Survey designer
What it does
Survey design is measurement work. You start from what you need to know, decide what would count as evidence for it, and only then write a question, because wording and response format together decide what the data can support. Most of the damage happens early: a double-barrelled stem, a leading phrase, a scale with nowhere neutral to land, an order that primes the answer before it's given. Academic work calls the same craft questionnaire design or instrument design, and the failure modes carry over unchanged.
This skill starts from your goal in your own words: validate a feature hypothesis, test a concept, take an NPS or CSAT reading, follow up an interview round with a much larger sample. It picks each response format to fit the analysis you'll want to run, then writes the question so it survives that analysis. The rationale sits underneath every item, so you can argue with the choice. Routing and screening logic come with it. The design is written backwards from the analysis, which is the part a cold prompt skips.
What you get back
- The question set in running order, with the response format for each item and the reason it's that format.
- Scale points written out in full, including the neutral point and the escape option where one belongs.
- Screening questions and skip logic between sections.
- An analysis plan mapping each question to the part of your goal it answers, and what a result would need to look like to count.
- The items most likely to be misread, with an alternative wording for each.
- A pilot checklist to run before it goes out.
How to use it
Example Prompts:
"I need a survey to find out whether people would use scheduled exports. Twelve questions maximum."
"Rewrite question 4, it's double-barrelled and I can't split the results."
"Turn these five interview findings into a survey that tests whether they hold across the whole base."
What it won't do
It won't tell you your sample is representative. Instrument quality and sampling are separate problems, and a clean survey sent to the wrong list still returns a clean answer to nothing. It also won't analyse the responses when they land. The plan for that analysis comes with the survey, and reading the result stays with you.
Questions
Will it keep NPS and CSAT in their standard form?
Yes. Both keep their canonical stem and scale so your numbers stay comparable with previous rounds and with outside benchmarks. If you ask for a modified version, it makes the wording change explicit and tells you what comparability you lose.
How long is too long?
Length costs you completion, and the cost lands hardest on the people you most want to hear from. The skill argues about length: ask for forty questions and it will tell you which ones do no work and what you could learn from an interview instead.
Can it design a concept test with images or prototypes?
Yes. It writes the exposure order and the questions around the stimulus, and you supply the material itself. It will also flag where a concept test needs a control condition to be worth running.
