Buyer guide
AI chatbot for Shopify: what to look for
By the RoutMe team. Last updated .
When choosing an AI chatbot for a Shopify store, look at four things: how it installs on your theme, what store data it can read and with which permissions, whether it is read-only by default, and how it hands off to a person. Ask for each answer in writing, and treat anything undated as not yet available.
Start with how it installs
There are two common routes into a Shopify store. A script snippet is pasted into the theme code, and it loads the chat widget on storefront pages. An app is installed from Shopify and may use theme app extensions. Shopify documents theme.liquid as the default layout file every theme includes, and recommends duplicating a theme before editing its code so you can discard changes. Ask a vendor which route they use, and whether anything requires checkout customization, which is a separate area.
RoutMe uses the snippet route and has no Shopify app today. The steps are in the Shopify setup guide.
Ask what data it reads, and with which permission
A chatbot is only as accurate as the data behind it. For a Shopify store the relevant sources are:
| Shopify API | What it covers | Access model |
|---|---|---|
| Storefront API | Products, collections, search, cart | Public data is available without a token; some features need a token |
| Admin API | Orders, products, customers, fulfillments | App access token with explicit scopes such as read_orders |
| Customer Account API | A signed-in customer's own orders and profile | OAuth 2.0 with the customer's consent |
Shopify's access scope documentation notes that read_orders covers orders created in the last 60 days, and that older orders need read_all_orders as well. A vendor who asks for write scopes to answer "where is my order" is asking for more than the job needs.
Prefer read-only
A chatbot on your public storefront speaks to anyone. If it can refund orders or edit addresses, a clever prompt becomes a business risk. A read-only design limits the damage of a mistake to a wrong sentence rather than a wrong action. Read what read-only AI means and how an agent looks up an order safely.
Check how answers are grounded
Ask whether answers come from live lookups, from documents you provide, or from the model's general knowledge. Good answers are a mix of the first two. RAG vs tools explains when each applies. Then try three tests: ask about an order that does not exist, ask for a policy your documents do not cover, and ask something unrelated to the store. A trustworthy bot says it cannot confirm, rather than improvising.
Test the handoff
Open a conversation, ask for a person, and see what happens. Does your team see the whole conversation? Does it arrive where your team already works? Is the customer told what happens next? The details are in human handoff best practices.
Look at the channels
Many Shopify shoppers use WhatsApp. If WhatsApp matters to you, check that the vendor uses the official WhatsApp Cloud API and enforces opt-in, the 24-hour customer service window and template rules. RoutMe enforces those as platform guardrails. See WhatsApp vs website chat.
Read the roadmap critically
Roadmaps are plans, not features. In RoutMe's case, turning a REST or OpenAPI API into MCP tools with one import, a custom components builder, web voice and a CopilotKit-based widget are in progress and labeled coming soon, and there is no native Shopify connector. A vendor that dates and labels its roadmap honestly is easier to trust.
A short evaluation checklist
- How is it installed, and can I roll it back?
- Which Shopify APIs and scopes does it need?
- Is it read-only by default?
- What does it do when it does not know?
- How does a person take over?
- Are lookups logged?
- What is live, and what is coming soon?
Sources
Explore the product: RoutMe for Shopify, solutions and the documentation.