This job is a customer survey: the AI writes five plain questions tied to one decision, and later reads the answers into a short list of what to change. It is a Skill and Tech job that runs across a few sessions, one short sitting to write and build the survey, a week for answers to arrive, and one sitting for the read-out. You give the AI the decision the survey serves and what you already know. It gives back the questions, the invitation message, and later a read-out with themes, counts and the customers' own words. A form tool you already have does the collecting.
- Shape
- SkillTech
- Sitting
- a few sessions
The shape: Skill and Tech
This job takes two shapes because it has two halves that need different things.
The Skill half is the writing and the reading, and both briefs below are yours to keep and reuse for the next survey. Good survey questions are a skill most owners were never taught. They ask one thing at a time, they do not lead the customer toward the answer you hope for, and they use the customer's words instead of yours. The AI is good at this, and good at the reading too: it can sort sixty open-text answers into themes faster than you can scroll through them.
The Tech half is the collecting, and once the form is built it gathers answers on its own while you get on with the week. The AI cannot send a survey, host a form or gather answers. A form tool does that, and you set it up by hand from the AI's draft. That part is mechanical: copy each question into the form builder, pick the answer type, check that responses can go to a spreadsheet, and send yourself a test.
What that means for you: the AI does the parts that need judgment about wording, and you do the parts that need a login. Neither half is long, but they cannot happen in the same sitting, because the answers need time to arrive.
What you give the AI
Gather these before you open the chat. The decision is the input that matters most.
- The one decision this survey serves. Write it as a sentence: "We are deciding whether to add Saturday hours" or "We want to know why people book once and do not come back." One decision per survey.
- What you already believe the answer is. Your best guess, stated plainly. The AI uses it to write questions that could prove you wrong, not questions that confirm you.
- Who will receive it. Past customers, current customers, people who asked for a quote and did not buy. Say roughly how many and how you will reach them: email list, text, a card at the counter.
- Your three or four most common customer questions. The ones you answer every week. They show the AI how your customers talk.
- Two or three recent reviews, copied word for word. Same reason: the customer's own language makes better questions than yours.
- What you will not change, whatever the answers say. Prices you are committed to, a service you will never offer. No point asking about them.
- Your form tool. Google Forms, Microsoft Forms, Typeform, or the form on your website. The AI can match its answer types to what the tool offers.
For the read-out, a week later, you bring one more thing.
- The exported responses, with names, emails and phone numbers deleted. A spreadsheet export pasted in as text, or the file itself if your AI accepts uploads.
What comes back
The job has two briefs, one for each sitting. Paste the first, then your inputs underneath it.
We are a small business writing a short customer survey.
It serves one decision, written below.
Write:
1. Five questions at most. One must be open text.
Each question asks one thing. No question suggests the answer we hope for.
Use the words our customers use (see the reviews and questions below).
2. For each question: the answer type (multiple choice, scale of 1 to 5,
open text) and, for multiple choice, the options, including "Other".
3. One line under each question saying how its answer feeds the decision.
If a question does not feed the decision, cut it.
4. The invitation message: three sentences, from the owner, saying why we
are asking, how long it takes, and what we will do with the answers.
5. A thank-you message for the end of the form, one sentence.
Rules:
- Do not ask for personal details beyond what the decision needs.
- Do not ask about anything listed under "will not change".
- Do not invent statistics or response rates.
Our decision, our guess, who receives it, our customers' questions,
two reviews, what will not change, and our form tool:
[paste here]What comes back is a draft survey you can build in the form tool in a few minutes, plus a message you can send as is or adjust. Build it, send yourself the link, fill it in on your phone, then send it to your customers.
A week later, open a new chat and paste the second brief with the exported answers.
Below are the answers to our customer survey, with personal details removed.
The decision it serves is: [paste the decision].
Return:
1. For each multiple-choice or scale question, the count for each option,
written as counts (9 of 22), not percentages.
2. For the open-text question, the themes that come up more than once.
For each theme: how many answers mention it, and two short quotes
copied exactly from the answers. Mark anything that appears only once
as a single mention.
3. Anything that surprised you compared with our guess, which was:
[paste your guess].
4. Three changes the answers point to, each tied to the answers behind it.
If the answers do not support a change, say so.
Rules:
- Quote customers exactly. Never reword a quote.
- Do not add percentages, benchmarks or industry comparisons.
- If the answers are too few or too mixed to decide, say that plainly.
The answers:
[paste the export]What comes back is a one-page read-out: counts, themes with real quotes, what went against your guess, and up to three changes.
How to judge it
A good draft survey is short, plain and could surprise you. A good read-out quotes your customers exactly and admits what it cannot tell. Reject either, or push back in the same chat, when you see any of these.
- A question that asks two things. "How satisfied are you with our price and service?" cannot be answered honestly. Split it or cut half.
- A question that leads. "How much did you love the new booking page?" assumes the answer. Ask what they thought of it instead.
- More than five questions. The brief said five. If the AI adds a sixth "just in case", cut it.
- Questions about things you will not change. That wastes the customer's time and sets an expectation you will not meet.
- Percentages on a small set. "64 percent prefer weekends" from fourteen answers is nine people. Ask for counts.
- A quote you cannot find. Search the export for a few words of each quote. If a quote is not in the answers word for word, the AI wrote it, and the whole read-out needs a second look.
- Counts that do not match your form tool. Google Forms and Microsoft Forms show their own summary of every multiple-choice question. If the AI's counts differ, the form tool is right.
- Themes that echo your guess back to you. If the read-out confirms everything you already believed, read twenty raw answers yourself before acting on it.
- Changes the answers do not support. Every recommendation should point at the answers behind it. One that does not is the AI's opinion, not your customers'.
What stays with you
The AI writes and reads. Four things stay with you, and they are the parts that decide whether the survey was worth sending.
The decision is yours. A survey without a decision behind it produces a pile of answers nobody acts on. If you cannot name the decision in one sentence, do not send the survey yet.
The asking is yours. Customers answer a survey because they know you, not because the questions are well written. Send it from your own name, to people who have actually bought from you, and say what you will do with the answers.
The privacy is yours. The AI does not know what your customers expected when they gave you their details, or what rules your industry follows. Strip identifying details before pasting anything, and if you work in health, legal or finance, check what you may share.
The change is yours. A read-out tells you what customers said. It cannot tell you whether the change is affordable, whether your team can deliver it, or whether a vocal few speak for everyone. That judgment, and the follow-through, is the whole point of the job. If you change something because of the survey, tell the people who answered. That message is often the most useful thing the survey produces.
An editorial example
This is an invented business, used only to show the method.
Say you run a small dog-grooming shop with one groomer and a front desk. The decision: whether to open on Sunday mornings. Your guess: working owners want weekends, and Saturday is already full.
The AI drafts four questions. Which day and time would you most like to book? How far ahead do you usually book? Have you ever not booked because no time worked for you? What would make booking easier for you?
It leaves out grooming prices, because prices are on your "will not change" list. You build it in Google Forms in a few minutes and send it by text to the people who booked in the last three months.
A week later you have thirty-one answers. The read-out counts eleven of thirty-one asking for an evening slot, against seven for Sunday morning. The open-text theme that comes up most is not the day at all: nine answers mention that booking by phone is hard because nobody picks up during grooming. You check two of the quotes and find each one in the export, word for word.
The survey answered a different question from the one you asked. The change it points to is a way to book without calling, and maybe one evening a week, before any Sunday. That is the job.
Which form tool to use
Use the one you already have. The best form tool is the one you can build in today without signing up for anything.
| If you already use | Collect with | Why |
|---|---|---|
| Gmail or Google Workspace | Google Forms | Responses can go straight to a Google Sheet you export for the read-out |
| Outlook or Microsoft 365 | Microsoft Forms | Responses open in Excel the same way |
| A website builder with a form block | Your site's own form | Customers see your site, not a new brand |
| None of these | Google Forms | It needs only a Google account |
| A survey that needs one question per screen | Typeform | It shows one question at a time; price on vendor site |
Whichever you choose, check before you send that it can export responses as a spreadsheet, because the read-out needs them in one place.
A survey often turns up the same questions your customers ask you every week. Our note on turning everyday customer questions into better content covers what to do with those, and the customer pain points generator gives you a starting list of problems to test before you write your questions. When this one is done, the full list of DIY marketing jobs has the next one.
Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free
