creel
For Shopify stores
Agents are buying. Are they buying from you?
AI agents now shop and pay for people. Ask one for a lotion and it names three, picks one, and buys it.
Your agent revenue is two numbers multiplied: how often agents pick you, and how often they can finish the purchase on your store. Most stores lose on both, and never see it.
Paste your store. In about 30 seconds you see which pages respond, what an agent can read, and what needs a closer look.
The agents people already buy with
ChatGPT agent
OpenAI
Browses your store, fills the cart, hands over at payment.
Gemini · AI Mode
Picks from the Shopping Graph and checks out on select stores.
Claude
Anthropic
Drives Chrome on the shopper's own machine.
Copilot
Microsoft
Copilot Mode in Edge browses and shops for the shopper.
Grok Bot
xAI
Its own cloud computer. Pays with Stripe Link once the shopper approves.
Comet
Perplexity
A free agent browser that shops end to end.
Alexa+ · Buy for Me
Amazon
Buys from your site with the Amazon wallet.
Instinct
Spear Street
A personal agent that acts without asking first.
OpenClaw
open source
A self-hosted personal agent with checkout skills.
Manus
Manus
A general agent that browses and buys.
Opera Neon
Opera
An agentic browser that navigates stores and fills forms.
Meta · Siri
due this fall
Announced, not shipped: Meta's personal agent and the rebuilt Siri.
As of September 2026. Behind them, the rails: Stripe Link, PayPal, Visa and Mastercard agent payments, Google UCP, Shopify Agentic Storefronts.
Agent revenue = picked × bought
Picked. Of the buy prompts in your category, how often an agent puts your product in its answer and sends the buyer to your store rather than a marketplace.
Bought. Once it picks you, how often it gets from your product page to the order.
Both are counts from dated, recorded runs. Each stop below costs you one of them.
Where you lose it
One ask, seven stages. At each one, something a shopper never notices can stop an agent.
0 Picked
"Recommend me a lotion." The agent names yours and points the buyer at your store.
Can stop here: your product missing from the feeds and catalogs agents read, price, stock, shipping or returns it cannot read, or a marketplace listing it finds first.
1 Found
The agent reaches your store and gets the same page a browser would.
Can stop here: a bot manager challenge, a robots.txt disallow, or a CAPTCHA on the first page.
2 Product
It picks the right item, variant and price, and reads shipping and returns.
Can stop here: product information it cannot read, or a variant picker it cannot work.
3 Cart
The cart holds what was asked for and nothing else.
Can stop here: an overlay it cannot close, a pre-checked add-on or subscription, or a blocked cart page.
4 Checkout
It reaches checkout as a guest and enters an address.
Can stop here: forced sign-up, an address field that rejects typed input, or a CAPTCHA.
5 Payment
It sees the correct total at the payment step.
Can stop here: a total that changed since the cart, or a challenge at the payment step.
6 Ordered
The order is placed. We only go this far in a separately agreed real-agent run, with your go-ahead.
Can stop here: an unsupported payment method or a card challenge that requires the buyer.
A stopped agent may need the shopper to take over, or find another way to buy. These interruptions can be hard to see in ordinary store analytics.
Three checks, each one deeper
Quick check. Free, on this page, about 30 seconds. HTTP only, read-only. Whether four agent user-agent strings get the same pages a browser request gets, whether robots.txt lets them in, whether product data is readable, and whether cart and checkout URLs answer. Signals around stages 0 to 3.
Full pass. Ask for it after the quick check. A browser walks product, cart and checkout with a test address and stops at the payment step. Stage reached and screenshots by email. We confirm scope and timing first. Stages 1 to 5.
Real-agent run. From $2,500. Consumer agents driven by hand, one instruction each, screen recorded, plus unbranded buy prompts in your category to count how often you are picked and where the buyer is sent. The stage each agent reached and why, the proposed fix for each stop, and a re-run after your changes. Agents and scope agreed before payment. Stages 0 to 6. hello@usecreel.com
What the fix looks like, on Shopify
Each confirmed stop maps to a change on your side, and on Shopify most are theme, settings, robots.txt.liquid and app-level changes. Get your catalog into the feeds agents read: Google Merchant Center, the Shopify catalog that ChatGPT and others pull from, Perplexity's merchant program. Review how verified agents are handled by the bot manager and robots.txt. Make product, price, stock, shipping and returns readable. Open guest checkout where appropriate. Remove unwanted overlays and pre-checked add-ons from the buy path. You get a proposed change for each issue, and a re-run once it ships.
Evidence & examples · How we check
The name: a creel is the basket an angler keeps the catch in.
creel. Counts, not guarantees. Each report names what was tested. · hello@usecreel.com