Square + ChatGPT Ordering: The Local Merchant Setup Playbook

Square’s ChatGPT and Claude ordering launch made local merchants a simple promise: if you are a US Food & Beverage seller with an activated Square Online Ordering profile, you are already in — no new API integration, no additional setup, no extra fees. That promise is true for eligibility. It is not the same thing as being ready.

An AI agent taking a customer’s order does not walk past your counter or smell the espresso. It reads your catalog — item names, descriptions, modifiers, prices, stock levels — and paraphrases what it finds to someone deciding what to eat. Data hygiene that used to be a back-office POS concern is now, quite literally, your storefront in a new channel.

This guide is the merchant-side companion to our coverage of Square’s ChatGPT and Claude ordering launch, which covers the announcement itself and the commission-versus-marketplace math. Here we walk the actual setup: activating the Online Ordering profile, preparing the catalog an agent will read, keeping inventory honest, configuring pickup guardrails, and understanding the delivery-partner economics — every step sourced from Square’s own support and developer documentation.

Square’s July 1 press release framed the launch in the most merchant-friendly terms possible: eligible sellers are opted in with “no additional work, setup, or fees required.” That framing is accurate — and it hides a distinction worth being precise about. Square automated eligibility: if you are a US Food & Beverage seller and your Online Ordering profile is active, your menu can surface when a customer asks ChatGPT or Claude where to order. Square did not — because it cannot — automate performance: whether your items read well, price correctly, and show as in stock when an agent looks.

The strategic logic behind the launch is worth one paragraph before the mechanics, because it explains why the setup burden lands where it does. Square’s stated position is that discovery itself is moving into AI assistants: “As consumers increasingly turn to AI to decide where to eat, shop, and book services, Square is extending this approach to help businesses show up with accurate information,” per the company’s press release. Square wins by being the rails under that shift — which is why it made joining the channel free and frictionless, and why the quality of what agents find on those rails is left entirely to the merchant.

Everything else in this guide is optimization. This step is the gate: without an activated Online Ordering profile, a seller is not eligible for the AI-ordering surface at all, no matter how long they have run Square POS. Two constraints apply before you even reach the activation button, both from Square’s own support documentation:

• Food & Beverage sellers only. Online Ordering profiles are not available to other seller categories — which also means the AI-ordering channel is, for now, a restaurant, café, and food-retail story.
• A menu must exist first. The profile cannot be activated until at least one menu is created with items available on it.

With those in place, activation is a short Dashboard sequence:

1. Open Square Dashboard → Channels → Online Ordering.
2. Confirm your menus are assigned to the profile and have items available.
3. Click “Enable ordering.” If your business location has a valid address, pickup fulfillment configures automatically.

Before the profile goes live, Square asks you to complete its configuration: branding (logo and colors), a profile name and Cashtag, a business description, and your pickup and delivery preferences. Treat the business description with the same care as your item copy — it is part of the data set that represents you in channels you do not render yourself.

The technical foundation matters here because it defines what an agent can and cannot see. Square’s Catalog API is the data layer the AI ordering surface reads from: it manages items, item variations, categories, modifiers, images, taxes, discounts, and measurement units. When a customer asks an assistant “what’s good at the café on Fifth,” the raw material for the answer is whatever that structure contains. Sparse descriptions produce sparse answers; missing modifiers produce orders your kitchen has to phone-correct.

Three mechanics deserve specific attention, all from Square’s developer documentation:

• Every item needs at least one variation. A CatalogItem must have at least one CatalogItemVariation before it can be added to a purchase or transaction. An item without one is effectively invisible to any ordering channel — including AI agents. Items support up to 250 variations each.
• There is no separate publish step. New items created via the Catalog API are immediately visible in the Square Dashboard and POS across all locations. Catalog fixes propagate as fast as you make them — and so do catalog mistakes.
• Third-party tools need the right scopes. Applications need the ITEMS_READ OAuth permission to read catalog objects and ITEMS_WRITE to create or modify them — worth checking if a menu-management or POS-sync tool pushes into your Square catalog on your behalf.

The trend beneath the mechanics is the part we think merchants should internalize. For years, the business case for cleaning up Square catalog data was indirect — nicer receipts, better website listings, marginal SEO. Agentic ordering makes the case direct: as one DTC-focused analyst at Eightx put it, “agents surface what they can read” — the framing being that product descriptions, pricing, and availability accuracy now function as the new SEO. That is an analyst’s characterization rather than a Square requirement, but it matches how the architecture works: the agent has no eyes, only your data. The same logic drives API-first commerce architecture generally — structured, machine-readable product data is the asset; every new channel is a renderer.

A practical pass for a typical local menu: rewrite one-line item names into two-sentence descriptions a stranger could order from, attach a photo to every top-20 seller, model options as modifiers rather than prose (“no onions” as a modifier is data; in a description it is a suggestion), and delete retired items instead of zero-pricing them.

The Catalog API works alongside the Inventory API so that items sold are automatically subtracted from stock — this is the mechanism that keeps an AI agent from offering an item that is actually gone. But the automation has a precondition most local merchants have never thought about: Square tracks stock as a running ledger of adjustments since the last physical count. If you have never done a physical count, the system starts from zero — and every downstream channel, the AI surface included, inherits that fiction.

The state model is simple enough to reason about directly. Inventory sits in defined states: IN_STOCK (available, bidirectional), SOLD (terminal — units exit tracking when a sale completes), NONE (the default for newly added variations, which cannot be returned to), plus WASTE and UNLINKED_RETURN for shrinkage and returns not tied to a transaction. When an order completes through Square’s own systems, inventory moves from IN_STOCK to SOLD with no manual step. Under the hood, changes are processed as atomic batches sequenced by client-generated RFC 3339 timestamps — which is how Square reconciles offline transactions and out-of-order updates without corrupting the count.

The pre-launch inventory checklist is therefore short: do one honest physical count so the ledger has a real baseline; enable tracking on the items you actually run out of (the daily-batch pastries, not the sugar packets); and if any non-Square system touches stock, make sure something pushes those adjustments into Square. Merchants juggling more than two channels should think about this structurally — our multichannel inventory sync decision matrix covers when a dedicated sync layer earns its keep. The AI channel raises the stakes on accuracy for a specific reason: a human browsing your site might forgive an out-of-stock click, but an agent that confidently offers an item you cannot make turns a failed transaction into a trust problem with the customer and the assistant’s ranking of your business.

Fulfillment for AI-channel orders runs on the rails you already have. One framing note before the settings: Square’s Orders API models three fulfillment types generally — PICKUP, SHIPMENT, and DELIVERY, each with its own detail object — and a single order created through it carries only one fulfillment. Square’s launch materials, however, describe the ChatGPT and Claude flow in terms of the merchant’s existing online-ordering profile, and coverage to date centers on pickup-style flows. Square has not published a definitive statement on whether AI-channel orders can dispatch delivery, so treat the delivery-partner section that follows as your general Square Online delivery setup — not as a confirmed AI-ordering feature.

Pickup, the fulfillment path that configures automatically when your location has a valid address, comes with three settings worth deliberately configuring rather than defaulting — all from Square’s pickup-options documentation:

• Prep-time buffer. You set the number of minutes you need to prepare an order for pickup. Set it to your real kitchen pace at peak, not your aspirational one — an agent will relay whatever promise this field makes.
• Order cutoff. An optional cutoff time after which incoming orders roll to the next day — the difference between a late-night agent order becoming tomorrow’s first ticket versus tonight’s unnoticed one.
• Per-window caps. You can cap orders per 15-minute pickup window, which is your throughput protection if a new channel adds volume faster than the line can absorb.

Two more settings function as economic guardrails: a minimum order amount and a per-order item and dollar cap. These exist for any online ordering, but they read differently once software can place orders — they are your bounds on an agent submitting an unusually large or trivially small basket. Set both deliberately before counting on the channel.

For delivery beyond pickup, Square merchants connect a third-party courier network from Square Dashboard → Settings → App Integrations → App Marketplace. The options include DoorDash Drive, Uber/Postmates, Nash, and Deliverect. The economics of the DoorDash Drive route are a long-standing Square Online product fact — this integration predates the July AI-ordering launch — and the structure matters: Square charges a flat $1.50 per order for DoorDash Drive-fulfilled deliveries dispatched through Square Online, on top of DoorDash’s own flat delivery fee (an amount Square does not specify). Fixed cost per order, not a percentage-of-basket commission.

Processing fees are likewise unchanged by the AI launch — there is no separate “AI fee” tier. Square’s standard online rates are 3.3% + $0.30 per transaction on the Free plan and 2.9% + $0.30 on the Plus and Premium plans, and Square takes no marketplace commission on AI-channel orders. Run the math on a $40 order: pickup on the Plus plan costs $1.46 in fees (2.9% × $40 = $1.16, plus $0.30) — about 3.7% of the order. Add DoorDash Drive dispatch and it is $2.96, about 7.4%, before DoorDash’s own unspecified flat fee. A marketplace app charging its published 15–30% commission range takes $6.00 to $12.00 from the same basket.

Two fine-print items from Square’s partner documentation:

• Gift cards don’t travel. Square Gift Cards cannot be redeemed on orders placed through the third-party delivery-partner integrations (DoorDash Drive, Uber/Postmates, Nash, Deliverect) — a real limitation if gift-card loyalty is part of your regulars’ habit.
• Cancellation economics have three phases. Cancel before courier dispatch and there is no fee — full refund. Cancel after dispatch but before pickup and the delivery service may charge a cancellation fee. After the courier picks up, the service keeps its delivery fee, and if the customer receives a full refund, the merchant absorbs the cost.

One adjacent layer to keep an eye on: payment. Square handles the money on its own rails today, but the broader agentic-checkout stack — tokenized credentials, agent-initiated payments, buyer verification — is being standardized across the industry. Our guide to agentic checkout tokenization covers where that layer is heading and why it matters even to merchants who never touch a payments API.

Existing coverage treats setup as “zero work,” which is true for eligibility and misleading for everything after it. The table below is our synthesis of Square’s support and developer documentation — the first checklist we are aware of that separates what Square automates from what is still your job.

On measurement: Square tags AI-channel orders as a distinct source inside its reporting, which means from day one you can compare this channel’s volume, basket size, and repeat behavior against walk-in, phone, and marketplace orders. Use it. The Eightx analysis makes a forward-looking argument worth hearing with its label attached: the firm recommends tagging and tracking AI-channel orders for a contribution baseline now, on its own reasoning that the current zero-commission pricing is adoption-seeding and may not be permanent. That is one analyst’s prediction, not a Square statement or any confirmed roadmap — but the underlying discipline costs nothing and pays off either way: if the channel stays free, you know what it is worth; if pricing ever changes, you negotiate from data instead of vibes.

One honest gap to close out the setup picture: Square has not published opt-out instructions as of this writing. The July 1 press release does not address declining AI-channel visibility at all, and we were unable to verify any of the specific dashboard paths circulating in secondary coverage. If you would rather not be discoverable through AI assistants, the reliable route today is Square support — not a settings toggle someone described on the internet.

Our own projection, clearly labeled as such: the merchants who win this channel over the next year will be the ones who treat catalog and inventory data as a maintained asset rather than a one-time cleanup — because every new agentic surface that launches will read from the same structured data, and the cost of being ready for the next one rounds to zero once the first cleanup is done. This is the operating model behind our ecommerce engagements, and for merchants wiring AI into operations more broadly, our AI transformation practice starts exactly here: data readiness before channel chasing. The early adopters already talk about it in those terms — Andrew Costaris, Digital VP at Partners Coffee, put it this way in a statement Square provided to VentureBeat: “With agentic commerce and AI tools working in the background, we're confident knowing that our business is being digitally discovered and is consistently growing in efficiency, while our customers can continue to enjoy a lo-fi, specialty coffee-first environment.”