In short
AI engines generate every answer from scratch, with an element of randomness, sources that shift and models that get updated. That’s why a single question doesn’t work as a measure. Grouping questions by topic, repeating them across several engines and combining several analyses gives a stable number.
Three reasons answers vary
- Generation involves randomness: models pick each word from several likely options. Two answers to the same question are rarely identical.
- Real-time search: when the engine searches the web, the pages it finds change with the day, the country and the language.
- Model updates: providers change models and rules without notice, and that changes which brands appear.
The most common mistake
Asking one question, seeing whether the brand shows up (or not) and drawing conclusions. With a single answer, you can’t tell a trend from a fluke. It’s like judging a campaign by a single click.
How we measure at Neural Ranks
- Topics, not promptsWe group 5 questions with the same intent, written the way customers ask them.
- Several enginesWe ask each question on ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews.
- Several analysesThe numbers combine the last three analyses, which smooths out the chance of any single day.
- Confidence levelEvery metric says whether the evidence is strong, moderate or weak.
Rule of thumb: only react to a change when it holds across more than one analysis and shows up on more than one engine.
You can read the full method in the lesson Measure by topic, not by prompt.
Key points
- A single answer is not a measure.
- Measure topics, across several engines and several analyses.
- Check the confidence level before you decide.
Go from theory to numbers.
See how AI talks about your brand on every topic.