LEARN

How to Build a Question Pool

A rigid basal framework with fixed nodes, alongside separate loose particles accumulating into an expanding cluster, embodying the question set's division into fixed monitoring and continuously expanding topic libraries
A question set should first converge into decision problems, then split into a fixed monitoring library and a continuously expanding topic library; they cannot be mixed

IN ONE SENTENCE

A question pool is not a keyword list: collapse surface phrasings into decisions, then split into a frozen monitoring pool and a growing topic pool.

A question pool is not a keyword list. Collapse many surface phrasings into a limited set of decision questions, then layer them by buying stage. After that, the monitoring pool stays frozen while the topic pool keeps growing — the two must stay separate.

OUR POSITION

Merging the two pools is the most common failure. Move the monitoring pool and the mention-rate denominator moves with it, making periods incomparable; freeze the topic pool and content planning runs dry. It has to be two pools, not one pool used two ways.

01

How to collapse them

Collect surface phrasings first: real user phrasings, related questions in search results, and questions sales keeps hearing. This step wants volume, not judgement.

Then merge by shared decision. Different phrasings asking about the same decision become one meta-semantic question.

Then layer by buying stage: awareness, interest, decision, purchase.

02

Which questions stay out of the monitoring denominator

Definitional and mechanism questions: in testing these show the lowest brand hit rate, so including them permanently depresses mention rate. Their value is in being cited — keep them in the topic pool.

Questions naming a brand in the prompt: asking 'A or B' steers the answer toward those two, and naming a competitor hands them the mention.

03

Size and maintenance

The monitoring pool does not need to be large, it needs to be frozen. Thirty-five to forty questions keeps the denominator clean and month-to-month comparable.

One pool per language. Phrasing structure and citation logic differ, and a translated question is often not what locals type.

Log every change: which questions, when, and the previous definition. Without that record no later trend can be explained.

Sources

  1. [1]GEO: Generative Engine Optimization．Aggarwal et al., KDD 2024．2024

Updated 2026-08-10