Survey Questions With AI: Write, Test and Analyse a Survey
Write better survey questions with AI. Get survey questions examples for customers, students and staff, then pilot, share and analyse responses step by step.
By the end of this workflow you'll have a short, unbiased survey built around one clear decision, tested on a handful of real people, shared through Google Forms and WhatsApp, and analysed with AI into a one-page summary of findings you can act on. You'll also get ready-to-adapt survey questions examples for customers, students and employees.
Time to complete: 2 to 3 hours of active work, plus a few days for responses to come in
Difficulty: Beginner
Tools needed:
Any AI chatbot (ChatGPT, Claude or Gemini; free tiers work)
Google Forms (free with a Google account) or a similar form builder
Google Sheets or Excel to hold the responses
WhatsApp, email or social media to share the link
Before You Start
Know who will act on the results. A survey nobody acts on wastes your respondents' time.
Check that a survey is the right tool. Surveys measure how many people think or do something, not why.
Keep personal data to a minimum. If you collect names, phone numbers or emails in Nigeria, the Nigeria Data Protection Act 2023 applies. The easiest fix is not to collect personal details you don't need.
Plan for short. Aim for 5 to 7 minutes and 8 to 15 questions.
Step 1: Define the Decision the Survey Will Inform
Good surveys start with a decision, not "let's find out what people think". What will you do differently depending on the answers?
I want to run a survey. Help me sharpen its purpose.
Context: [who you are and the situation, e.g. I run a small bakery in Surulere, Lagos, and sales of our weekday breakfast packs have dropped]
Who I'll survey: [e.g. customers who have bought from us in the last 3 months]
Ask me 4 questions to clarify the decision this survey should inform. Then write:
1. One sentence stating the decision
2. 3 to 5 specific things I need to learn to make it
3. Anything I should NOT try to learn with this survey because another method would be betterThe third item is useful: AI often points out that something is better answered by your own sales data.
Checkpoint: You have one decision sentence and no more than five learning goals. If you have more, you're running two surveys; cut back.
Step 2: Choose the Right Question Types
Each learning goal needs a question type that fits.
Question type | Best for | Example |
|---|---|---|
Multiple choice | Facts, behaviour, picking from known options | "How often do you buy breakfast on the way to work?" |
Likert scale (e.g. 1 to 5, strongly disagree to strongly agree) | Attitudes and satisfaction | "The delivery time was acceptable." |
Net Promoter Score (0 to 10) | Overall loyalty, tracked over time | "How likely are you to recommend us to a friend?" |
Open-ended | Reasons, ideas, surprises | "What's one thing we could do better?" |
Here are my survey learning goals: [paste from Step 1]
For each goal, suggest the best question type (multiple choice, Likert scale, NPS, rating or open-ended) and explain why in one sentence. Limit open-ended questions to 2 or 3 in total, because they take longer to answer and analyse.Checkpoint: Each learning goal has a matching question type, and you have no more than three open-ended questions.
Step 3: Write Clear, Unbiased Questions
Now draft the actual questions.
Write survey questions for these goals and question types: [paste from Step 2]
Audience: [e.g. Lagos office workers aged 22 to 40, most answering on their phones]
Rules:
- Plain English a secondary school leaver would understand
- One idea per question (no double-barrelled questions)
- Neutral wording that doesn't suggest a "right" answer
- Answer options that don't overlap and cover every case, including "Other" or "Not applicable" where needed
- Balanced scales with a neutral midpoint
- Put easy, non-sensitive questions first and demographics at the end
- Total of 8 to 15 questionsCheck the AI's draft yourself for three classic errors:
Leading questions
"How much did you enjoy our new menu?" assumes enjoyment. Use "How would you rate our new menu?" with a balanced scale instead.
Double-barrelled questions
"Was our staff friendly and fast?" asks two things. Split it into two questions.
Overlapping or missing options
Age bands like "18 to 25, 25 to 35" overlap at 25. Income bands that stop at ₦500,000 a month leave higher earners nowhere to go. Add "Prefer not to say" for sensitive questions.
Checkpoint: You have a draft of 8 to 15 questions, each asking one thing, with complete answer options.
Step 4: Get AI to Critique Your Survey
Before you show the survey to anyone, ask the AI to attack it.
Act as an experienced survey researcher reviewing this questionnaire. Here it is: [paste all questions with answer options]
Check for:
- leading or loaded wording
- double-barrelled questions
- overlapping or incomplete answer options
- jargon or words my audience may not know
- questions that don't serve the stated decision: [paste decision sentence]
- anything that could feel intrusive or collect personal data I don't need
List each problem, the question number and a suggested fix. Then estimate how many minutes the survey takes on a phone.Accept fixes that make sense; you know your audience better than the AI does.
Checkpoint: Every question has passed the critique or been fixed, and the estimated time is under 7 minutes.
Step 5: Pilot Test With Five Real People
AI can't tell you how a real person reads a question. A pilot test can. Build the survey in Google Forms, then send it to five people who match your audience. Ask where they hesitated or wanted an option that wasn't there.
Here is the feedback from 5 pilot testers on my survey: [paste their comments]
And here are my questions: [paste]
Summarise the problems they hit, rank them by how much they would affect the results and suggest the smallest change that fixes each one.Pilots often reveal a missing option (usually "I don't use this") or a survey that's slow on a weak connection. Delete the pilot responses before you launch.
Checkpoint: You've fixed every issue at least two testers raised, and the survey still fits in 5 to 7 minutes.
Step 6: Distribute the Survey
In Nigeria, WhatsApp is often the fastest route to responses. Use AI to write a short, honest invitation.
Write a WhatsApp message (under 70 words) inviting [audience, e.g. our customers] to complete a short survey.
Include: what it's about, how long it takes ([X] minutes), that answers are anonymous, what we'll do with the results and the link: [link].
Friendly but not pushy. Then write a shorter reminder message to send 3 days later.Share the link where your audience already is (class groups, customer broadcast lists, staff channels) and send one reminder, not five. If you offer airtime as an incentive, collect contact details in a separate form so answers stay anonymous.
Checkpoint: The survey is live, the invitation has gone out and you've scheduled one reminder.
Step 7: Analyse Responses With AI, Safely
Once responses are in, export them from Google Forms to Google Sheets. Then clean the data before it goes anywhere near a chatbot.
Privacy first. Delete any columns with names, phone numbers, email addresses or anything else that identifies a person. Under the Nigeria Data Protection Act, you're responsible for personal data you collect, and pasting it into a third-party AI tool is a form of sharing.
For numbers (multiple choice, scales, NPS), use spreadsheet charts or formulas. Spreadsheets do this well, and with ChatGPT for Excel, AI can write the formulas for you. To calculate NPS, subtract the percentage of detractors (scores 0 to 6) from the percentage of promoters (9 to 10).
For open-ended answers, AI is a big time saver:
Below are [number] anonymised open-ended responses to the question: "[question]"
1. Group them into 4 to 7 themes. For each theme, give a name, a count of responses and 2 short representative quotes copied exactly.
2. Note any responses that don't fit a theme but seem important.
3. Do not invent quotes or counts. If you're unsure where a response fits, say so.
Responses:
[paste responses, one per line]Chatbots sometimes miscount or merge themes, so spot-check at least 20 answers yourself. Our AI fact checker workflow has a good routine for checking outputs like this.
Checkpoint: You have charts for your closed questions, themes with verified counts for open-ended ones, and no personal data has left your spreadsheet.
Step 8: Turn Responses Into Findings
"62% chose option B" is a result. "Most customers want earlier delivery, so we should test a 7am slot" is a finding.
Here are my survey results: [paste summary numbers and themes]
The decision this survey was meant to inform: [paste decision sentence]
Write a one-page summary with:
- 3 to 5 key findings, each backed by a specific number or theme
- what each finding suggests we should do
- limits of the data (sample size, who didn't respond, possible bias)
- one recommended next step
Keep it plain and direct.Be honest about limits: forty responses from your own WhatsApp contacts tell you something, but they don't represent everyone. If you need to share the results more formally, run them through the steps to turn research into a clear brief for a manager or client.
Checkpoint: You have a one-page summary that answers your original decision, with clear limits stated.
Survey Questions Examples
Adapt these to your context.
Customer survey questions
How did you first hear about us? (Friend or family / Instagram / WhatsApp status / Walked past / Other)
How often do you buy from us? (Weekly / Monthly / Less than monthly / This was my first time)
How satisfied were you with your most recent order? (1 Very dissatisfied to 5 Very satisfied)
How likely are you to recommend us to a friend? (0 to 10)
What's the main reason for your score? (Open-ended)
Student survey questions
"The course materials helped me understand the topics." (Strongly disagree to Strongly agree)
Which resources do you use most? (Lecture notes / Textbooks / YouTube / AI chatbots / Study groups / Other)
How often do you use AI tools for coursework? (Never / Rarely / Weekly / Daily)
What's the biggest obstacle to your studies this semester? (Open-ended)
For final-year projects that rely on a questionnaire, plan the survey as part of your wider write-up; our essay outline workflow helps you structure the chapters around it.
Employee survey questions
"I understand what's expected of me in my role." (Strongly disagree to Strongly agree)
"I have the tools I need to do my job well." (Strongly disagree to Strongly agree)
How likely are you to recommend this company as a place to work? (0 to 10)
What's one change that would make your work easier? (Open-ended)
For employee surveys, anonymity is everything. In small teams, avoid demographic questions that could identify someone, such as department plus job title. Then turn the results into targets with our SMART goals examples workflow.
Worked Example: Funmilayo Surveys Her Bakery Customers in Surulere
Funmilayo runs a bakery in Surulere, Lagos. Her weekday breakfast packs (bread, egg and tea for about ₦3,500) had been selling less since the start of the term, and she wasn't sure why.
In Step 1, the AI helped her turn "find out what customers think" into one decision: should she change the price, the contents or the delivery time of the breakfast pack?
Her first draft included "Don't you think our breakfast pack is great value?" The AI critique in Step 4 flagged it as leading, and she replaced it with a five-point value-for-money scale. Her pilot test with five regulars revealed that nobody understood her "delivery window" question, so she rewrote it as "What's the latest time you'd want breakfast delivered?"
She shared the survey on her WhatsApp broadcast list of about 400 customers and got 87 responses in five days. She deleted the phone number column before pasting open-ended answers into a chatbot. The AI grouped them into five themes, the largest being "arrives too late" (31 responses, which she checked by hand). The one-page summary recommended testing a 6:45am delivery slot for two weeks before touching the price.
Common Mistakes
Asking everything you're curious about. Long surveys get abandoned. Cut any question that doesn't serve the decision.
Leading the respondent. Words like "great", "new and improved" or "don't you think" push answers in one direction.
Skipping the pilot. Five testers catch problems that no amount of AI review will.
Pasting raw data into AI. Remove names, phone numbers and emails first. It protects your respondents and keeps you on the right side of data protection rules.
Trusting AI theme counts blindly. Spot-check them against the raw answers.
Next Step
Pair your findings with a look at what rivals are doing using our competitor analysis with AI workflow, then package everything into a decision-ready document.
FAQ
How many questions should a survey have?
For most purposes, 8 to 15 questions that take 5 to 7 minutes on a phone.
What are good survey questions?
Good survey questions ask about one thing, use plain neutral wording, offer answer options that don't overlap and cover every case, and link directly to a decision you need to make.
Can AI write my survey questions for me?
AI is very good at drafting and critiquing survey questions, but you should still review every question and pilot test with real people.
Is it safe to paste survey responses into ChatGPT?
Only after you remove personal details such as names, phone numbers and emails. In Nigeria, the Nigeria Data Protection Act 2023 makes you responsible for personal data you collect. Anonymised responses carry far less risk, and you should also check your AI tool's data settings.
What's the difference between a Likert scale and NPS?
A Likert scale measures agreement or satisfaction, usually on five points from strongly disagree to strongly agree. NPS asks one specific question (how likely are you to recommend us, from 0 to 10) and turns the answers into a single score by subtracting detractors from promoters.
How many responses do I need?
It depends on what you're deciding. For a small business or class project, 50 to 100 responses from the right people can reveal clear patterns. Always report your sample size.
More in Research Workflows.


