Decision questions and informational questions: why the difference matters

Decision questions ask an AI to choose or name products. Informational questions ask it to explain, and a good answer may name none. Measurement must keep them apart, because mixing them waters down recommendation figures and hides a different, useful signal.

What is the difference in practice?

Compare two questions a marketer might ask.

• "Which AI visibility tool should a small agency use?" This is a decision question. A good answer names products.
• "How do AI assistants decide what to cite?" This is informational. A good answer explains a process, and naming any product would be a side note.

Same subject, different job. The first asks the AI to pick. The second asks it to teach.

Why does it matter for measurement?

Imagine measuring how often an AI recommends a brand, over a set of questions that is half decision and half informational. On the informational half, the AI correctly recommends nobody, because nobody was asked for. That half contributes zeros for every brand.

The result is a recommendation figure that is lower than it should be and that depends on the mix. Change the proportion of informational questions and every brand's number moves, though nothing about the brands changed. Two reports with different mixes cannot be compared.

How does Citewake's published method treat them?

It tags every scored question before it is ever run, with three tags: intent, context and type. The type is either decision or informational. Then:

• Recommendation measures are computed over decision questions only. These are the recommendation rate, the rate at which a brand is the expressed first choice, and the brand's share of all positive recommendations.
• Informational questions are reported separately, as how often brands appear when no product was asked for.

Tagging in advance prevents the temptation to sort questions to suit the results.

What does each type tell you?

Decision questions tell you whether an AI will pick you when someone is ready to choose. They are the closest thing to the moment of purchase.

Informational questions tell you whether you are part of the picture earlier. A brand that appears in answers to "how do I do X" is shaping how the buyer thinks about the problem, and may be remembered when the buyer later asks for a recommendation. This is not the same as being recommended, and it should not be reported as if it were.

A brand can be strong in one and weak in the other. Some brands appear in how-to answers but are not chosen when people want a product. Others are chosen when asked but are invisible in the explanations that come first. Each pattern calls for different work.

Do informational questions ever lead to products?

Often. Many answers to how-to questions end with a suggestion or a short list of tools, even when the question did not ask for one. That is worth recording, since it shows a brand being named in a setting where the buyer may not yet be shopping. But it is a mention in passing, not a response to a request for a choice, and the method counts it that way: informational results are reported as appearances, apart from recommendations.

How can you tell which type a question is?

Ask what a helpful answer would look like.

1. If it would be a short list of products with reasons, it is a decision question.
2. If it would be an explanation, a method or a definition, it is informational.
3. If it could go either way, rewrite it so it clearly does one thing.

Words such as "best", "which", "compare", "alternatives" and "recommend" point to decisions. Words such as "how", "why" and "what is" point to information, though not always. Read the question as a person would.

Intent tags help. The method's intents are recommend, compare, implement, evaluate price and learn. Recommend, compare and price questions usually sit on the decision side. Learn questions usually sit on the informational side. Implement questions can be either, depending on whether the buyer wants a tool or a method.

What happens to comparisons between brands?

They get fairer. When two brands are compared on decision questions only, the comparison asks the question that matters to buyers: who gets picked when a pick is requested. Informational results are shown alongside, so no one is hiding the other half.

This also helps with reading change. If a brand's recommendation rate falls while its informational appearances rise, you can see that the story is not simple. How to read an AI answer like an analyst shows how to look at the detail behind figures like these.

What if a question is mixed?

Some questions ask for an explanation and a choice at once, such as how to do a task and which tool to use. These muddy the data. The cleaner approach is to split them into two questions, one of each type, and tag each. If you must keep a mixed question, tag it by what the buyer most needs, and be consistent from month to month.

How can you apply this to your own tracking?

1. Sort your list of questions into decision and informational.
2. Report your recommendation rate over the decision questions only.
3. Report appearances on informational questions separately.
4. Keep the sorting fixed, so month-to-month comparisons hold.

How to build a benchmark set of questions covers the wider design of the set, and the three ways AI names a brand explains how a mention differs from a recommendation.

Where can you read the full method?

On the methodology page, described as a method ahead of the first published edition in early 2027. To see how assistants treat your own site today, try the Citewake checker.