How to choose an AI support assistant for your online store

Once you've decided your store needs a support assistant, there's a strong pull to grab the cheapest one you can find, paste in the code, and tick the box. I'd gently talk you out of that.

The wrong assistant doesn't just underperform — it actively costs you, because customers judge your entire store by the answers it gives. So here's how I'd choose one, in plain terms, whether you end up building it, buying it off the shelf, or having someone set it up for you.

Better no bot than a bad one

This is the part people skip, and it matters most. A support assistant that gives a confident wrong answer — invents a delivery time, promises a returns policy you don't offer, or replies "sorry, I don't understand" to a perfectly normal question — quietly teaches your customers to stop trusting your store. That damage is real, and you rarely see it happen.

A cheap, simple bot that answers badly will lose you more than the money it saved. Put bluntly: if the only thing in your budget is a bot that guesses, you're genuinely better off with a clear FAQ page and an email address. The goal was never "a bot." It's a reliable one — smart enough to help, honest enough to admit when it can't, and good enough that you'd happily stand behind every answer it gives.

Choose for your store, not the hype

The right pick depends on your situation, not on whose feature list is longest. Before you compare anything, get clear on your own store:

• How much support you actually get — and when (evenings? weekends? overnight?)
• Which languages your customers really use
• What they ask about most — delivery, sizes, ingredients, returns
• How often your products and policies change

A shop selling one product to one country needs something very different from a 500-product store shipping worldwide. Buy for yours — and for where you're heading, not just where you are today.

The checklist — what to insist on

Whatever you're weighing up, don't accept anything that can't do all of this:

• Uses approved store information. Grounding and guardrails should reduce unsupported answers.
• Has a clear handoff for uncertainty. It should say when information is unavailable and pass the question to a person.
• Supports the languages your customers use — with important answers configured and tested.
• Captures a name and email when someone isn't ready to buy yet.
• Fits your technical setup and has a clear process for keeping information current.
• Can grow with you — with deliberate updates and testing as you add products, channels and traffic.

If a tool wobbles on the first two points especially, walk away. Taking "just anything" here does more harm than good — an assistant that makes things up burns more trust than it will ever save you in time.