Shopify AI Agents, Chatbots and Agentic Commerce

Say which of the two jobs you are hiring for. A storefront assistant is conversation design and knowledge writing. Agentic commerce is product data and channel setup. Some specialists do both; ask which they have actually shipped on a live store.

Ask for real transcripts, not a demo. A demo answers the questions the builder picked. Anonymised transcripts from a live store show how the assistant handles vague questions, off-topic ones and the moments it should hand off.

Make your own questions the acceptance test. Pull 20 to 30 real pre-purchase questions from your inbox: sizing, ingredients, compatibility, delivery times. The assistant goes live when it answers them correctly or hands them off cleanly.

Get the rules written before the prompts. What may it never do? Typical answers: offer unapproved discounts, make medical or safety claims, promise delivery dates, change an order. Ask to see that list, and how each rule gets tested.

Check the hand-off. When the assistant can’t answer, the shopper should reach a person, leave an email or open a ticket in one step. Out of hours, the fallback should say when a person will reply. A dead end is worse than no assistant at all.

Ask what customer data it can see. A pre-purchase assistant rarely needs order history or customer records. Less access means less to go wrong.

Red flags. Conversion claims with no transcripts to back them. “Trained on your store” with no explanation of what that means. No plan to review conversations after launch.

The on-site assistant. Try Shopify Inbox’s AI agent before paying for a custom build. Bring in a specialist when answers depend on data Inbox doesn’t read, when you want guided selling across a large catalogue, or when the assistant needs to take actions such as building a cart.

Whichever you use, the knowledge the agent answers from matters more than the model. A sizing chart written as plain text, shipping cut-offs by region and returns rules with the exceptions spelled out do more for answer quality than switching models. When an answer is wrong, check the source content first.

Agentic commerce. Through Agentic Storefronts, eligible stores can sell inside AI channels. Shopify’s help docs currently list ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, managed under Sales channels → Agentic. Eligibility is mostly based on US buyers, so check what your store can switch on before scoping any work.

Shopify Catalog is the product data layer those channels read, and data quality affects whether you are shown. Specific titles, complete attributes, clean variants, GTINs and good images matter more than anything clever.

A title that reads well to people but leaves out size, material or compatibility gives an AI channel little to match against a shopper’s request. This is catalogue work, and the size and state of your product data drive the effort.

Budget for two things: the setup, and the monthly work of reading conversations and fixing answers. Setup and tuning below are typical market ranges, not directory data; the last tile is what the 15 listed specialists publish as starting prices (September 2026).

The work splits between the assistant on your storefront and the product data AI channels read.