Skip to content
← Back to the blog

Store operations

Should an AI Create a Shopify Draft Order Without Approval?

No. Do not let an AI create a customer-facing Shopify draft order (or send its invoice / complete it) without approval. Draft and flag for review are fine. Human yes before liability. Preview the package first; the merchant owns create, invoice, and complete in Shopify admin.

A Shopify draft order is how merchants stage phone, chat, wholesale, and custom-quote sales before the shopper pays. Shopify's draft orders and invoices help covers phone, email, and in-person sales, custom prices, and invoices with a secure checkout link. That path is useful for D2C phone and chat orders and for B2C custom quotes, not only Plus B2B.

An AI store operator is not Shopify Sidekick, not a storefront chatbot, and not SMS marketing. It is also not Shoper, the Polish ecommerce platform. Close spelling. Different product. Shoperator does not create, invoice, or complete Shopify draft orders. It also does not pause, launch, or edit Meta or Google ads. For the category, read What Is an AI Shopify Store Operator?. For related approval gates, read Should an AI Change Shopify Prices Without Approval? and Should an AI Refund a Shopify Order Without Approval?.

Should an AI create a Shopify draft order without approval?

No. An AI should not create a customer-facing Shopify draft order without a human yes, and it should not send the invoice or complete payment either. Assemble and flag a preview first. Human approval comes before create, before invoice send, and before complete. The merchant acts in Shopify admin after each yes.

Shopify Admin GraphQL names those write paths draftOrderCreate, draftOrderInvoiceSend, and draftOrderComplete. Naming the gates is not permission to fire them from a freeform chat model. See Shopify's draftOrderCreate reference.

On Shoperator you text one coordinator (Mark). SMS is standard. Slack is Enterprise. Mark applies supported changes after your yes. Draft order create, invoice send, and complete aren't among them, so you do those in Shopify admin.

Safe ask: "Assemble a draft order package for this phone order: customer, line items, proposed prices, inventory flags, and payment terms. Do not create it." Unsafe as a live execute: "Create draft order and invoice customer X."

What is the difference between preview, create, invoice, and complete?

Preview is a package that has not written to Shopify yet. Create writes a draft order record. Invoice send emails a secure checkout link. Complete turns the draft into a real order and reserves inventory. Those are four stages, not one button. Collapse them and you skip the human gates that matter.

Shopify's creating draft orders help covers building the draft, setting payment terms, and choosing whether to send an invoice or accept payment. Once the invoice goes out, the buyer can pay. Complete creates the order on the Orders page. API language matches that split: draftOrderCreate, draftOrderInvoiceSend, then draftOrderComplete. A yes to preview is not a yes to invoice. A yes to create is not a yes to complete.

What risks does a freeform draft order create?

A freeform draft order can invent custom or off-list prices, attach the wrong customer or company, set payment terms the store did not mean, and oversell units that were never held. Invoice send makes that package a customer promise. Complete locks inventory and creates a real order. Freeform natural-language builds always need human-in-the-loop.

Custom prices and wholesale rates are a normal draft-order use case in Shopify help. That flexibility is also the hazard when a model invents the number. Wrong customer or company can send the checkout link to the wrong buyer. Payment terms such as net 30 change when cash is due.

Inventory is a separate trap. Shopify's draftOrderCreate docs state that creating a draft does not reserve or hold inventory by default, though you can set reserveInventoryUntil. On draftOrderComplete, inventory is reserved for the items in the order. If AI creates drafts without a stock check, storefront sales can still take the last units. For inventory approval habits, read Should an AI Change Shopify Inventory Without Approval?.

What can AI safely assemble before a human yes?

AI can safely assemble a draft order package as a preview: customer or company, channel, line items and variants, proposed prices, inventory flags, and payment-term suggestions. It should not create the draft, send the invoice, or complete payment. A preview is work product. The human yes is the write gate. The merchant then acts in Shopify admin.

A useful package keeps those fields separate. Name the buyer and the channel (phone, chat, wholesale, or custom quote). List each variant and quantity. Call out any custom or off-list price. Flag Available risk. Note proposed payment terms without applying them. Do not instruct the system to "create and invoice" as one fire. Shoperator can follow a preview-then-approve habit in text. It does not claim to create, invoice, or complete draft orders today.

When are bounded draft-order rails ever okay?

Bounded draft-order rails can be okay later when the merchant already locked a fixed price list, a known customer or company, and a capped SKU set. That is not freeform natural language inventing prices and buyers. Rails are merchant-defined order rules. Freeform builds that invent line items, discounts, or payment terms still need human-in-the-loop.

A bounded rail might look like: known wholesale account, SKUs from an approved list, prices from a published price list, and no custom overrides. Shopify's B2B draft orders help shows how company location can pull catalog and payment terms once the company is assigned. The safety comes from locked inputs, not from a model freestyling the quote. Keep freeform phone quotes and invented payment terms behind a human yes before create, a second yes before invoice send, and a third yes before complete.

How does text approval work across the three gates?

Text approval across create, invoice, and complete is preview first, then a separate human yes at each gate, then the merchant executes in Shopify admin. Nothing customer-facing fires because a chat model assembled a quote. On Shoperator, Mark applies supported changes after your yes. Draft order create, invoice send, and complete aren't among them.

Use this gate checklist before anything writes:

  1. Confirm the customer or company and the channel (phone, chat, wholesale)
  2. Assemble line items, variants, and proposed prices as a preview only
  3. Flag inventory, custom/off-list prices, and payment terms
  4. Send that preview for a human yes or no before create
  5. Merchant creates the draft order in Shopify admin after that yes
  6. Get a separate human yes before the merchant sends the invoice in Shopify admin
  7. Get a separate human yes before the merchant completes payment, then verify in Shopify admin

Invoice send is the customer promise. Shopify's draftOrderInvoiceSend mutation emails a secure checkout link. Complete creates the order and reserves inventory per Shopify's draftOrderComplete docs. For the same preview-then-approve habit on catalog work, read How to Run a Shopify Store by Text.

If you want HITL text approval as a pattern example, Shoperator is that example for supported catalog work. Shoperator does not create, invoice, or complete draft orders. The merchant still owns those writes in Shopify admin.

FAQ

Should an AI create a Shopify draft order without approval?

No. An AI should not create a customer-facing Shopify draft order without a human yes, and it should not send the invoice or complete payment either. Assemble and flag a preview first. The merchant acts in Shopify admin after each yes.

What is the difference between preview, create, invoice, and complete?

Preview is a package that has not written to Shopify yet. Create writes a draft order record. Invoice send emails a secure checkout link. Complete turns the draft into a real order and reserves inventory. Treat each as its own approval moment.

What risks does a freeform draft order create?

Custom or off-list prices, wrong customer or company, payment terms the store did not mean, and overselling when stock is not held on create. Invoice send makes the package a customer promise. Complete locks inventory and creates a real order.

What can AI safely assemble before a human yes?

Customer or company, channel, line items and variants, proposed prices, inventory flags, and payment-term suggestions. It should not create the draft, send the invoice, or complete payment. The merchant then acts in Shopify admin.

How does text approval work across the three gates?

Preview first, then a separate human yes before create, before invoice send, and before complete. The merchant executes each approved write in Shopify admin. Nothing customer-facing fires because a chat model assembled a quote.

Does Shoperator create Shopify draft orders?

No. Shoperator does not create, invoice, or complete Shopify draft orders. Mark applies supported changes after your yes. Draft order create, invoice send, and complete aren't among them, so you do those in Shopify admin.