questions.

Author
Xingge
Published
2026-11-09
Updated
2026-11-09
Level
L2 Reproducible
Version
1.0

Term details

Name
Citation Rate: Definition, Calculation, and Limitations
Type
Metrics
ID
citation-rate-en

Citation Rate: Definition, Calculation, and Limitations

Citation rate measures, within a fixed question set, the share of questions whose AI answer contains a clickable link to this site. It is calculated as cited questions divided by total questions and recorded weekly, to observe citation changes for the same question set over time.

Definition and calculation standard

The denominator is the total number of questions included in the statistics, and the numerator is the number of questions where a clickable link to this site appears at least once. If a question shows a link multiple times, the numerator still counts once.

ItemValue
Metric nameCitation rate
FormulaCited questions divided by total questions
UnitPercentage
Recording frequencyWeekly
Judgment unitSingle question
Record fieldsQuestion number, model, time, cited or not, link address, screenshot

Collection and judgment

  1. Fix the question set. Use the same questions and original wording each period, with no last-minute additions or removals, so the denominator does not shift.
  2. Fix the model and entry point. Record the model name, version or interface label, and whether web access is on.
  3. Fix the time window. Ask questions in the same period to reduce the effect of load and policy fluctuations.
  4. Keep a trace for each item. Save the answer text or a screenshot for each question, and record whether the link is clickable and points to this site.
  5. Cross-check in pairs. Have a second person review items judged as cited, and keep both sides' reasoning when they disagree.

Data sources and record fields

Data comes from a manual or semi-automated question log, not from undocumented platform interfaces. Fields include question number, the question as asked, model, time, cited or not, link address, screenshot path, and judge.

This site's current question set is 30 questions covering 6 entry points: DeepSeek, Kimi, Doubao, ERNIE, Tencent Yuanbao, and Qwen. Calculate per model first, then aggregate.

The log should match the evaluation table in the checklist index, so the same metric does not carry two sets of fields.

Limitations and misreadings

  • Platform redesigns change question entry points and display formats, so the standard drifts easily; record the interface version.
  • Different sampling times cannot be compared directly; cross-period comparisons must use the same question set.
  • A citation is not adoption. A link appearing only shows it was used, not that the conclusion is correct.
  • Citation rate depends on the difficulty of the question set, so different sites are not directly comparable.

Relationship to other metrics

Citation rate and mention rate complement each other: mention rate includes appearances without links, while citation rate counts only clickable links. Combined with the primary answer rate, you can separate being cited from being treated as the conclusion.

For judgment details, refer to the evaluation method in the method index; for term standards, see the metric index and glossary index.

Related sections: metric index, method index, glossary index, research index, checklist index.

Evidence

Author

Xingge

GEOA method research, focused on how Chinese LLMs cite and describe brands accurately.

Entity https://shingle.com.cn/#person

Author profile