Key Takeaways
To build a questionnaire with ChatGPT, first state who you are asking and which decision the answers should inform. Request questions that address one concept at a time without suggesting an answer. Check the response options and routing, then use a small trial completion to learn how real readers interpret the draft.
This guide is useful for collecting feedback after one practical online session. The example is fictional: it describes no real course and contains no participant responses.
Connect the questionnaire to one decision
The goal could be to improve the exercise in the next session. For that, you need to understand whether the instructions were clear, how much of the exercise participants completed, and whether the allotted time was sufficient.
This questionnaire does not measure the instructor’s popularity or acquired knowledge. Completing an exercise does not by itself establish that someone has developed the skill independently.
OpenAI’s prompting guide recommends describing the goal, context, output, and boundaries (source 1). Here, the boundary is specific: prepare the questionnaire without inventing answers.
Prompt: five questions with explicit routing
Give ChatGPT this task:
Draft a five-question questionnaire for participants after one completed
practical online session. The goal is to improve the next session's exercise:
instruction clarity, completion, and allotted time.
Use IDs Q1–Q5. Each question should address one concept.
Do not assume people liked the session or everyone completed the exercise.
For closed questions, provide complete answer options and specify one choice.
Allow a response when someone cannot assess the item.
Q1 checks attendance: "No" ends the questionnaire.
Q2 concerns instruction clarity.
Q3 concerns exercise completion; "I did not attempt it" skips Q4.
Q4 concerns the time allotted for the exercise.
Q5 is an optional open question about a possible change.
Address one reader directly. Do not request a name or email address.
After the draft, explain the routing rules and remaining uncertainties separately.
Do not create survey findings, publish the questionnaire, or send it to anyone.
Five questions define this example’s scope; they are not the correct number for every survey. If you are making a different decision, change the goal first and reconsider which questions are necessary.
Find these problems in the first draft
Pew Research Center’s methodology guide emphasizes clear wording, one concept per question, and response options that do not overlap (source 2). These rules also help you review a ChatGPT draft.
| Draft problem | Why it is ambiguous | How to clarify it |
|---|---|---|
| “How clear and useful was the exercise?” | It asks about clarity and usefulness together | In this questionnaire, assess instruction clarity only |
| “How much did you like our excellent instructions?” | It describes the instructions positively in advance | Ask how clear the instructions were |
| Only “Good / Very good / Excellent” | It leaves no room for a different experience | Include unclear, middle-rating, and unable-to-assess options too |
Changing a word is insufficient if the question measures something else. For example, “I liked it” and “I completed it” are different answers.
Reference questionnaire for one online session
The following questionnaire was prepared as a reference for this article, not as a transcript of an actual ChatGPT response. Before real use, put the session’s name and date in the questionnaire title.
The questions concern the single online session identified in the title.
Participation is voluntary. Choose one answer for each closed question.
Q5 is optional. Follow the routing instructions for the other questions.
Q1. Did you attend this session?
A. Yes → Q2
B. No → End of questionnaire
Q2. How clear were the exercise instructions?
A. Very unclear
B. Unclear
C. Neither clear nor unclear
D. Clear
E. Very clear
F. I did not review the instructions
After any answer → Q3
Q3. How much of the exercise did you complete during the session?
A. I completed all of it → Q4
B. I completed part of it → Q4
C. I attempted it but could not complete any of it → Q4
D. I did not attempt it → Q5
Q4. How would you assess the time allotted for the exercise?
A. I did not have enough time
B. I had enough time
C. I had more time than I needed
D. I cannot assess it
After any answer → Q5
Q5. What, if anything, would you change for the next session?
Open response, optional. The questionnaire ends after this.
Q2’s middle rating and “I did not review the instructions” are different answers. The latter avoids forcing someone to rate clarity. Likewise, Q4’s “I cannot assess it” does not mean the time was sufficient.
Q3 separates “I attempted it but could not complete any of it” from “I did not attempt it.” In the first case, a time assessment may be useful; in the second, Q4 is skipped. Partial completion also has its own option.
Check the routing with fictional paths first
When building a form, copying the questions is insufficient: configure the routing too. Then walk through five reference paths:
- Choose “No” in Q1: Q2–Q5 should not appear.
- After confirming attendance, choose “I completed all of it” in Q3: Q4 should appear.
- Choose “I completed part of it” in Q3: Q4 should still appear.
- Choose “I attempted it but could not complete any of it” in Q3: Q4 should appear in this case too.
- Choose “I did not attempt it” in Q3: you should move directly to Q5.
Leave Q5 blank on every applicable path and confirm that the questionnaire can finish. This is a trial walkthrough, not real participant data. If you cannot configure a route in your form tool, do not treat it as verified.
Not requesting a name does not itself establish anonymity. Check the actual form’s account and response-collection settings before making a promise to participants.
Check the wording with real readers too
Pew’s guide also emphasizes pretesting (source 2). Use a small trial completion to see what readers understand a question to mean and which answer they cannot find.
Ask a few people similar to your audience to work through the questionnaire aloud: how they interpret “part,” when they would choose “I cannot assess it,” and where choosing becomes difficult. This helps find problems; it does not by itself establish scientific validity.
Q5 follows detailed questions in this example, so its answers already have the context of instructions and practice. If you need unaided recall of general problems, try the open question after the attendance check and before the details. The same methodology source discusses order effects (source 2).
When to keep an existing question or choose a conversation
If comparing results across sessions, record every change to wording and answer options. Avoid rewriting the questionnaire each time merely to make the text sound better.
If you do not yet know what is getting in a participant’s way, a short conversation may be more useful than predefined rating options. Ask ChatGPT for alternative wording, but make the final choice based on your goal and trial completions.
After collecting responses, keep the questionnaire and its version so you know what participants answered. The feedback-analysis guide can help you group the collected text and connect conclusions to quotes.
Full source addresses
To open a source, copy its full address into your browser.
- OpenAI, Prompting:
https://learn.chatgpt.com/docs/prompting - Pew Research Center, Writing Survey Questions:
https://www.pewresearch.org/writing-survey-questions/
Updated: October 19, 2026



