Record the conditions.
Store the exact prompt, collection date, engine, locale and relevant settings alongside the observed response. Explain missing or excluded results.
Our methodology
Trust begins with
a clear definition.
A transparent framework for reading AI search data. Understand the sample, the calculation and the limits before making a decision.
Measure a defined
slice of the conversation.
AI search is not one fixed list of rankings. Answers may vary by question, model, time, locale and context. A useful report defines these conditions and records what was actually observed.
The definitions below describe the intended Glinty approach. They are a framework for the concept, not a claim that this website is collecting live data.
Make each number
explain itself.
| Metric | Definition | Example from a fictional sample |
|---|---|---|
| Brand mention rate | Answers naming a brand ÷ eligible answers34 / 50 = 68%||
| Domain citation rate | Answers citing a domain ÷ eligible answers12 / 50 = 24%||
| Recommendation rate | Answers explicitly recommending a brand ÷ eligible answers20 / 50 = 40%||
| Percentage-point change | Current rate − previous rate68% − 55% = +13 points
An answer can mention multiple brands. Mention rates across brands do not need to total 100%. Recommendation classification requires a documented rule and review of ambiguous answers.
Preserve a baseline.
Use a stable prompt set for comparisons. Annotate changes in collection method or sample composition instead of treating them as organic growth.
Keep the caveats close.
Disclose sample size and response variability. Distinguish correlation from causation when discussing content changes or business results.
Your next chapter in search
Great brands deserve
to be discovered.
Make the next answer a little more about you.