For the last two years, the assumption in retail marketing has been simple: AI chatbots are eating search, so retailers need to worry about getting recommended by ChatGPT the way they used to worry about ranking on Google. New eMarketer data complicates that story in a useful way. When US shoppers are actually buying something, they'd rather use the retailer's own AI assistant than a general-purpose one.
That distinction matters more than it sounds like it should. It splits AI shopping into two separate jobs: one for browsing and ideas, another for the moment money is about to change hands. A general chatbot is fine for the first job. It has no real advantage at the second.
Two different jobs, two different tools
Ask ChatGPT for gift ideas or a comparison between two blenders and it'll give you a reasonable answer. It's trained on a huge amount of public information about products, categories and general consumer preferences. What it doesn't have is your order history, your return window, or a live read on whether the item you want is actually in stock at the warehouse fifteen miles from your house.
A retailer's own AI assistant has a narrower job but a deeper one. It knows the return policy because it's built on top of it. It can check real inventory instead of guessing. It can see that you bought the same brand of running shoes eight months ago and ask if you want the same size. None of that is exotic technology. It's just proximity to the data that actually decides a purchase.
US consumers say they'd rather use a retailer's own AI assistant than ChatGPT when buying. Source: eMarketer
Why this is good news, not more work
The easy read on this data is that small retailers now need to build a sophisticated AI assistant to compete. That's the wrong takeaway, and it's worth naming why. The finding isn't that shoppers need a flashy AI feature. It's that shoppers already trust the retailer's own channel more, at the exact moment that trust is worth the most.
The purchase moment was never up for grabs. Shoppers were always going to trust the business they were buying from over a general tool that's never seen their order history.
That's a relief for anyone who's been quietly worried that owning a website, a checkout, and a direct relationship with customers was becoming a legacy advantage instead of a real one. It isn't. The channel a small retailer already controls, its own site, its own emails, its own product pages, is still where the buying decision gets made. General AI chatbots are a discovery layer sitting on top of that, not a replacement for it.
What this actually changes for a small retailer
- Don't treat 'getting mentioned by ChatGPT' as the goal. Traffic that starts in a chatbot still has to land somewhere and finish the sale there.
- Put the order-history, return-policy, in-stock details where a buyer can find them fast, on your own site, because that's the trust a general chatbot can't replicate.
- Resist the pressure to build a complex AI assistant before you've made sure the basics, a clear page, a clear return policy, a fast checkout, are solid.
None of this requires a new department or a six-figure AI project. It requires the retailer's own site to actually do the job of closing the sale once someone lands there, whether they arrived from a search engine, an ad, or a chatbot that pointed them in your direction.
That's the layer DayClerk's audience simulation is built to strengthen before a single ad runs. It reads a business's brief and models how a segment will actually behave on the page, surfacing the friction points and drop-off risks that decide whether a visitor who lands on your site buys or leaves, regardless of what sent them there.
The eMarketer data is a point in favor of something a lot of small retailers already had right: the site you control is still where the sale gets made. The job now is making sure it earns that trust once someone actually arrives.