You need a backpack for day hikes. It should cope with a bit of rain, carry a jacket and a water bottle, stay comfortable for a few hours and still be compact enough to take on a train. You'd rather spend less than €100.
That's a perfectly normal thing to say to someone in a shop. You don't need to know the ideal capacity in litres or whether the fabric has to be fully waterproof. A member of staff who knows the range can ask for one missing detail and point you towards a couple of sensible options.
Buying the same thing online often puts more work on the shopper. You open the backpack category, work out which specifications matter, set a few filters and compare several product pages before the original requirement turns into an actual product. An AI shopping assistant is useful in that gap, when the shopper already understands the problem they're trying to solve but hasn't yet identified the product.
What an AI shopping assistant does
An AI shopping assistant lets shoppers describe what they need in ordinary language and uses that context to narrow a catalogue. Depending on the product, it can consider requirements such as intended use, budget, preferences and practical constraints before suggesting suitable options.
Our backpack request already says a lot: day hikes say how much it needs to carry, the rain means it has to keep things dry, the €100 rules out part of the range and the train makes size matter more. None of this requires the shopper to know which filter names the retailer has chosen.
ChatGPT's shopping research feature works in a similar way, across the whole web rather than one shop. OpenAI says it's designed for purchases involving comparisons, trade-offs or several constraints, and it asks follow-up questions about needs, preferences and budget to refine its recommendations. Inside a retailer's own site the job is narrower: the conversation has to lead back to products the retailer can actually sell.
Search, filters and assistance do different jobs
Online stores have spent years improving search and filtering, and for good reason. If you know the exact model you want, typing its name is hard to beat. If you need a black jacket in medium for less than €100, a handful of filters will take you to the right part of the catalogue in seconds.
The backpack sits in a third category. You know the outcome you want, but you may not know whether that means 18 or 25 litres, whether it needs to be fully waterproof or just shower-resistant, or which models strike the right balance between hiking and travel.
These paths overlap, and a good store needs more than one of them. Baymard Institute's research into ecommerce product lists and filtering shows how important filtering and sorting remain when shoppers need to bring a large range down to a manageable set of relevant products. Assisted shopping comes in earlier, when the shopper hasn't yet worked out which attributes to filter by.
The useful part of the shop-floor experience
There's little reason to pretend physical shopping was better at everything. Ecommerce gives people larger ranges, easier comparisons and products that would never fit on a local shop's shelves, and a salesperson can give poor advice too. One part of the shop-floor experience is still worth keeping, though: being able to explain your situation before you know the answer. The customer can talk about a day hike, rain, what they need to carry and their budget, and a knowledgeable member of staff connects all that to the range without asking them to learn the technical vocabulary first.
Online, shoppers usually do that translation themselves, working through buying guides, filters and product pages until they understand how their need maps onto the catalogue. Assisted shopping gives some of that work back to the store. The customer still makes the choice, but the retailer takes a more active part in narrowing it.
Only recommend what the shop sells
A recommendation inside an ecommerce store has one obvious constraint: the retailer must actually sell the product. Suppose the ideal answer to our backpack request would be a lightweight 20-litre model under €100 with some weather protection. If the shop carries nothing like that, naming a plausible backpack sold elsewhere doesn't help the customer buy from that shop.
SmartShop AI only recommends products that exist in the merchant's catalogue, and when nothing fits it says so rather than inventing an answer. It also excludes out-of-stock products from its recommendations, so shoppers only see what they can buy right now. A good member of staff works the same way: they can understand exactly what would suit you and still tell you they don't stock it.
What the conversation could look like
A shopping assistant shouldn't turn a simple purchase into an interview. When the shopper has given enough information, there's little value in asking questions for the sake of it; one clarification is worth it only when the answer would really change the recommendation. For our backpack, the exchange might look like this:
- CustomerI'm looking for a backpack for day hikes. It should cope with some rain, not be too large and I'd like to stay under €100.
- AssistantWill you mainly use it on trails, or do you also want something comfortable for travelling and getting around town?
- CustomerMostly easy trails, but I'd like to take it travelling too.
- AssistantI have two that fit well: one is lighter and more compact, while the other gives you a little more room and better rain protection without becoming bulky.
The two backpacks appear as product cards below the reply, each with its own add-to-basket button.
What the conversation does is modest but concrete: a fairly loose request has become two backpacks the shop actually stocks. Sending the shopper another page of twenty would hand most of the decision straight back to them.
Sometimes search should win
Imagine the same customer comes back later to replace the backpack and knows the exact model. Search is probably the quickest route. The same applies when two straightforward filters solve the problem or when the catalogue only has a handful of products.
Conversation can create friction too: asking five questions before showing a product somebody has already named would make the store harder to use. That's why finding a product and choosing one deserve to be treated as related but distinct tasks. The right tool depends on how much the shopper already knows.
Another route through the catalogue
Conversational shopping is also showing up outside individual stores. In March 2026, OpenAI extended its Agentic Commerce Protocol, the standard merchants use to share product data with ChatGPT, to product discovery, so that ChatGPT can draw on fuller, more up-to-date product information. That doesn't mean every store needs to turn its catalogue into a chat window: navigation, filters, search and product pages still handle a large share of shopping journeys extremely well. The change is that they no longer have to be the only ways in.
SmartShop AI, a sales assistant for Shopify and WooCommerce stores, brings assisted shopping to each merchant's own store.
Ecommerce has become very good at helping people find products once they know what to look for. Assisted shopping deals with a different moment: when the customer has explained what they want but still needs help choosing.