What to ask an AI assistant about your store
The questions worth asking an AI assistant connected to your ecommerce data, and what the answers reveal
Published
The Hello Retail MCP server is how you connect an AI assistant to your store data. Once it is connected, the first questions worth asking are the ones that would otherwise take several dashboard reports. Which searches return nothing. Whether the last product feed synced. Which fields are searchable and sortable. What a template currently says. Each is answerable in one request.
This guide is the practical companion to our explanation of how MCP servers and APIs differ. It assumes the connection is already made and covers what to do with it.
Start by confirming what you are connected to
The first request should be to list the websites for your company. It is a small question that settles three things at once: the connection works, the assistant is looking at the right account, and you now have the site reference you will use in everything that follows. On the Hello Retail MCP server this needs a Company ID, found in my.helloretail.com under the profile icon and then Company Profile.
It is worth doing even when you are confident the setup is correct. Most confusion in the first session comes from an assistant answering accurately about a site you did not mean.
Four questions worth asking first
Which searches came back empty?
When a search comes back empty, a customer has told you exactly what they wanted and got a blank page. There are usually three explanations: you do not stock it, you do stock it but call it something else, or the search engine has not learned that the two words mean the same thing.
Read the list, then ask the follow-up that a dashboard makes awkward: are any of these products actually in the catalog already? That question crosses two tools, and it separates a merchandising gap from a findability problem.
What are people searching for most?
Top searches tell you what the store is actually for, as opposed to what the navigation assumes it is for. When the most frequent queries do not match the categories in your main menu, the menu is the thing that is wrong. This is also the fastest read on seasonal shift, since the top of the list moves weeks before revenue does.
Did the last product feed sync?
Feed problems are quiet. Nothing breaks visibly, prices simply drift out of date and new products fail to appear. Checking sync status takes one question, and it is worth asking before you conclude anything from the other three. An analysis of stale data is worse than no analysis, because it reads as authoritative.
Which fields can customers actually search and sort on?
Every store has product attributes it collects and never exposes. Asking which fields are searchable and sortable produces a list you can compare against what shoppers plausibly want to filter by. Material, fit, capacity, and compatibility are common gaps, and each one is a filter the store could have offered and did not.
Which tool answers which question
You do not need to name a tool when you ask. The assistant reads what is available and chooses. It is still useful to know the shape of what it can reach, because that is the boundary of what it can answer. The Hello Retail MCP server documents seven categories today, with more rolling out.
| What you want to know | Tool category it comes from |
|---|---|
| Which sites am I connected to | Website management |
| Empty searches, top searches, filter and sort usage | Search analytics |
| Whether the last feed synced | Feeds |
| What is indexed, and at what price | Product data |
| Which fields are searchable and sortable | Data fields |
| What a search template says, and changing it | Search |
| What a recommendation template says, and changing it | Recommendations |
How to ask a question the assistant can answer well
- •Name the site and the period. These are the two things the assistant cannot infer, and the two that most often make an answer misleading rather than wrong.
- •Ask for the reasoning alongside the answer. Asking which tools it used and what it found turns an answer you have to trust into one you can check.
- •Ask one question, then follow it. The value is in the second question, the one the first answer suggests. Bundling five questions into one request produces a summary rather than a finding.
- •Read before you approve a change. Write-enabled tools ask for confirmation. That prompt is the review step, so treat it as one rather than a formality.
What this is not for
An assistant reasoning its way to an answer is the right tool for investigating and the wrong one for anything that has to happen identically every time. The search box on your storefront, the recommendation slot on a product page, and the nightly feed job all belong in code against the API. Our guide to MCP servers and APIs covers where that line sits.
It is also not a substitute for knowing your catalog. The questions above surface problems quickly, and every one of them still needs a person to decide what the store should do about it. That decision is the part worth your afternoon, which is the argument for spending less of it clicking through reports. The same structured product data that makes these questions answerable is what makes a catalog legible to the buyer-side assistants described in our guide to agentic commerce.
Frequently asked questions
What should I ask an AI assistant connected to my ecommerce store?
Start with the questions you would otherwise have to open several dashboard reports to answer. Which searches return no results. Whether the last product feed synced. Which product and content fields are searchable and sortable. What a search or recommendation template currently says. These map directly onto the tool categories a connected assistant has available, so they are answerable in one request rather than a session of clicking.
What is the first thing to do after connecting an MCP server?
List the websites for your company. It confirms the connection works, it confirms the assistant is looking at the right account, and it returns the website you will refer to in every question afterwards. The Hello Retail setup guide suggests exactly this as a first request, using the Company ID found in my.helloretail.com under the profile icon and then Company Profile.
Can an AI assistant change my search configuration?
On the Hello Retail MCP server, yes, within limits. Most tools are read-only. The write-enabled ones, such as editing a search or recommendation template, ask for confirmation before they run, so a change is never made silently. In practice this means you can ask the assistant to draft a change, read what it proposes, and approve it as a separate step.
Why do zero-result searches matter?
Because the customer has told you what they wanted, in their own words, and got a blank page. The reason is either that you do not stock it, that you stock it under a different name than the one they used, or that the search engine has not learned that the two words mean the same thing. All three are fixable, and checking the list regularly is one of the cheapest sources of merchandising insight a store has.
Do I need to know the tool names to ask a good question?
No. The assistant reads the descriptions of the available tools when it connects and chooses which to call based on your question. Naming the website you mean and the period you care about matters far more than naming a tool, because those are the details the assistant cannot infer.
How is this different from an analytics dashboard?
A dashboard answers the questions it was designed to answer, in the shape its designer chose. An assistant with tool access can combine sources in response to a question nobody anticipated, such as asking whether the queries returning nothing correspond to products that are in the catalog but not indexed. The dashboard remains better for tracking a known number over time.
Connect your store and ask
The Hello Retail MCP server works with Claude, ChatGPT, and OpenAI Codex. Setup steps for each are in the developer documentation.
Read the MCP setup guide