Your customers' 20 questions
How to collect them, sort them and put them to work in one afternoon
By the end of the afternoon you'll have three things: your list of 20 real questions, the 10 test prompts you'll use to check whether AI names you, and your first three answers written for your site and your bot. This guide won't promise ChatGPT recommends you tomorrow. It gives you the thing that made the difference in our test: a written answer to the specific question the customer asked.
If you haven't read the article this guide goes with, start there; the test and the screenshots are in it. To get through steps 3 to 6 faster, use the Question Bank: paste your list and it hands back the three lists, sorted.
What you need: your phone, access to your email and DMs, a blank doc or spreadsheet, and a bit under three hours.
Pull your last 20 questions (45 minutes)
Go through the places where people ask you things, in this order. Start with the most recent and stop at 20.
| Where | How to search fast |
|---|---|
| Business WhatsApp | Search inside your chats for "?", "how much", "price", "can I", "do you", "is there", "included", "rain" |
| Personal WhatsApp and texts | Same search. This is where referrals and friends-of-friends tend to ask |
| Instagram and Facebook DMs | Filter for messages from people, not from accounts trying to sell you something |
| Email and web forms | Search "question", "wondering", "info", "quick question" |
| Reviews | Read for phrases like "wasn't sure if", "I was worried", "I asked". That's the doubt someone had before they bought |
| Calls and walk-ins | Write down the ones you got this week from memory. From today on, write them down the same day |
Can't get to 20? That's fine. Twelve is enough to start; twenty shows the patterns. Here's the signal that you're missing some, though: if most of what you find is from salespeople, your real customer questions are coming in somewhere you're not saving. That's exactly what we found when we went through our own CRM.
Write them down exactly as asked (15 minutes)
Three rules, no exceptions:
- Don't clean them up. "My mom doesn't speak English, does the captain speak Spanish?" is worth more than "Do you offer Spanish-language service?" The customer's version is the one that sounds like what they'll type into ChatGPT.
- Strip personal details. Names, phone numbers, addresses, the amount on a specific invoice. Keep the question, drop the person.
- Keep the language it came in. If someone asked in Spanish, write it down in Spanish. Don't translate it. A customer who asks in Spanish phrases things differently, and that difference is information.
Sort them into six intents (20 minutes)
Almost every question a service business gets falls into one of these six. Put each question in its group.
| Intent | What it sounds like | Example (fishing charter) |
|---|---|---|
| 1. Price and terms | how much, what's in the price, deposit, tip | "Is the price per person or per boat?" |
| 2. Availability | when, any openings, hours, how long it lasts | "Do you have anything Saturday morning?" |
| 3. Who it's for | age, language, skill level, accessibility, pets | "Can my 7-year-old come?" |
| 4. What's included and what to bring | gear, food, parking, what to wear | "Can we bring our own food?" |
| 5. What happens if | cancellation, bad weather, guarantee, refund | "What happens if the seas are rough?" |
| 6. Trust | experience, licenses, insurance, reviews, who shows up | "Is the captain licensed?" |
Then mark two things on each question:
- How often it repeats. The ones that repeat most go first.
- How much money rides on it. Some rare questions decide a big sale. One family asking for a Spanish-speaking captain can be a whole boat booked.
Keep the 10 that come out on top when you combine both marks.
Turn the top 10 into test prompts (20 minutes)
A test prompt is the customer's question, filled in with what AI needs to go looking: your type of business and your city. Template:
[The customer's situation, in their words] + [the service they want] + in [your city] + [the condition that matters to them]
Examples with the charter:
| What the customer asked | Test prompt |
|---|---|
| "Can my 7-year-old come?" | "We're a family with a 7-year-old. Which fishing charters in Fort Lauderdale take kids that age?" |
| "What happens if the seas are rough?" | "Which Fort Lauderdale fishing charters give you a refund or let you reschedule if the seas are too rough?" |
| "¿El capitán habla español?" | "Somos una familia y mi mamá solo habla español. ¿Qué charter de pesca de medio día en Fort Lauderdale tiene un capitán que hable español?" |
The middle one is word for word a prompt we ran in our test. ChatGPT answered it with six charters, and five of them hadn't shown up in either of its answers to the generic question.
Two things to watch:
- Don't put your name in the prompt. The customer doesn't know you exist yet. The point is to see whether AI finds you without a nudge.
- If you serve customers in two languages, write both versions. Write each one the way a customer who speaks that language would say it, not as a translation of the other.
Write the answer and publish it (40 minutes)
Start with the three questions that repeat most. For each one, write a two- or three-sentence answer in this shape:
[Direct answer: yes, no, or how much] + [the concrete fact that backs it up] + [what the customer does next]
Example:
"Yes, kids 4 and up are welcome. We carry life jackets in every kid size, and Captain Luis speaks English and Spanish. For families we recommend the half-day trip at 8:00 a.m."
(Made-up example for this guide. Only write what's true for your business.)
Where each answer goes:
| Intent | Where you put it |
|---|---|
| Price and terms | On each service page, next to the price, and on any booking platform where you're listed |
| Availability | On the service page, your Google Business Profile and your listings |
| Who it's for | On the service page, your team page (languages, experience) and your listings |
| What's included and what to bring | On the service page, as a list |
| What happens if | On a policy page linked from every service, and on your listings |
| Trust | On your team page and your About page |
Then put all of them in your FAQ section too, written as question and answer.
Don't skip the listings. In our English test, the charters ChatGPT and Perplexity recommended for the detailed question were ones whose own site or booking listing said, in plain words, "English and Spanish." The text the tools relied on was on booking listings as often as on the businesses' own sites. If you're listed on Google, Yelp, a booking site or a directory in your industry, those pages get read too. Keep them as complete as your own site. We covered why the first answer matters so much in AI search has no page two.
Match your bot and measure monthly (20 minutes)
Consistency. Copy every answer from step 5 into three places:
| Place | What you do |
|---|---|
| Your site and listings | Already done (step 5) |
| Your bot or saved WhatsApp replies | Paste the same answer. If you use an AI assistant, add it to its knowledge base |
| Your team | Send them the list in writing. Same question, same answer |
If a fact changes (a price, a minimum age, a schedule), change it in all three places the same day. If your bot and your site say different things, the customer believes neither. We dug into this in what your AI agent answers at 11 p.m.
Measurement, one afternoon a month:
| Month | Questions answered on your site | Appearances (out of 60) | Notes |
|---|---|---|---|
| Month 1 (starting point) | |||
| Month 2 | |||
| Month 3 |
This number isn't statistics. In its January 2026 study, SparkToro suggests running each prompt 60 to 100 times for a reliable average. Yours is a compass. What matters is the trend, using the same prompts every month.
What not to do
- Don't write answers that aren't true just to show up in AI. If your page says you speak Spanish and you don't, AI sends you the wrong customer.
- Don't stuff the page with made-up questions. Use the ones people actually asked. Invented ones read as invented.
- Don't track your "ranking." The lists shift from one answer to the next. Track whether you show up.
- Don't translate your answers word for word. If you serve two languages, write each version the way a customer in that language would say it.
A full example, start to finish
Sample business, made up for this guide: a fishing charter in Fort Lauderdale.
Question that came in on WhatsApp: "Hi, there's 5 of us, two little kids, and my mom doesn't speak English. Do you do half days?"
Intent: who it's for (age, language) and availability.
Test prompt (EN): "We're taking our two kids, 7 and 9, fishing for the first time, and my mom only speaks Spanish. Can you recommend a half-day fishing charter in Fort Lauderdale with a captain who speaks Spanish?"
Test prompt (ES): "Somos una familia con dos niños de 7 y 9 años y mi mamá solo habla español. ¿Qué charter de pesca de medio día en Fort Lauderdale nos recomiendas, con un capitán que hable español?"
Answer for the site, listings and bot: "Yes. We run half-day trips (4 hours) at 7:00 a.m. and 1:00 p.m., kids 4 and up are welcome, and the captain speaks English and Spanish. Pricing is per boat, up to 6 people."
Where it goes: the half-day trip page, the team page (the captain's languages), the FAQ, and every booking listing.
What to measure: whether this prompt names you in ChatGPT, Gemini and Perplexity, twice each, this month and next.
If you'd rather do it with us
This is what we set up for our clients at Hub365: we pull questions from every channel into the CRM, write the answers in English and Spanish, put them on the site and in the WhatsApp agent, and measure every month. If you want to see it applied to your business, book a free 30-minute call.
Sources
- Rand Fishkin with Patrick O'Donnell (Gumshoe), "NEW Research: AIs are highly inconsistent when recommending brands or products...", SparkToro, January 27, 2026 · https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
- Amanda Natividad, "Why Everyone Prompts AI Differently (Even When They Want the Same Thing)", SparkToro, February 11, 2026 · https://sparktoro.com/blog/why-everyone-prompts-ai-differently-even-when-they-want-the-same-thing/
- Hub365's own test in ChatGPT and Perplexity, November 17, 2026, sample business (fishing charter in Fort Lauderdale) · /blog/your-customers-questions-and-ai






