An AI chatbot on an eCommerce site is not much use if it answers like a generic chat window. It can be polite, fast and articulate, but if it does not know your products, availability, variants, reviews, facets, delivery, returns and commercial rules, it risks becoming just another widget on the site.
The point is not "having AI". The point is having an assistant that genuinely helps the user choose, removes doubts before the purchase and takes repetitive questions off the customer care team.
To do that, the chatbot has to be connected to the real catalogue. It has to know what is on sale, what is in stock, which products are similar, which differences matter, which questions customers ask most often, and when it should stop rather than invent an answer.
AI eCommerce chatbots: the difference between a demo and a sales tool
Plenty of AI chatbots look impressive for the first five minutes. They answer generic questions well, summarise text and hold a fluent conversation. Then the real question arrives:
- Is this product available in the variant I need?
- Which of these three models would you recommend?
- Is this item compatible with what I already own?
- How long do I have to return it?
- Which product has better reviews for my case?
- What is the alternative if this product is out of stock?
If the chatbot is not connected to the data, it answers vaguely. Connected properly, it can become a buying guide rather than just an automatic responder.
What an eCommerce chatbot needs to know to be useful
An assistant connected to the catalogue has to be able to read and use at least this data:
- products: title, description, images, features, variants and attributes;
- categories: where the product sits and which product families exist;
- facets: need, use, material, format, size, colour, price bracket, availability or other useful criteria;
- brands: the differences between makes, ranges and the most requested products;
- availability: stock, out-of-stock items, pre-orders, replacement products and restocking times where known;
- prices and promotions: without inventing discounts or conditions that are not on the site;
- reviews: what customers genuinely praise or criticise;
- product FAQs: recurring doubts already surfaced on product pages, in chat, in email and through customer care;
- policies: delivery, returns, warranty, payments, support and commercial rules;
- orders: only when the user is authenticated and permissions are clear, for order status, tracking or after-sales requests.
This is the difference between "a chatbot on the site" and an eCommerce virtual assistant. The first one answers. The second reasons over the shop's data.
The catalogue has to be clean, or the chatbot inherits the chaos
An AI chatbot plugged into a disorganised catalogue does not solve the problem. It amplifies it.
If the facets are wrong, if descriptions are thin, if brand names are written inconsistently, if variants are unclear, if stock levels are not synchronised, the assistant is working from weak foundations. It may phrase the answer well, but it starts from fragile data.
That is why the chatbot should come after, or alongside, serious work on the catalogue:
- tidy categories;
- useful, consistent facets;
- product pages with clear information;
- well-maintained brand pages;
- reviews and FAQs managed in a structured way;
- synchronised feeds and stock levels;
- clear rules on what it may and may not say.
We covered this in the article on automatic eCommerce facets with AI: when the catalogue is well organised, AI becomes far more useful. When the catalogue is confused, AI risks polishing the confusion.
A good chatbot should not always be selling
Here is an important point: an AI eCommerce assistant should not always say "buy this". It has to know when to recommend, when to clarify, when to offer alternatives and when to stop.
For example, it should be able to say:
- "this product may not suit your case";
- "I don't have enough information to recommend confidently";
- "check this specification before you buy";
- "this alternative matches what you asked for more closely";
- "this question needs a human agent".
That builds trust. A chatbot that always pushes the sale can look convenient in the short term, but if it produces the wrong purchases, returns or unrealistic expectations, it damages the shop.
The questions the chatbot should be able to handle
An eCommerce chatbot that is properly connected can help at several stages of the journey.
1. Product discovery
The user does not yet know what to buy. The assistant has to ask questions, understand the need, the budget, the use, preferences and constraints, then suggest categories or products that fit.
2. Comparing products
The user is torn between several products. The chatbot has to compare real specifications, reviews, differences, limitations and use cases. It should not simply say they are all excellent.
3. Doubts on the product page
The user is on the product page but has a specific question. Here the assistant can use the description, FAQs, reviews and attributes to answer quickly, as we described in the article on product pages with AI, reviews and FAQs.
4. Alternatives to out-of-stock products
If a product is unavailable, the chatbot can suggest similar alternatives by category, brand, price bracket, specification or reviews, without sending the user off the site.
5. After-sales support
When securely connected to orders, it can help with order status, tracking, returns information and practical instructions, leaving the more delicate cases to human support.
The chatbot has to be able to use tools, not just text
A genuinely useful assistant does more than "know things". It has to be able to query tools and up-to-date data.
For example:
- search the catalogue for products;
- filter by availability, brand, category and attributes;
- retrieve the current price and promotions;
- read aggregated reviews;
- show alternative products;
- open a ticket or hand the conversation to a human;
- retrieve order status only with correct authentication;
- log frequent questions to improve FAQs and product pages.
If the chatbot has no access to current data, it can answer well linguistically and badly commercially. And on an eCommerce site, commercial accuracy matters more than an elegant sentence.
Security: prompt injection, customer data and invented answers
A chatbot connected to company data has to be designed carefully. It is not enough to paste it into the site and hope it behaves.
The main risks are:
- invented answers: availability, discounts, delivery times or policies that do not exist;
- prompt injection: users trying to make it ignore instructions, policies or limits;
- improper data access: order or customer information shown to people who should not see it;
- unverified advice: especially in sectors where compatibility, safety or correct use matter;
- badly handled logs: conversations containing personal data stored with no clear criteria;
- automatic actions with too much freedom: cancellations, order changes or coupons granted without rules.
The answer is not to avoid AI. It is to design limits, controls, roles, logging, human escalation and properly separated data. An assistant has to know when it may answer, when it must ask for confirmation and when it has to pass the request to a person.
What to ask your agency before installing an AI chatbot
If someone proposes an AI chatbot for your eCommerce site, don't just ask "how much does it cost". Ask what it knows, what it can do and how it is controlled.
- Is the chatbot connected to the real catalogue, or does it only answer from static text?
- Does it know availability, variants, prices, promotions and out-of-stock products?
- Can it read reviews, product FAQs and frequently asked questions?
- Can it tell similar products apart and explain concrete differences?
- Can it suggest alternatives when a product is unavailable?
- How does it avoid inventing answers about discounts, delivery and returns?
- Which customer data can it see, and under what authorisation?
- How is prompt injection handled?
- When does it pass the conversation to a human agent?
- Are user questions used to improve product pages, FAQs and facets?
If the chatbot does not improve the catalogue, support and conversions, it is just a gadget. It may look modern, but it builds no advantage.
How we handle it at BitHub
At BitHub we see the AI chatbot as part of the platform, not as a separate accessory. It has to talk to the custom eCommerce build, the catalogue, facets, product pages, brands, reviews and commercial data.
Our work on AI and automation starts from the company's real data: products, categories, rules, content, customer questions and processes. From there you can build an assistant that does not merely converse, but helps people choose, cuts repetitive requests and gathers useful signals for improving the catalogue.
This also connects to the digital strategy side: the questions people put to the chatbot can reveal new doubts, products that are hard to understand, missing facets, pages that need improving and content worth creating.
A good AI chatbot does not replace the strategy. It makes it faster to apply, because it works every day on the questions users actually ask.
Useful sources
For further reading, useful material includes the Google documentation on AI-generated content, the guidelines on helpful, reliable content, Product structured data, the OWASP Top 10 for LLM Applications project and the OWASP Prompt Injection Prevention Cheat Sheet.
FAQs on AI chatbots for eCommerce
Does an AI eCommerce chatbot need to be connected to the catalogue?
Yes, if it is meant to genuinely help you sell. Without a connection to products, availability, variants, categories, reviews and policies, the chatbot can only answer generically and risks giving information of little use.
Can the chatbot replace customer care?
It can cut a lot of repetitive questions and help before the purchase, but it should not completely replace human support. Complaints, complex cases, personal data and sensitive requests need escalation to a person.
Can it recommend products automatically?
Yes, if it uses real data and clear rules. It has to be able to compare products, facets, availability, reviews and specifications. Without up-to-date data, it risks recommending products that are unsuitable or unavailable.
What is the main risk with an AI eCommerce chatbot?
The main risk is that it invents answers or acts on incorrect data: discounts that do not exist, wrong delivery times, policies that are not real, or improper access to customer data. You need limits, controls and logging.
Can chatbot conversations improve SEO?
Indirectly, yes. User questions can reveal real doubts, missing facets, incomplete product pages and FAQs worth adding. Those signals can improve content, categories and product pages.
Where is the best place to start?
With the most frequent questions and the categories that generate the most sales or the most doubts. First connect the catalogue, availability, FAQs and policies; then add more advanced functions such as product comparison, alternatives and after-sales support.