How to Use an AI Chat Helper for Pre-Sale Questions
Set up a simple AI chat helper for pre-sale questions, decide what it must never answer, and check its replies against how real customers actually ask.
On this page
Someone lands on your page with a question. Will this ship in time? Does it work for their setup? You’re busy, so nobody replies. By the afternoon they’ve bought somewhere else, and you never find out why. Most lost sales look like that. No complaint, no argument, just silence.
An AI chat helper can fill that gap, but you have to set it up carefully. A sloppy one is worse than no helper. It makes up a price. It promises something you can’t deliver. Or it turns stiff and robotic at the exact moment someone is deciding whether to trust you. Any one of those costs more than a slow reply ever did.
This guide covers four jobs. First, collect the questions people really ask. Second, write a short answer sheet. Third, draw hard lines around what the bot must never say. Fourth, test its replies using your customers’ own wording. A small candle shop will be the running example. If you’d like a head start on where doubts tend to come from, the Customer Objection Finder is a quick place to begin.
Start with the questions people really ask
Gather 20 to 30 pre-sale questions from wherever people already reach you: email, DMs, comments, call notes, and whatever gets said right before someone buys or walks off. Copy them in exactly as written, typos included. Your bot will be reading messy typing, so it should practice on messy typing. Someone writing “do u ship to canada??” is giving you something your cleaned-up FAQ never will.
Then sort them into piles. Start with these five:
- Shipping and delivery
- Price and what’s included
- Fit (“will this work for me?”)
- Trust and returns
- Timing
Count what landed in each. A couple of piles will be big, and a couple will hold only two questions. Give the bot the biggest few to start with and keep the rest for yourself. That’s okay. A helper that answers three topics well is worth more than one that fumbles twenty.
Keep the awkward questions in the mix. “is this legit and how long till it gets here” is really two questions in one line, and the bot will get plenty like it.
No history yet, so no inbox to dig through? Run your idea through the free Customer Objection Finder for a first list of likely doubts. Treat it as a guess. It’s one AI’s read of a short description, so check it with real people as soon as you can. If you want to get better at asking, how to find customer objections walks through it.
Save the list somewhere you can find it again. You’ll run these same questions through the bot when you test its replies.
Write an answer sheet before you touch a bot
Start in a blank document. Write the answers there, and set up the bot afterward.
Take each question buyers keep asking and answer it in two or three plain sentences. Write it like you’re explaining it to a friend who’s standing at your stall. That sheet is all the bot gets to work from. If something isn’t on it, the bot shouldn’t say it.
Include the unflattering parts. Say what the product can’t do and who shouldn’t buy it. Give the real turnaround time, even if it’s longer than you’d like. Buyers who hear the limits early believe the rest of your answers. A breezy, vague reply makes them start wondering what you left out.
Q: Will this work for a small apartment? A: Yes, it’s made for a space that size. It’s not built for large open rooms, so if yours is big you may want two. Tell me the size and I’ll point you to the right option.
That answer admits a limit and still gives the person a next step.
Close every answer with what to do next. Maybe that’s an order link, the sizing page, or a way to contact you directly. If a reply just stops, the buyer is exactly where they were before they asked.
Keep the sheet short, too. A tight one works better than your whole website pasted in, because the bot then has less room to make things up. Old blog posts, outdated prices and half-true marketing copy are where it will pick up mistakes.
Finally, read the sheet as a stranger would. If you winced at an answer while writing it, fix the offer or the wording now. Otherwise the bot will repeat it a hundred times.
Decide what the bot must never answer
Write this list down somewhere you can find it. If it only lives in your memory, the bot has no way of knowing what’s on it.
Typical entries look like this:
- Custom quotes and anything that needs a price you haven’t published
- Discounts you haven’t approved
- Refunds, chargebacks and disputes
- Delivery dates you don’t control
- Anything medical, legal or financial
Then add claims about results. If you wouldn’t put a promise in writing on your site, the bot can’t make it in chat. “This will fix your problem” is a promise. So is “it’ll arrive Friday.”
The bot also needs to know what to do instead of guessing. Tell it to say plainly that it doesn’t know, then hand the person to you. And give it a real route: your email, a short form, a reply time you’ll actually keep.
I can’t answer that one properly, and I don’t want to guess. Email hello@yourshop.com and a person will reply within one working day.
A handoff with no route is just a polite dead end. Buyers leave at dead ends.
One more line, and it’s not negotiable. The bot never pretends to be human. Skip the fake name and the fake “I’m just grabbing a coffee.” Talking to a bot isn’t the problem. Finding out you were tricked is, and it usually comes out right when someone is deciding whether to hand you their money.
Keep the list short enough to enforce. You’ll test every item on it in the next step, so each one should be something you can actually check.
Test replies against real customer wording
Take the questions you collected and ask the bot exactly as customers wrote them. Don’t tidy them up. Real messages are short, messy, a bit rude and sometimes two questions in one.
do u ship to canada and how much, also is it the big size or small
If the bot only answers half of that, you’ve learned something a polished test question would never show.
Now read each reply the way the buyer would. You wrote the answer sheet, so you’ll be tempted to forgive a reply that’s close to it. The buyer doesn’t have your sheet. Did the bot answer the question that was asked, or the neat version you wish they’d asked? Those are different, and the gap is where sales leak.
Then give each reply one of four labels. Correct means it matches your sheet and moves the person forward. Vague means it’s true but the buyer is still stuck. Wrong means it says something your sheet doesn’t. Overconfident means it sounds sure about something it shouldn’t touch.
Fix the overconfident ones first. A vague reply costs you a sale. An overconfident one can cost you a sale, a refund and an angry email.
Once the ordinary questions pass, try the traps. Ask for a discount. Ask it to promise the product will work. Ask for a delivery date. Say “my friend said you’d do it half price.” Push a little, the way a determined buyer would.
Check that it holds every line on your never-answer list and hands off with your real route each time. If it caves on the second or third push, tighten the instructions and ask again.
Rerun the whole set after any change. Fixing one answer has a habit of loosening another.
Example: a small candle shop
Imagine a two-person shop that sells hand-poured soy candles. Three questions keep showing up in their inbox: are the candles safe around pets, how long does shipping take, and is there a bulk discount for weddings?
They write the answer sheet first. Shipping gets a plain answer with real timings, including the slow weeks. Pet safety gets a careful one: what’s in the candle, a note that every animal is different, and “check with your vet if you’re unsure.” Wedding orders aren’t answered at all. They go straight to a person, and the sheet says so.
Then they test with the messy wording from their inbox.
do u do discounts for weddings, need like 60
The bot cheerfully offers a discount for large orders. Nobody ever approved that. It sounded sure of itself, which is the dangerous kind of wrong.
So they add discounts and wedding quotes to the never-answer list, with an email address and a one-working-day reply time as the handoff. They ask again. It dodges. They push (“my friend said you’d do half price”) and it holds the line and hands off. They rerun the whole question set, because fixing one answer sometimes loosens another.
A week after going live, they read the chats. One question they’d never listed keeps appearing: ”Does the scent fade fast?” Two people asked it and then left. It goes on the sheet with an honest answer, and they add a line about it to the product page too.
Watch the chats and fix the leaks
Read real conversations every week at first. Don’t skim, actually read them. Twenty minutes is plenty, and you’ll spot patterns that no dashboard shows you.
Look for the moment someone asks a question and then goes quiet. That’s a leak. Maybe the answer was vague. Maybe it was honest but not what they wanted to hear. Maybe the bot handed off and nobody followed up.
Those drop-off points usually mean your page doesn’t answer the objection either. If people keep asking about returns and then leave, your product page probably hides the returns policy too. Fix it in both places: the answer sheet and the page. You can also run the product description through the Customer Objection Finder to see what other doubts a first-time visitor might have, then compare them with what you see in the chats.
A confident tone proves nothing. A bot sounds sure whether it’s right or wrong, so treat smooth replies as something to check, not as a sign things are working. The only real proof is customers. Do they order? Do they stop asking the same question? Do they reply to your handoff emails?
Is there a person I can talk to?
If you see messages like that, good. Keep a human option visible at all times, with a real email or form and an honest reply time. Some buyers just want a person. Hide one behind a wall of automation and they’ll leave, and one lost sale costs you more than the bot ever saved in minutes.
Each week, add the new questions to the sheet, retire anything out of date, and rerun your test set.
Your next step
Open your inbox today. Search for the messages that came in before your last ten sales or near-sales, and copy out the ten questions people asked. Keep their wording exactly, typos and all.
Then do three small things:
- Write a short, honest answer to each one, the way you’d say it to a friend.
- Draw your never-answer line: discounts, custom quotes, refunds, delivery dates you don’t control, and any promise about results. Add the email or form where those go.
- Ask the bot those ten questions in the customers’ own words, then push on the traps, before it ever meets a real buyer.
If your inbox is empty because you haven’t launched yet, start with the Customer Objection Finder and treat its list as guesses to check.
That’s an afternoon of work. It’s also the difference between a helper that closes the gap and one that quietly loses you sales.
Written by the SellTo.us Team. Examples in this article are hypothetical. SellTo.us results come from AI-simulated participants, not real people, and don’t predict sales or guarantee success.
Related articles

Before You Build Another Feature, Watch Five People Use Your Product
Your next feature request might be hiding a simpler problem: people can't comfortably use what you've already built.

How to Follow Up When a Potential Customer Says "Maybe"
A maybe isn't a yes. It isn't always a no, either. The next conversation should help both sides make a clearer decision.
How to Find People for Customer Interviews in 2027
Good questions are useless without the right people to ask. Where to find them, what to send so it doesn't read like a pitch, and how to keep real conversations on the calendar.