How to write useful survey questions with AI

Give every question a purpose

Start with the decision the answers should support. For every question, finish this sentence: “I will use this answer to…” If there is no clear ending, consider removing the question.

Tellseed defaults to ten generated items and ten shown in a round. Ten is a starting point, not a target to fill at any cost. Ask for fewer if the decision only needs a few answers, or a longer questionnaire when the extra work has a clear purpose.

Remove leading wording and double questions

“How much did you enjoy our excellent service?” assumes a positive experience. “How would you describe your experience?” leaves the answer open. Give the agent examples of the difference and ask it to explain proposed revisions.

“Was the booking process fast and easy?” asks about two things. Someone may find it easy but slow. Separate the questions or choose the dimension that matters to your decision.

Avoid asking people to recall details they are unlikely to remember. A question about their most recent visit is often easier to answer than one about every visit in the last year.

Match the answer format to the task

Free text is useful for explanations and unexpected issues. Classification buttons are useful when the allowed categories are known. A fixed set of labels makes counting straightforward but can hide answers that do not fit.

For a classification campaign, define each label and decide whether Other or a skip option is needed. For open questions, use plain wording and avoid asking for long essays when a sentence would do.

Name and email fields are collection settings, not a reason to add contact requests to every question. Both are disabled by default and optional when enabled. Explain the purpose when collecting contact information.

Try the full round with a fresh reader

Read the contributor page on a phone. Check the introduction, answer controls, question order, progress and completion message. Ask someone unfamiliar with the project to tell you what they think each question means.

In Tellseed, review the draft through the preview and approval links. After human approval, ask the agent to publish. Submit a test round and check the answers view before inviting your full audience.

If the pilot reveals an unclear question, revise it and record the change. Avoid combining results from materially different wording without acknowledging the difference.

Keep interpretation tied to the evidence

A campaign can collect real answers without producing a representative sample. Report who was invited, who answered and how many responses you analysed. Treat missing answers and skipped questions as missing data.

When asking an agent for analysis, request the observations first, then its interpretation and suggested actions. Keep an export so you can inspect the underlying answers independently.