Personalization
AI ecommerce personalization
What to personalize, what to skip, and how to measure it with a holdout group.
AI and conversion
AI shopping assistant: when to add one to your store, and how to test it
An AI shopping assistant is a conversational tool that helps people find, compare and choose products. There are two kinds a store owner has to think about: the assistant you could add to your own store, and the assistants shoppers already use, such as ChatGPT, Perplexity and Amazon's Rufus, which decide which products to recommend from across the web. This guide covers both.
Short answer
An AI shopping assistant is a chat tool, usually built on a large language model, that answers shoppers' questions in plain language and helps them find, compare and choose products. It can run on one store's site or be a general assistant used across many sellers. Its answers are only as good as the product data and policies behind it.
- Stores with a large or technical range, many questions before purchase and weak site search are the most likely to benefit.
- Answer the common questions on the product page first, and let the assistant cover the long tail.
- Set firm rules for what it may say, and make clear to shoppers that they are talking to an AI.
- A tribunal held Air Canada responsible for what its website chatbot told a customer, so plan as if you answer for yours.
- Judge it with a holdout test on revenue per visitor, not chats started or assisted revenue.
What is an AI shopping assistant?
An AI shopping assistant is a conversational tool, usually built on a large language model, that helps people find, compare and choose products by answering questions in plain language. It can run on a single store's website, or it can be a general assistant a shopper uses to research products from many sellers before deciding where to buy.
The two kinds need different decisions from a store owner. With your own assistant, you choose whether to add one, what it may say and how to measure it. With the assistants shoppers bring, you do not control the conversation at all: you control only the information on your site that they read. Both are covered below.
| Question | Your store's own assistant | The shopper's own assistant |
|---|---|---|
| Examples | A chat or guided-finder tool added to your site | ChatGPT, Perplexity, Google's AI features in search, Amazon's Rufus on Amazon |
| Who controls it? | You and your vendor | The AI company or marketplace |
| What it reads | The catalog, policies and instructions you give it | Your public pages, structured data, product feeds, reviews and what others say about you |
| What you decide | Whether to add one, what it may say, how to test it | How readable, complete and consistent your product information is |
| How you measure it | Holdout test on revenue per visitor | Referral traffic and sales from AI assistants, and whether they describe you accurately |
What does an AI shopping assistant do on a store?
On a store, an AI shopping assistant answers product questions, asks what the shopper needs and suggests matching products, compares options side by side, helps with size, fit and compatibility, explains shipping and returns, and on some platforms adds items to the cart. It does all of this only as well as the product data and policies it is given.
- Answer product questions. "Is this jacket waterproof or water resistant?" "Will this lens fit my camera?"
- Guide the choice. Ask two or three questions, then narrow a large range to a few suitable products, like a good salesperson or a product quiz.
- Compare. Lay out the differences between two or three products the shopper is deciding between.
- Explain policies. Shipping times, return windows, warranties and how to get help, quoted from your real policies.
- Hand over. Pass the conversation to a person when the question is about an order, a complaint or anything it should not answer.
AI shopping assistant vs ecommerce chatbot
Older ecommerce chatbots follow scripted decision trees: the shopper clicks a button, the bot gives a fixed answer. An AI shopping assistant uses a language model to understand open questions and write answers from your catalog and policies. It handles more questions, and it can also get more of them wrong. What it is allowed to answer from matters more than the label.
Does your store need an AI shopping assistant?
A store is most likely to benefit from an AI shopping assistant when it has a large or technical range, shoppers ask many questions before buying, and site search often fails. It is least likely to benefit when the range is small, purchases are simple, traffic is too low to test, or product pages are missing basic information.
| Factor | May help | Probably will not help |
|---|---|---|
| Range | Hundreds or thousands of products, many similar | A few dozen products a shopper can browse in minutes |
| Purchase type | Considered or technical: fit, compatibility, specifications | Simple or impulse purchases |
| Pre-sale questions | Your inbox and chat logs are full of the same product questions | Shoppers rarely ask anything before buying |
| Site search | Many searches return nothing or the wrong products | Search already finds what people look for |
| Product data | Complete attributes, sizes, materials and policies | Thin descriptions the assistant would have to guess around |
| Traffic | Enough to test against a holdout group in a few weeks | Too little to tell whether it helped |
If most rows fall in the right-hand column, the better first move is usually to answer the common questions on the product page itself. That helps every visitor, including those who never open a chat window, and it helps the AI assistants that read your pages from outside. A CRO audit shows what is missing.
How to add an AI shopping assistant in 7 steps
To add an AI shopping assistant, start from the questions shoppers already ask, fix the product pages, clean your product data and policies, set firm rules for what the assistant may say, test it yourself with hard questions, launch it as a holdout test measured on revenue per visitor, and review its conversations every week.
-
Collect the questions shoppers already ask
Read your support inbox, chat logs, product reviews and site search terms. List the questions people ask before buying. This list tells you whether an assistant is needed, and it becomes the test script for any assistant you try.
-
Answer the common ones on the page first
Add the top answers to product pages, size guides and FAQs. Information that sits on the page helps every visitor and every outside AI assistant. An assistant should cover the long tail of questions, not hide the basics in a chat window. The Factory to Home ecommerce case study tested exactly this: shipping, warranty and payment FAQs on the product page.
-
Clean your product data and policies
An assistant only knows what you give it. Complete attributes, dimensions, materials, compatibility, stock status, and current shipping, returns and warranty policies. Gaps here become wrong or vague answers.
-
Set firm rules for what it may say
Decide what it must never do: promise discounts, make exceptions to policies, give medical, legal or safety advice, or guess when it does not know. Set when it hands over to a person, and make clear to shoppers that they are talking to an AI.
-
Test it yourself with hard questions
Before launch, put your question list to it, plus awkward ones: questions about products you do not sell, policy edge cases, and attempts to get a discount. Record every wrong answer and fix the source data or the rules.
-
Launch it as a test with a holdout group
Show the assistant to a random share of visitors and not to the rest. Compare revenue per visitor and conversion rate between the groups over full weeks. Plan the sample size first with the A/B test sample size calculator.
-
Review conversations every week
Read a sample of transcripts. Look for wrong answers, questions it could not handle and products people asked for that you do not stock. Each one is either a fix for the assistant or a finding for your product pages and range.
The risks of an AI shopping assistant
The main risks of an AI shopping assistant are confident wrong answers, especially about policies and product facts; legal exposure for what it says; disclosure rules where you sell; slower pages from extra scripts; personal data in chat transcripts; and useful information that ends up inside a chat window instead of on the page, where most visitors and outside AI assistants never see it.
You answer for what it says
In Moffatt v. Air Canada (2024 BCCRT 149, decided 14 February 2024), a British Columbia tribunal found Air Canada responsible for wrong bereavement fare information its website chatbot gave a customer. The tribunal rejected the idea that the chatbot was separate from the rest of the airline's website. Plan as if the same applies to you: limit what the assistant may answer and ground it in your real policies.
Shoppers may need to be told it is AI
The EU's AI Act requires providers of AI systems that talk directly with people to make sure those people are informed they are interacting with an AI system, unless it is obvious (Regulation (EU) 2024/1689, Article 50). If you sell in the EU, ask your vendor how it meets this, and label the assistant clearly anyway.
It can slow the page
Chat widgets add scripts. Check page speed on a phone before and after, and load the assistant in a way that does not delay the product page itself.
Transcripts contain personal data
Shoppers type names, order numbers, addresses and health details into chat windows. Know where transcripts are stored, who can read them and how long they are kept.
How to measure an AI shopping assistant
Measure an AI shopping assistant with a holdout test: compare revenue per visitor and conversion rate for visitors who can use it against a random group who cannot. Track answer quality and handovers alongside. Do not judge it by chats started or "assisted revenue", which include sales that would have happened anyway.
| Metric | Type | What it tells you |
|---|---|---|
| Revenue per visitor, assistant vs holdout | Main result | Whether the assistant adds revenue across all visitors, not only those who chat |
| Conversion rate, assistant vs holdout | Main result | Whether more visitors buy |
| Average order value and returns | Side effects | Whether it pushes bigger baskets that come back |
| Wrong answer rate | Quality | From weekly transcript review: how often it says something untrue |
| Handover rate | Quality | How often it passes to a person, and whether that is too often or too rarely |
| Pre-sale support tickets | Cost | Whether it takes load off your team or creates new questions |
| Chats started, assisted revenue | Usage only | How much it is used. Not proof that it helped |
Agent readiness
How shoppers' own AI assistants choose products
When a shopper asks a general AI assistant for a product, it builds its answer from what it can read about you: your pages, structured data, feeds, reviews and what other sites say. You cannot control the answer, but you can control how clear and complete that information is.
- product.name
- Linen duvet cover, queen
- PASS
- offers.price
- 89.00
- PASS
- offers.availability
- InStock
- PASS
- material
- null
- MISSING
- dimensions
- null
- FAIL: IN IMAGE ONLY
- returnPolicy
- null
- FAIL: FOOTER LINK ONLY
The panel is an illustrative example of what an outside AI assistant can read from a product page's HTML and structured data. The name, price and stock are there. The dimensions are only in a size chart image and the returns policy sits behind a footer link, so the assistant cannot quote either.
What you can do
- Put product names, prices, stock, sizes and materials in the page HTML as text, not only in images or scripts.
- Keep structured data accurate and matching what the page shows.
- State shipping, returns and warranty terms in plain text an assistant can quote.
- Keep product information consistent across your site, product feeds and marketplaces.
- Decide which AI crawlers you allow in your robots file, and check your firewall is not blocking the ones you want.
This is a new and fast-moving area, and nobody has years of data on it yet, including Convertica. The guide to agentic commerce explains how AI agents shop on a person's behalf. The agent readiness checklist and the guide to llms.txt cover the practical checks.
Audit and advisory
Where Convertica fits
Convertica has worked on conversion for online stores since 2017, with more than 2,000 clients. The free CRO audit, an app you can run now, checks your page for people and for AI agents, including whether an AI assistant can read what you sell, the price and the policies. CRO advisory, led personally by founder Kurt Philip with Convertica's CRO team, covers whether an on-site assistant is worth testing, how to measure it, and how visible your products are in AI search and AI assistants (see generative engine optimization services). In advisory your own team builds the changes; with full implementation, Convertica's team builds the fixes from your audit. Both are priced after your audit.
In this series
More on AI and conversion
Testing
AI A/B testing
What AI can and cannot do for your tests, and a 7-step workflow.
Checklist
Agent readiness checklist
Check what AI agents can read on your product pages.
Questions about AI shopping assistants
What is an AI shopping assistant?
An AI shopping assistant is a conversational tool, usually built on a large language model, that helps shoppers find, compare and choose products by answering questions in plain language. Some run on a single store's website. Others, such as ChatGPT, Perplexity and Amazon's Rufus, are used by shoppers to research products from many sellers.
What does an AI shopping assistant do?
An AI shopping assistant answers product questions, asks about needs and suggests matching products, compares options, helps with size, fit and compatibility, explains shipping and return policies, and on some stores adds items to the cart. How well it does this depends on the product data and policies it is given.
What is the difference between an AI shopping assistant and a chatbot?
Older ecommerce chatbots follow scripted decision trees and answer a fixed set of questions. An AI shopping assistant uses a language model to understand open questions and answer from your catalog and policies. The line between them has blurred, and what matters most is what the assistant is allowed to answer from.
Do AI shopping assistants increase conversions?
They can, for some stores, but there is no reliable general figure, and vendor numbers usually count sales that would have happened anyway. The only way to know for your store is to show the assistant to a random share of visitors and compare revenue per visitor with a holdout group that does not see it.
What is the best AI shopping assistant?
There is no single best one. For a store, the right assistant is the one that answers accurately from your own product data and policies, hands over to a person when it should, does not slow your pages, and lets you test it against a holdout group. Judge it on transcripts and revenue per visitor, not on a demo.
Is there a free AI shopping assistant?
For shoppers, several general AI assistants can help with product research at no cost, with features that vary by plan. For store owners, some ecommerce platforms include assistant features in their plans and specialist tools charge separately. Whatever the price, the cost of wrong answers to customers also counts.
Is a store responsible for what its AI shopping assistant says?
Generally, treat it as if you are. In Moffatt v. Air Canada (2024), a Canadian tribunal held the airline responsible for wrong refund information its website chatbot gave a customer. Limit what your assistant may answer, ground it in your real policies, and review its transcripts.
How do I get recommended by AI shopping assistants like ChatGPT?
Make your product information easy for AI systems to read and trust: complete product details, prices and stock in the page HTML, accurate structured data, return and shipping policies in plain text, consistent information across your site and product feeds, and robots rules that allow the AI crawlers you want.