Connecting an AI assistant to your shop gives it real access to your data and your tools. This article explains how that access works, what you control at all times, and what stays your responsibility once the connection is set up.
How does the connection work?
When you connect Claude or ChatGPT to your WiziShop account, the connection goes through the same authorization mechanisms that normally protect access to your shop. In practice, you're redirected to a WiziShop page where you approve the access yourself. At no point are your password or your login details shared with Claude, ChatGPT or any third party.
Once you approve the connection, the assistant receives limited, identified access. WiziShop knows exactly which assistant is connected to which shop, with which permissions, and since when. This access isn't permanent by default, and it can't extend itself: the assistant can't go and get a permission mid-conversation that it wasn't given at the start.
💡 Good to know
This works on the same principle as allowing a third-party app to access your Google or Facebook account. You give your approval once, for a specific scope, and you can withdraw it at any time without changing your password.
What access you grant
View your active connections
You can see at any time the list of assistants connected to your shop, the date they connected, and the permissions granted to each. If you manage several shops, each connection belongs to one shop only. Connecting an assistant to "Shop A" doesn't automatically give it access to "Shop B".
Restrict access
Not all permissions are the same. You can, for example, limit an assistant to read-only actions: it can check your statistics, your orders and your product listings, but can never edit them. It's a good option when you're starting out, while you get used to what the assistant offers before letting it make real changes.
Revoke access
Revoking cuts the connection immediately. The assistant can then no longer view or change your shop, exactly as if you'd removed a team member's or a third-party app's access. If you want to restore a revoked connection later, you have to set it up again from scratch.
⚠️ Watch Out
Any action that changes your shop (a price changed, a product created, an order status updated) always needs your explicit confirmation before it runs. A deletion goes further: it shows a preview of what's about to be deleted, then asks for your final confirmation. This reduces the risk of mistakes, but it doesn't remove it. You approve every sensitive step, so you stay the one who decides.
Control your AI agents' access
Each AI agent connected to your shop (Claude, ChatGPT or any other agent using MCP) can only reach the data and actions you allowed when you connected it. You stay in control of this access and can adjust it whenever you like.
Edit an agent's permissions
Go to My Account > My Shop > Manage Users [à vérifier]. Every connected AI agent appears there alongside the other people with access to your shop. Select the agent to see its permissions and edit them. For example, you can let it view your products while taking away its ability to edit them.
Remove an agent's access
From the same section, you can remove an agent's access completely. It can then no longer view or change your shop. To use it again later, just start the connection again from the AI section [à vérifier].
💡 Useful Tips
Only grant the permissions the agent needs for what you ask it to do. If you only use it to write product listings, there's no reason to give it access to your orders.
What stays your responsibility
WiziShop provides the MCP server that connects your shop to an AI assistant, but WiziShop doesn't provide support for Claude, ChatGPT or any other assistant. How they work, any bugs, their limits in interpreting instructions, and any question about using them are up to Anthropic, OpenAI or the provider concerned, not WiziShop.
Once an action has been approved and carried out on your shop (a price changed, a promo code activated, an order status updated, content published), it's treated exactly as if you'd done it yourself from your administration interface. WiziShop doesn't reverse an action made by mistake through an AI assistant, or the consequences of an instruction that was badly worded or misread by the assistant. There's no automatic restore feature for these cases.
If a mistake happens, you fix it like any mistake made by hand: you take back control of the action concerned, yourself or through your assistant, and correct it again. A wrong price can be corrected, an incorrect order status can be changed back, but WiziShop doesn't technically "undo" the original action.
So you remain responsible for what the assistant does in your name, just as you would be for anyone you gave access to your shop. Delegating a task to AI doesn't delegate responsibility for the result.
Be careful with generated content
Some content needs extra care before you publish it, beyond a simple proofread.
An AI-generated image can give a misleading impression of the real product, with a color, a size or a material that differs from what will actually be delivered. Ask your assistant to flag any image that departs noticeably from the original product.
A customer review copied word for word from another platform is neither allowed nor advisable. Always ask for a rewording or a summary rather than an exact copy.
False urgency or false scarcity (a countdown that matches nothing, a stock level shown as limited when it isn't) puts your shop at risk of non-compliance. A countdown must reflect a real deadline, and the stock you display must match your actual stock.
Best practices
A few habits noticeably cut the risk of mistakes day to day.
Always read what the assistant is about to do before you approve it, especially for an action that touches a price, a stock level, a promo code or an order status.
Give precise instructions rather than vague ones. A request like "lower this price a bit" leaves room for an interpretation that isn't what you had in mind. "Lower the price by €5" leaves no room for doubt.
Limit access to what's strictly needed. If you only need to check your statistics, a read-only connection is enough.
Revoke a connection you no longer use rather than leaving it active for no reason.
Test on low-stakes actions first, such as a secondary product listing or a statistic, before trusting the assistant with tasks that directly affect your prices or your current orders.
Always ask your assistant to tell you what's published and what's still a draft. That goes for any action your visitors can see, not just a design change: a new category, a marketing text, a generated image.
👉 Resources
I create and manage my shop with AI (Claude, ChatGPT, Mistral)
What the agent can do (products, design, orders…) and its limits
Practical use cases
💬 Feel free to contact your Business Coaches if you have any questions!