← The Maker Growth Blog

Your inbox is the same eight questions on repeat. Build the assistant that drafts the answers — and keep yourself in charge.

AI for makersChatGPTcustomer servicehandmade business

Open your Etsy messages, your Shopify inbox, and your Instagram DMs on any given morning and you’ll see the same thing: eight questions you’ve answered a hundred times, worded eighty different ways. Can you do it in navy? How long does a custom one take? Do you ship by the 14th? What’s the price for 25? The questions aren’t hard. But typing a careful, warm answer to each one, across three apps, is a slow tax on every single day. This is what AI customer service for a small business is supposed to solve. The fix most people reach for is a bot that answers for you, and that’s the wrong fix. The right version isn’t a bot you hand the customer to. It’s a supervised assistant that drafts the reply, flags what it doesn’t know, and hands it back to you to send. You stay in charge of every promise. Here’s how to build it.

Why a bot is the wrong answer (and a draft assistant is the right one)

A customer-facing bot promises things on your behalf before you’ve seen them. It quotes a price from last season, promises a turnaround you can’t hit this week, or confirms a rush order while your machine is down. One confident wrong answer to a buyer costs you a refund, a bad review, or a remake: the exact things you got into this to avoid. A supervised assistant flips the risk: it does the typing, you do the deciding. Nothing reaches a customer that didn’t pass your eyes. That’s the same healthy split that runs under the whole AI back office for a maker business — AI organizes, you judge.

Step 1: collect 15–20 real inquiries you’ve actually answered

Open your message history and pull 15 to 20 real questions from the last couple of months. Cover the full spread: personalization options, materials, bulk and wholesale pricing, turnaround time, shipping and deadlines, artwork and file formats, rush orders, and returns. Strip out names, addresses, and anything payment-related before it goes anywhere near a chatbot — you’re saving the question, not the customer’s identity. This set is your raw material. It’s also quietly useful on its own: the questions that show up most are the ones your listings aren’t answering yet.

Step 2: build the source sheet — the one thing that makes this work

This is the step everyone skips, and it’s the step that makes AI customer service trustworthy instead of risky. Write one document, your source sheet, that holds only answers you’ve approved: your real policies, your current turnaround times, your production limits, your materials, your shipping cutoffs, and a short list of questions that must always be escalated to you (custom pricing, rush availability, anything about refunds). The assistant is only ever allowed to answer from this sheet. If the answer isn’t in it, the correct output is “owner review needed,” not a guess. That one rule is the difference between a helper and a liability.

Here’s how a quick reply goes wrong (an example, not a real customer): a buyer asks, “Can you do 25 of these by the 14th?” It’s a busy morning, the answer feels like yes, and you type “Absolutely!” Then you check the queue. Two other orders are already due that week and you’re short on blanks. Now you’re either working nights or writing an apology. That’s why rush dates and bulk quantities live on the escalate list. A good source sheet holds your bulk pricing too, worked out once from your real costs in the Laser Pricing Calculator (it covers UV jobs as well), so the assistant can quote the price table you approved and nothing else.

Step 3: run the draft-assistant prompt

Now paste a new inquiry in with your source sheet and ask for three things, not one:

Act as a draft assistant for my handmade business. Using only the policies and product information below, write a warm reply to this customer. Then list: 1) information still needed, 2) any promise I should not make yet, and 3) the next internal action. If the answer is not in my source material, say “owner review needed” instead of guessing.

You get back a ready-to-edit reply, a checklist of what’s missing, and a note on what to hold back, plus a flag any time it hits the edge of what it actually knows. The “only from my source material” line is doing the heavy lifting; without it, AI will fill gaps with plausible-sounding guesses, which is exactly what burns you. If you’d rather not write the prompt from scratch, the free Etsy Listing Prompt Builder has a custom-order reply mode that builds this kind of grounded prompt from your real product facts. If you’ve ever gotten confident-but-wrong answers out of ChatGPT, it’s almost always a grounding problem. Here’s how to prompt it so it stops making things up.

Step 4: add the human line, then send it yourself

The draft gets you 80% there in seconds. Your job is the last 20%: make it sound like you, add the one detail only you know (the dye-lot note, the “I’ll have your proof over by Thursday”), and confirm every number against reality before you hit send. A reply that reads like a maker who’s glad they reached out beats a flawless-but-robotic one every time. This is the same discipline that turns a vague custom-order DM into a clean quote — if that’s most of your inbox, here’s the custom-order version of this workflow, and the 12-minute version for corporate and bulk inquiries.

Step 5: feed the recurring questions back into your shop

After a couple of weeks, you’ll see the same three or four questions driving half your inbox. That’s a gift. Every recurring question is a hole in your listing, your FAQ, or your shop policies — patch it at the source and the question stops arriving. Update the listing description, add the answer to your saved replies, tighten the quote form. If the question is about size, materials or how to order, run the listing through the free Etsy Listing Checker. It checks your title, tags and description against Etsy’s current guidance, which puts size, materials and how to order near the top of the description. That’s exactly where a repeat question should be answered. A customer-reply helper is one of the first three AI workflows every maker should set up precisely because it compounds: the better your source sheet gets, the less the inbox needs you at all.

The line AI never crosses

An assistant can draft, sort, and flag. What it must never do on its own is promise inventory, confirm a delivery date, approve a rush, quote a custom price, or authorize a refund — unless it has verified, current access to that data, which a pasted source sheet is not. Those are judgment calls tied to your real queue, your real stock, and your real capacity this week. Let AI carry the typing so you have the head-space for the deciding. Keep the relationship; automate the keystrokes.

Draft it with AI. Send it as you.

Keep reading

Ready to put this to work? Join the Maker Growth Hub.

Build this into your shop

The Maker Growth Hub gives you the templates, prompts, live calls, and community to put tactics like this to work — $25/mo, locked for life.

Join the Hub →
← All posts