AI Shopping Assistant vs Chatbot vs Site Search vs Agent: The Four Are Not the Same
Four things get sold under one name. A chatbot deflects tickets, an assistant sells, site search retrieves, and an external agent decides whether a shopper reaches your site at all.
Viren Inaniyan · September 23, 2026 · AI Agent Storefronts
An AI shopping assistant is a conversational sales agent on a merchant's own site that recommends products from that merchant's catalog and helps complete the purchase. A support chatbot resolves service questions. AI site search retrieves products for a query. An external agent, like ChatGPT or Alexa for Shopping, decides which brands get named before a shopper picks a destination at all.
Four different products, one vocabulary, and a category where the naming is doing real damage to buying decisions. What is an AI powered shopping assistant and how is it different from a chatbot drew 107 citations across five AI engines in September 2026, which tells you how many buyers are stuck on the first question.
The four, side by side
| Support chatbot | AI shopping assistant | AI site search | External agent | |
|---|---|---|---|---|
| Job | Resolve a question | Sell a product | Retrieve for a query | Decide the shortlist |
| Runs on | Your site | Your site | Your site | Someone else's surface |
| Triggered by | A problem | Intent to buy | A typed query | A shopper's question |
| Success metric | Deflection rate, CSAT | Conversion, AOV | Search conversion, null-result rate | Whether you are named |
| Sees | Order and policy data | Your full catalog | Your product index | Feeds, crawls, marketplaces |
| You control | Everything | Everything | Everything | Almost nothing |
| Typical vendors | Gorgias, Fin, Zendesk | Rep AI, Alhena, Envive, Dialog | Algolia, Constructor, Luigi's Box | not purchasable |
The column that people skip is the fourth, and it is the only one where being absent costs you the shopper entirely rather than costing you a percentage of a session.
Chatbot and assistant: the honest distinction
Both are a text box in the corner. The difference is what they are optimised for, and it shows up in the metric each vendor leads with.
Support-first products report deflection and CSAT. Gorgias, whose Shopping Assistant is the clearest case of a support company moving into selling, reports in its own September 2026 research that "1 in 7 (14%) Shopping Assistant conversations ends in an attributed order" and that the assistant "influences 1.9% of online-store GMV".
Sales-first products report conversion and basket. Rep AI publishes "10-30% Lift in CVR" and "16%+ Lift in AOV from AI-guided sales". Envive publishes a "4X AVg Conversion Lift" for visitors who engage with its storefront.
Read a vendor's own headline metric and you learn which product you are actually being sold. We took those claims apart in the audit of every published conversion number in this category, because the denominators behind them are not comparable.
In practice the two are converging. Most serious products now do support and selling in one interface. The question worth asking a vendor is not whether they do both but which one their pricing, their reporting and their roadmap are built around.
Assistant and site search: different shoppers, not different quality
Site search serves the shopper who can already name what they want. It fails on the shopper who can say "something warm for a wet commute that does not look like hiking gear" and cannot say "waterproof insulated jacket".
An assistant handles that second shopper by asking, narrowing, and proposing. It also handles the objection that kills the sale after retrieval succeeds, which no search box does: the sizing question, the returns question, the will-it-arrive-Friday question.
The line between them is blurring fastest of all four. Algolia, Constructor, Bloomreach and Luigi's Box all now ship conversational layers over their search infrastructure, and every pure-play assistant is building or buying retrieval. Assume in two years this is one product.
The external agent is the one nobody sells you
The first three columns are things you buy and install. The fourth is not.
When a shopper asks ChatGPT, Gemini, Perplexity, Copilot or Alexa for Shopping what to buy, an answer is generated that names a small set of brands. Nothing on your site takes part in that. Your assistant cannot intervene, because the shopper has not arrived and may never arrive. The shortlist has already been decided.
It is not a small surface, and it does not behave like search. In our own September 2026 capture across five engines and three countries, the answers to buying questions in a single ecommerce category drew on 1,825 citations spread over 419 domains, and the engines disagreed with each other almost completely about which sources to trust. We published that data in full in which AI shopping assistants the engines themselves cite.
The practical consequence: an onsite assistant improves the yield on traffic you already won. It does nothing about the traffic that was assigned to a competitor one step earlier.
Which one to buy first
Work backwards from where the loss actually is.
Qualified traffic arrives and does not convert. An onsite assistant addresses that directly, and this is the case where the vendor conversion claims, read carefully, are relevant to you.
Shoppers arrive, search, and leave empty-handed. That is a retrieval problem. Check your null-result rate before buying a conversational layer over a search index that is not working.
Support volume is drowning the team. That is a chatbot, and the deflection metrics are the ones that matter.
Your brand is not named when shoppers ask assistants what to buy. No onsite tool reaches that shopper. This is what Citation Rank measures and Share of Voice sizes, and House of Zelena moved from 22% to 56% Share of Voice on ChatGPT over six months of that work.
Most brands have more than one of these problems. Few have diagnosed which one is costing the most, and the order you fix them in changes the return on all of it. The free Citation Rank scan settles the fourth question in a day, without a call.
Sources
- Gorgias, "AI shopping assistants are now officially influencing sales", gorgias.com/research, captured 23 September 2026.
- Rep AI homepage, hellorep.ai, captured 23 September 2026.
- Envive homepage, envive.ai, captured 23 September 2026.
- Tru Commerce citation corpus, 26,629 citations across 250 prompts, 5 engines, US/UK/India, captured 16 September 2026.
FAQ
An AI shopping assistant is a conversational sales agent on a merchant's own site that recommends products from that merchant's catalog and helps complete the purchase. Its job is conversion. A support chatbot's job is resolution, which is why the two are measured with different metrics.
A chatbot answers questions about an order, a policy or a problem, and succeeds by resolving without a human. A shopping assistant narrows a shopper's choice and moves them to purchase, and succeeds by selling. Most products in market do both, but they are optimised for one.
They solve different halves of the same problem. Site search answers a query the shopper already knows how to phrase. An assistant handles the shopper who cannot phrase it, works through constraints in conversation, and handles objections. Several vendors now ship both behind one interface.
ChatGPT is an external agent, not an onsite assistant. The difference is whose catalog it sees and who owns the session. An onsite assistant only ever sells your products to traffic that already arrived. An external agent chooses which brands to name before the shopper picks a destination.
It depends on where the loss is. If qualified traffic arrives and does not convert, an onsite assistant addresses that directly. If your brand is not being named when shoppers ask assistants what to buy, no onsite tool reaches that shopper, because they never arrive.
Continue reading
September 23, 2026
Every Published AI Shopping Assistant Conversion Claim, Audited
Rep AI publishes 10-30% CVR lift. Envive publishes 4x. Alhena publishes 20% AOV. Gorgias publishes 14% of conversations. All four are probably true and none of them answer the question you are asking.
September 23, 2026
What It Actually Takes to Put an AI Shopping Assistant on Your Store
A script tag on a clean Shopify catalog is a two-day job. A custom stack with location-bound inventory and seventy attributes per SKU is a two-to-three week job. Here is what separates them.
September 23, 2026
An Onsite AI Assistant Does Four Jobs. Most Teams Buy It for One
Every vendor in this category sells conversion. The teams getting the most out of an onsite assistant are the ones using it as a demand asset, a support channel and a listening post at the same time.