Which question intents does KI Score distinguish?
Recommendation, comparison, problem, information: how KI Score classifies the questions in a question set by intent and weights them.
Last updated last month
For the KI Score, every question in a market’s question set is assigned one of four intents. This article explains what those intents are, how much each one counts, and why your own monitoring works without this classification.
Four question intents
Each question in the question set gets exactly one intent:
| Intent | What the question is about | Weight |
|---|---|---|
| Recommendation | The person wants a brand, product, provider, or tool recommended to them. | 5 |
| Comparison | The person is comparing options or alternatives, or asking which is better. | 2 |
| Problem | The person has a specific problem or is looking for help with an error. | 2 |
| Information | A factual question, for example about price, process, duration, or terminology, with no intent to get a recommendation or comparison. | 1 |
Only the KI Score team can see the question set and the intent of individual questions internally. No customer role has access to them, not even workspace admins.
Why recommendation questions count more
Someone who asks an AI for a recommendation is often close to making a choice. If your brand is named in that kind of answer, it matters more for your visibility than a mention in a general factual question. That is why the KI Score weights answers by the intent of the question.
The weights apply to your mention rate and to the Share of AI Voice in your market, among other things. On the KI Score page, the info text for the Erwähnungsrate im Zeitverlauf (mention rate over time) card says that recommendation questions count five times as much as purely informational questions. Comparison and problem questions count twice as much as informational questions.
To see how the weights feed into the KI Score formula, read How the KI Score is calculated.
How the intent is determined
Two methods determine a question’s intent together: a language model that reads the question, and fixed rules that look for typical phrasing. The weight follows from their results:
- Both methods agree: The question gets that intent. A question only receives the full weight of 5 if both methods classify it as a recommendation.
- The methods disagree: Until the KI Score team has reviewed the question, the lower of the two weights applies.
- The language model returns no result: If the rules detect a recommendation, the question is weighted as an information question until it has been reviewed.
This keeps an uncertain classification from artificially inflating the score.
No weighting by intent in monitoring
The intent classification applies only to the KI Score. Your own prompts in monitoring have no intent label. There is no filter or weighting by intent there, and the mention rate in monitoring is unweighted.
To organize your own prompts, use topics and tags. If you want to mark recommendation questions, for example, you can give those prompts a tag of your own.
What teams say about KI Score
“We finally see week by week where our brand shows up in ChatGPT and Perplexity.”
“The score finally gave us one number the whole team actually understands.”
“Instead of guessing which prompts matter, we now have real data.”
“Within the first week we already saw which competitor was beating us in AI answers.”
“Setup took one afternoon, not weeks of onboarding.”
“Finally a tool that takes AI Overviews as seriously as classic SEO.”