7 · Documentation
Questions
The nine question types, which count towards presence, and the principles for a question set that yields reliable numbers.
The questions are the measuring instrument. A poor question set yields precise numbers about the wrong thing. This page records the question types and the principles by which augenmerk checks questions, in the input module as you type and for the whole set.
Nine question types
| Type | Example | Counts towards presence | Delivers |
|---|---|---|---|
| Discovery | "Who is a good provider of X?" | yes | presence, new brand names |
| Ranking | "What are the best X?" | yes | presence, position |
| Recommendation | "Which X would you recommend?" | yes, double | presence as recommendation |
| Attribute | "Which X are considered sustainable?" | yes | attributes the AI links to brands |
| Comparison | "Compare A and B" | no | readable side-by-side |
| Reputation | "What are the drawbacks of X?", "Who should not buy X?", "Is X worth the price?" | no | purchase check: topics, brand image |
| Strengths/weaknesses | "What is positive and negative about X?" | no | topics, brand image |
| Need | "What does diesel cost in Hamburg today?" | no | whether the AI takes on a task |
| Perception | "What is X known for?", "Why not X?" | no | brand image |
Ranking is the only type from which a position arises. A mere enumeration does not order. Recommendation counts double: whoever is named here is recommended, not merely listed.
What counts
Only questions with no brand in the wording and no segment form presence. Segment questions count in the segment. Questions that name a brand deliver topics and brand image. Need questions are evaluated separately. See Presence and position.
Principles
These rules come from measured question sets, not from assumptions. Where a rule can be checked in the wording, the application checks it as you type.
Name compulsion
Without the word brand, manufacturer or provider in the wording, a question yields no brands but varieties. "Which rice is best for risotto?" produced not a single brand in eleven measurements; "Which rice brand would you recommend for risotto?" named the measured brand in 36 % of answers. This holds for every industry with generic terms: rice, flour, oil, mattresses. For products that are brands themselves, such as apps, the rule is uncritical.
One deliberate exception: a single open recommendation question without name compulsion per project. It measures whether the AI names brands at all on its own, and lowers the values of all brands in doing so. Hence exactly one, with a history of its own.
Ambiguity belongs out of the wording: rice and travel, rice pudding as dessert or raw rice, "best pizza" as restaurant or frozen product. The cue to the industry sits in the question text, never only in the project settings.
Type and wording match
"… is recommended?" is a recommendation, not a ranking question. Set up as ranking, the judge invents an order from one to three names. Ranking questions ask for "the best", "name the five best", "which has the best …". Questions whose order comes from third parties, such as test results, are attribute, never ranking. Plural yields lists and thus usable positions; singular overstates place 1.
Segments
Every question that names a product segment gets a segment and counts only there. A segment question in the overall value shifts the ranking in favour of whoever occupies the niche. No segment in which your own brand does not compete; no segment with only one question. Need questions stay without a segment, otherwise they appear in no view and still cost.
Comparison and reputation
A comparison without your own brand yields nothing countable. One comparison per project, with your own brand, as a readable raw answer. A reputation question "Is X good?" measures only sentiment, and sentiment in AI answers is constantly positive. Reputation questions pay off as a purchase check (below): drawbacks, "not for whom", worth the price. Topics come from questions that provoke a verdict: strengths/weaknesses, perception and purchase check.
Redundancy
Questions that yield the same thing are one question. Three discovery questions with 93 to 100 % share of the same brand bring no third observation but cost three times. The question that yields the most other brands per answer stays. After two runs the questions view shows such pairs as a hint.
Character-identical
Retained questions stay character-identical. The evaluation recognises a question by its wording; every rewording is, from the next run, a new question with a new history. Reword only when the old history consists of zeros anyway. Deactivating stops the cost and deletes nothing.
Cost
Every active question costs one measurement per channel per run. Brands cost nothing, with one exception: strengths-and-weaknesses questions name the brand in the question text, so every brand asked about that way costs a question of its own. There is no per-question frequency; every active question runs in every run.
The blueprint of a question set
This is how onboarding builds a question set from brand, industry, range and competitors. First the positioning — target groups, price tier, buying criteria, buying occasions — then the questions by block:
| Block | Number | Type | counts | in the test |
|---|---|---|---|---|
| Open recommendation without name compulsion, with a buying occasion | 1 | recommendation | yes | yes |
| Recommendation with name compulsion and target group or price tier | 1 | recommendation | yes | yes |
| Name collection of the market, plural, no qualifier | 1 | discovery | yes | yes |
| Ranking questions, one buying criterion each | 3 | ranking | yes | 2 |
| Attributes ("Which … brands …") with further criteria | 2 | attribute | yes | 1 |
| Per core segment recommendation and ranking, with segment word and occasion | up to 4 × 2 | — | in the segment | 2 |
| Strengths and weaknesses of your own brand (fixed wording) | 1 | perception | no | yes |
| Need before the purchase decision, without asking for brands | 2 | need | no | 1 |
| Comparison with the strongest competitor | 1 | comparison | no | no |
| Purchase check: drawbacks, "not for whom", worth the price | 3 | reputation | no | no |
| Competitor perception: strengths/weaknesses of the three strongest competitors | 3 | strengths/weaknesses | no | no |
That makes 26 questions, ten of them in the free test. Around twelve counting questions give a noise band near ±8 points; for ±5 points it would take around 32 counting questions. That is a question of budget, not of questions. See Noise band.
Purchase check
Most tools measure whether a brand is recommended. Buyers also use AI for the cross-check once they have a brand in mind: "What are the drawbacks of X?", "Who should not buy X?", "Is X worth the price?" According to a US consumer survey from July 2026, a chatbot has already talked more than half of AI users out of a purchase. That is why these three questions belong in every question set. They name the brand, do not count towards presence, and deliver topics with evidence and source into the brand image — where it shows whether the AI advises against, with what, and from which source.
Competitor perception
The topics view compares brand against market. For that it needs one judgement-provoking question per brand: for your own brand the perception question, for the three strongest competitors "What is especially positive and negative about X?" each. Without these questions the topics view would stay empty.
Questions from Google queries
With Search Console connected, augenmerk derives further questions from the most-seen queries of the website. Queries are bundled by need — synonyms such as petrol prices, fuel prices and pump prices are one group; "near me", a specific city, a product type, a point in time are separate groups — and each group yields one question in a person's wording. The principles on this page apply here too: no brand in the question, no duplicate of the existing set, no question from too few searches. Every suggestion carries its origin: the queries and their impressions. Activation is manual. See The application.
The check in the application
The input module "Manage questions" checks every question against these rules as you type and reports three levels: error (type and wording contradict each other, need with segment), warning (segment term without segment, segment with only one question) and hint (no name compulsion, similar question present, comparison without your own brand, strengths-and-weaknesses question about a brand that is not tracked). The same rules run across the whole question set and appear in the questions view under "How well is the question set built?". They warn, they do not block.
Updated: 2026-09-16