The questions decide the result

• You wrote twenty questions in ten minutes and none of them sounds like what a real customer types.
• Your presence is 80 per cent, and that is because fourteen of twenty questions contain your name.
• Next quarter you changed three questions and can no longer compare with what came before.

Whoever picks the questions picks the result. That is why questions get audited.

1. 1

   Six intents, not one

   Direct brand, category discovery, competitor comparison, purchase decision, product questions and trust questions. Each answers a different moment, and none alone describes your visibility.

2. 2

   The split gets validated

   Before the set is accepted we check that no intent eats the split, that there are no duplicate questions, and that the ones containing your brand stay inside the expected band. If they do not, you are warned before spending.

3. 3

   It is frozen and versioned

   When you launch, the set is frozen with its version number. Changing a question creates a new version instead of editing the old one, so two quarters remain comparable, or the dashboard tells you they are not.

• The set with each question's intent written beside it.
• The actual split against the recommended one, with the gap flagged.
• The estimated cost of the sweep before you launch it, not after.

• Consultants who must justify why those questions and not others.
• Teams repeating the measurement every quarter who need it to be the same one.
• Anyone who already has their list in a spreadsheet and wants to bring it over without retyping.

It is on every plan, trial included: it is the part of the product that consumes nothing because it calls no model. What the plan limits is how many questions fit in a set and how many active sets you can keep at once.

See the full pricing

How many questions should my set have?

Between 20 and 40 covers a normal-sized category well. Below 10 the denominator is so small that one changed answer moves the figure several points. Above 60 the cost rises and the new information falls: most questions 41 to 60 say the same as the first twenty.

Can I include questions with my brand in them?

Yes, and you should: knowing what AI says when asked about you is valuable. What you should avoid is them being the majority, because then the presence rate measures whether the model can read your name, not whether it recommends you. The system flags that skew instead of letting you discover it in the meeting.

Are the questions sent to the model exactly as written?

Exactly as written, with no added context about your brand. Adding it would be handing over the answers before the exam: the model would mention what we just told it and the measurement would be worthless. The only things set apart are language and country, and those are recorded in the sweep configuration.

What if I want to change the set halfway through the year?

You can, and sometimes you must because the business changes. What you cannot do is pretend nothing happened: the new version starts a new series, and the chart marks the break. Comparing before and after a set change is comparing two different exams and calling it progress.