How to Draft Gorgias Replies with AI (From Your Help Center)

What "drafting with AI" means in Gorgias

Gorgias gives you two distinct AI surfaces, and it helps to separate them before you build anything. Copilot is agent-assist: it sits beside a human, drafts a reply, summarizes a long thread, and reads sentiment so the person sending the message moves faster. AI Agent (the Automate side) is autonomous: it reads an incoming conversation and, within limits you set, resolves it end to end without a human touching it.

Both lean on the same source material — your Help Center articles, your store data, and the guidance you write — but they pull it differently. The important nuance, well documented by eesel's teardown of Gorgias AI Agent actions, is that the AI Agent writes its own replies from Skills and Guidance and knowledge retrieval — it does not use your Macros at all. Macros and Quick Responses stay a human-only tool. So if your team has spent a year perfecting a macro library for returns, exchanges, and shipping delays, the native AI Agent won't reuse a word of it; it composes fresh text and you steer tone through Skills and Guidance instead.

Step 1: Ground the AI in your Help Center

The single highest-leverage move is making sure the AI has your real knowledge to draw on. In Gorgias, that means:

1. Connect your Help Center so published articles become retrievable knowledge. FAQ-style answers (return window, shipping zones, sizing) are handled by knowledge retrieval rather than a scripted action.
2. Add Guidance — short reference documents you write for the agent — to cover the nuance an article doesn't, like how to phrase a goodwill exception or which tone to use for an angry WISMO customer.
3. Connect your store data (Shopify) so order status, tracking, and customer history are available to the draft, not just static text.

The payoff is that a shopper asking "can I return these boots, they're the wrong size?" gets a reply built from your actual returns policy — window, condition rules, prepaid-label logic — instead of a plausible-sounding guess.

Step 2: Shape tone and behavior with Skills

Because the native AI Agent ignores Macros, tone lives in Skills — instructions written in a WHEN/IF/THEN structure that trigger on intent. A return Skill might read: when the intent is "return request," if the order is inside the 30-day window, then explain the process and offer a prepaid label; if outside the window, then apologize and offer store credit as a goodwill gesture. You're effectively rebuilding the judgment that used to live in your macro library and your team's heads, but as explicit rules the agent follows.

This is genuinely powerful for repeatable, high-volume intents. It's also where the work is: every branch of behavior you care about has to be authored, tested, and maintained, and there's a real risk of mutating live customer data while you test against production.

Step 3: Decide draft-and-review vs. auto-send

Not every reply should ship on its own. A sensible rollout for an ecommerce team looks like a ladder:

1. Draft only for the first few weeks — the AI writes, a human reads, edits, and sends. This is where you learn whether the grounding is right before anything reaches a customer.
2. Auto-send the safe intents — order status, "where's my package," basic return eligibility — once you trust the drafts.
3. Keep judgment calls human — refund disputes, damaged-item claims, anything with a goodwill decision — routed to an agent with a strong draft waiting.

Gorgias itself caps AI automation at roughly 60% of email and chat conversations, per MyAskAI's Gorgias Automate guide, which is a useful reminder that the native tool is designed to assist a human team, not replace one.

The honest limits — and where an AI layer picks up

Native Gorgias Automate and AI are legitimately good, and worth crediting plainly. Copilot makes human agents faster; AI Agent can fully resolve routine, well-defined intents; and the knowledge retrieval genuinely pulls from your Help Center. If you're on Shopify and your volume is dominated by a handful of repeatable intents, the native tools may take you a long way.

But there are seams worth being clear-eyed about:

• It doesn't reuse your Macros or Quick Responses. The library your team trusts is invisible to the AI Agent; you re-author that behavior as Skills.
• It's Shopify-centric, with essentially one agent per store — awkward for multi-brand or multi-store operations.
• Deflection isn't the same as resolution. An answer that closes an FAQ is not the same as looking up an order, checking a subscription, and taking the right action — and the native billing model charges for full resolution regardless of how the outcome lands.

That billing model is the part most teams underestimate. Gorgias AI Agent is priced per automated interaction — a conversation the AI resolves entirely on its own. Per Gorgias's own AI Agent pricing page, that's roughly $0.90 per resolved conversation on most annual plans and $1.00 on Starter, with overages billed higher — around $1.50 per interaction for Support + Shopping Assistant. Human-resolved conversations, spam, and unanswered emails aren't charged, and on handoff the ticket shifts to your Helpdesk plan rather than double-charging. Details and current numbers are laid out in our Gorgias pricing explained breakdown.

Here's why the unit matters. Per-resolution pricing is clean when the AI truly closes a conversation, but it couples your cost to an outcome you don't fully control — and it doesn't distinguish between a one-line FAQ deflection and a multi-step automation. A dedicated agent layer can price differently:

Macha is that layer. It runs on top of the Gorgias you already use through a live connector — it does not replace your help desk. You connect Macha to Gorgias, and its Customer Reply Assistant drafts a reply grounded in your Help Center and your own systems, then hands you a ready-to-send draft you can paste straight in (respecting the tone and templates your team already relies on). Crucially, Macha's credits are consumed per AI action — a draft, a lookup, a send — not per resolution, so the cost tracks the work performed rather than an outcome you can't fully steer. See the pricing page for how credits map to actions.

The lookups themselves come from custom tools: a return-eligibility check or an order-status call becomes something the agent can invoke mid-draft, which is exactly what the screenshot above shows — two tools feeding one KB-grounded return reply. For the full setup, our guide on connecting Gorgias to Macha to automate ticket replies walks through it end to end, and the best AI agents for Gorgias compares the options if you're still evaluating.

The clean division of labour: let Gorgias stay the system of record for tickets and let its native AI handle the intents it's tuned for, then layer an agent on top for KB-grounded drafting that reuses your templates, taps your live tools, and bills for the work rather than the outcome.

FAQ

Does Gorgias AI use my Macros and Quick Responses to draft replies? The native AI Agent does not — it writes its own replies from Skills, Guidance, and knowledge retrieval, and Macros stay a human-only tool. Copilot (agent-assist) helps humans who can still apply Macros manually. If reusing your existing templates matters, an agent layer like Macha can draft in your templates and tone rather than composing from scratch.

How does Gorgias draft replies from my Help Center? Connect your Help Center so published articles become retrievable knowledge, then add Guidance documents for nuance. FAQ-style questions (return windows, shipping zones) are answered by knowledge retrieval; store data from Shopify supplies order and tracking details.

How is Gorgias AI Agent billed? Per automated interaction — a conversation the AI resolves entirely on its own. Public figures put this around $0.90 per resolved conversation on annual plans and $1.00 on Starter, with overages higher (about $1.50 per interaction). Human-resolved, spam, and unanswered messages aren't charged. Confirm current numbers on Gorgias's pricing page.

What's the difference between deflection and resolution? Deflection means a customer's question is answered so they don't need an agent (often a one-line FAQ). Resolution means the whole conversation is closed — sometimes requiring a lookup or action. Automation is the broader category of doing the work either way. It matters because native Gorgias bills per resolution, while Macha bills per AI action.

Can I add AI drafting to Gorgias without replacing Gorgias? Yes. Macha connects to Gorgias as a live connector and runs on top of your existing help desk — it doesn't replace it. It drafts KB-grounded replies you can review and send, reusing your tone and templates, while Gorgias stays the system of record.

Ready to draft grounded Gorgias replies your agents can actually send? Start a free trial of Macha and connect it to your Gorgias in minutes.