AI Customer Support for Shopify: What We Actually Run
AI customer support for Shopify from a five-store operator — Shopify Inbox, Gorgias, Storefront MCP, and the drafts-only agent we actually run.
Key points — AI summary
- A week of scrolling the support channel showed most volume was three zero-judgment questions — where is my order, address change, ship date — all answerable from live order and tracking data (roughly 80% by the author's eyeball, not a measurement)
- Shopify Inbox is the honest starting point for a single store — free, native FAQ and order-status answers — but five stores means five separate inboxes, which breaks when your team lives in Discord
- Gorgias reports its AI resolves roughly 60% of inquiries automatically — the vendor's own number; shop on escalation behavior and draft quality instead, and ask a demo to show exactly what happens when a customer is furious
- What the author runs instead — an agent that drafts replies in Discord while a human clicks send on every message; a wrong draft costs an edit, a wrong sent message costs a customer
- Nothing that moves money should ever act without a human click — an AI with refund authority is an AI a manipulated conversation can spend from, and that rule has no expiry date
Summarized from this article by our writing pipeline; reviewed by the author.
On this page
- Where our support time was actually going
- First stop: what Shopify hands you before you spend anything
- The helpdesk aisle: Gorgias, Tidio, and the automation number
- What we run instead: drafts in Discord, a human clicks send
- How I'd choose if I were starting today
- The multi-store wrinkle no pricing page mentions
In February 2026, customer support was the job I most wanted off my plate. We operate five Shopify stores with a small team that lives in Discord, and the support queue had a rhythm I could set a clock by: messages piling up overnight, most of them variants of "where is my order," a few address changes, and one genuinely hard case buried somewhere in the middle. So I went shopping for AI customer support for Shopify — the native tools, the helpdesk platforms, the build-it-yourself route — and we ended up in the least fashionable corner of the market: an agent that drafts replies in Discord and never sends anything without a human click.
This post is the map I wish someone had handed me back then: what each category actually does, where each one stopped making sense for us, and why "how much does it automate" turned out to be the wrong question to shop on.
Where our support time was actually going
Before evaluating anything, I scrolled back through a week of our support channel and sorted the messages by eye — no spreadsheet, so treat what follows as an operator's impression, not data. The bulk were three questions: where is my order, can I change my address, when will this ship. Every one of them is answerable from live order and tracking data with zero judgment required. The messages that genuinely needed a human — an upset customer, a customs mess, anything touching money — were a small minority, but they were the only ones that could really hurt us if a bot fumbled them.
That split drove every decision afterward. A tool that automates the easy ~80% — my eyeball number from one week of scrolling, not a measurement — only earns its keep if it reliably gets out of the way on the hard remainder.
First stop: what Shopify hands you before you spend anything
Shopify Sidekick was the first thing I looked at, and the first thing I ruled out for this job — not because it's bad, but because it faces the wrong direction. Sidekick is a merchant-side assistant: it helps you analyze customer issues and draft response templates inside the admin, but it doesn't talk to your customers. Useful, just not support automation; our Sidekick guide covers where it does earn a place.
Shopify Inbox is the real customer-facing freebie: native live chat with instant answers for FAQs and automatic order-status lookups, powered by Shopify Magic and trained on your catalog and policies. My honest take — if I ran one store, I would start here and feel no shame about it. It costs nothing and needs no integration work. Where it stopped for us is structural: five stores means five separate inboxes, and my team doesn't sit in the Shopify admin all day. It sits in Discord.
The third native piece isn't a tool at all. Storefront MCP is Shopify's protocol for letting outside AI — ChatGPT, Claude, or an agent you build — read store information and handle order status and returns conversationally. Since March 2026, Agentic Storefronts have been on by default for US merchants, which means some of your customers are already asking an AI assistant about their order instead of your chat widget. MCP isn't a support product; it's the plumbing the build-your-own route runs on, including ours.
The helpdesk aisle: Gorgias, Tidio, and the automation number
If you want to buy rather than build, Gorgias makes the strongest Shopify-native case I found: chat, email, and social in one interface, a pre-sale shopping assistant, and a post-sale support agent wired into order data. Gorgias reports that its AI resolves roughly 60% of support inquiries automatically — that's the vendor's own number, and ours is lower by design, because a human approves every send. I also haven't measured our rate rigorously, so I won't pretend to a comparison I can't back. Tidio plays a similar game at a smaller scale, and platforms like eDesk focus on the routing layer — tagging tickets by urgency and sentiment so the angry ones surface first. At the aggressive end, eesel AI lets the agent process refunds and create replacement orders directly, not just report status.
That last capability is the moment I closed the tab. I've written elsewhere about what I refuse to delegate to an AI agent, and the top of that list is anything that moves money. On our stores that's a permanent stance, not a maturity milestone we haven't reached yet — an AI with refund authority is also an AI a manipulated conversation can spend from.
The mundane reason we didn't buy a helpdesk was cheaper than the philosophical one: I got as far as working the per-store pricing math across five stores, looked at where my team actually spends its day, and stopped. Moving support out of Discord into a separate platform would have added a tool to check, not removed one.
What we run instead: drafts in Discord, a human clicks send
In February 2026 we wired an AI agent into our Discord with partial control of the real stores. Six weeks in, it was running the morning health check, drafting most first-line support replies — "most" is me eyeballing the channel, not a computed percentage — and flagging stuck shipments from our tracking pipeline before the customer writes in. Every draft waits for a human to approve it before anything reaches a customer. The full build story, wiring and mistakes included, is in the Discord AI agent post; I won't retell it here.
What I will tell you is why drafts-only survived contact with reality. Our agent, like every agent I've tested, occasionally states wrong things with complete confidence. A wrong draft costs me an edit. A wrong sent message costs me a customer. I considered promoting the easy tier — pure order-status replies — to auto-send, and decided against it: the approval click costs a few seconds and buys an audit trail plus a human glance at every outbound word. I keep revisiting that decision, and I keep re-making it the same way.
How I'd choose if I were starting today
- One store, modest volume: turn on Shopify Inbox and spend the money elsewhere. Genuinely.
- Real volume, team already living in a helpdesk: a Gorgias-class platform is defensible — but shop on escalation behavior and draft quality, not the headline automation rate. In the demo, ask to see exactly what happens when a customer is furious.
- Team in chat, or more than one store: build on Storefront MCP and the Admin API, and start drafts-only. You'll know within weeks whether the drafts deserve more autonomy.
My contrarian position after this whole exercise: automated-resolution rate is a vendor metric, and optimizing for it points you the wrong way. The number I actually care about is how many hard cases reach a human untouched and on time. And whatever you deploy — nothing that moves money should ever act without a human click. That rule has no expiry date on our stores.
The multi-store wrinkle no pricing page mentions
Everything above gets harder with each additional storefront. Five stores means five sets of policies, brand voices, and escalation quirks, and an AI answering support questions is only as good as its access to the right store's live data — the same consolidation problem as managing multiple Shopify stores from one dashboard. Our agent works because it stands on a unified data layer across the fleet: orders, shipments, and tracking in one place. If you want the build details, an AI agent working across your entire store network covers drafts, approvals, and logs end to end.