AI for Online Stores: Answer Product Questions, Close the Sale

An AI sales assistant for an online store makes money by answering the question a shopper asks right before they buy: does it come in my size, is it in stock, what does shipping cost, can I send it back. In Salesforce's 2026 holiday research, 41% of shoppers said a brand-owned AI assistant answering their questions made them "much more confident" in a purchase, and retailers running their own shopper agents grew 2025 holiday sales by 6.2%, against 3.9% for retailers without one. The catch is accuracy. An assistant that guesses about stock or returns costs you more than no assistant at all.

Speed counts too: in Entagl's own data, the first seller to reply won 78.4% of competitive e-commerce inquiries that ended in a sale. This post covers which pre-purchase questions an AI should answer, which source each answer has to come from, and the lines a store's assistant should never cross.

Why do product questions decide online sales?

Traffic is no longer the problem for most stores. The Salesforce Shopping Index shows global digital traffic up 18% in Q2 2026 while order volume rose just 1%, with cart abandonment at 82%. Baymard Institute's running average across 50 studies puts documented cart abandonment at 70.22%.

Look at why people leave. In Baymard's 2026 survey of 1,083 US online shoppers:

• 40% abandoned because extra costs like shipping were too high
• 20% because delivery was too slow
• 13% because the return policy wasn't satisfactory
• 12% because they couldn't see the total order cost upfront

Several of those are questions in disguise. "How much is shipping to Leeds?" "Will it arrive before the 14th?" "Can I return it if it doesn't fit?" A shopper who gets a fast, correct answer has one less reason to close the tab. A shopper who has to hunt through a footer link often doesn't bother.

Be honest about the ceiling, though. The same Baymard survey found 42% of shoppers had abandoned a cart because they were "just browsing." No assistant fixes window shopping.

Which questions should an AI answer, and from what source?

Every answer needs a source of truth. If the assistant writes the answer from memory or from a product description someone pasted into a prompt six months ago, it will eventually be wrong in a way that costs money.

The shopper asks	The answer must come from	What happens if the AI guesses
"Does this come in navy, size M?"	The product's real variants in your catalog	It promises a variant you don't make
"Is it in stock?"	Live inventory, checked at the moment of asking	An oversold order, a refund, a bad review
"How much is shipping, and how long?"	Your published shipping policy	The hidden-cost surprise that drives 40% of abandonment
"Can I return it?"	Your published returns policy	A dispute when the customer quotes the chat back to you
"Which one is better for running?"	Product descriptions and specs	A wrong recommendation, then a return
"Where is my order?"	The order record, after an identity check	Order details shared with the wrong person

The pattern in that right-hand column is the reason "just add a chatbot" goes badly for stores. Shopify's Q2 2026 data points the same way from another direction: when AI search tools drew on structured Shopify Catalog data, the shoppers they referred converted at 2x the rate of those referred from scraped or third-party product feeds. Clean, current product data is what makes an AI answer worth trusting, whether that AI belongs to a search engine or to you.

We covered the general method in how to train an AI agent on your own business knowledge.

Why does reply speed matter as much as accuracy?

Because shoppers compare. In the Entagl Response Velocity Study (2026), we looked at 6,138 cases in the e-commerce and marketplace strata where one buyer messaged two or more sellers about a comparable product within four hours. Among the sequences we could link to a purchase, the first seller to reply captured 78.4% of sales. Across all 32,581 conversations in the study, replies inside 60 seconds converted at 35.1%, against 7.1% for replies that took one to 24 hours.

That study is observational, not a controlled trial, so read the numbers as a strong association rather than a promise. For a store whose inbox goes quiet from 6pm to 9am, any overnight question that is also going to a rival seller becomes one of those races, started with a 15-hour handicap.

Where do these questions actually arrive?

Not only in the chat bubble on your product page. Salesforce found that social media drove 29% more ecommerce visits in Q2 2026 and that purchases through social media rose 17% year over year. For many small brands the busiest sales desk is the Instagram DM inbox, with WhatsApp close behind.

Two things follow from that:

1. The assistant has to work where the question lands. A shopper who asks about sizing in an Instagram DM won't move to your website chat to get an answer.
2. The language has to follow the shopper. A store that ships internationally gets questions in whatever language the buyer thinks in.

Shopify's data also shows where shoppers arrive. In Q2 2026, 50% of AI-referred sessions to Shopify stores landed directly on a product page, which means the shopper skipped your homepage and category pages and arrived with a specific question. If you want the bigger picture on assistants that shop on a customer's behalf, see our piece on agentic commerce and AI shopping agents. This post is about the other side: your own assistant, answering your own customers.

What should a store's AI assistant never do?

Here is where most of the risk sits. Four rules worth writing into any setup:

• Never invent stock, prices, or policy terms. If the live data can't be reached, the assistant should say so and offer a person, not fill the gap. Our guide on how to stop AI agents hallucinating goes deeper on grounding.
• Never share order details without checking identity. An order number alone proves nothing. Ask for the email or phone number on the order too.
• Never take card details in the chat. Send the shopper to your store's own checkout, where payment is handled properly.
• Treat product text as data, not instructions. Product descriptions, reviews, and policy pages are written by people, and some text can carry hidden instructions aimed at AI systems. We explain the risk in prompt injection and how to secure your AI agent.

You might be thinking this sounds like a lot of caution for a sizing question. It is, and that's the point: the sizing question is easy. The returns question with an angry customer at 11pm is where an unguarded assistant does damage.

How does Entagl answer store questions?

Entagl's Receptionist is the agent that handles customer messages, and it connects to Shopify so its answers come from your store rather than from memory. Once connected, it can:

• Read your catalog: product names, variants, descriptions, prices, currency, and images. When the synced catalog might be out of date, it searches your store's live public catalog.
• Check live stock before saying "in stock," including availability by location (once you grant inventory access when connecting Shopify).
• Answer from your own policy pages for returns, refunds, shipping, and warranty, instead of paraphrasing what it thinks your policy is.
• Send a checkout link for the exact variant the shopper chose. It asks for the quantity and the shopper's name first, and the shopper pays on your Shopify checkout using whatever methods your store offers.
• Look up order status and tracking only after verification: the order number plus the email or phone number on that order. Orders older than 60 days are handed to a person.
• Ask which store when you run more than one, so a UK shopper isn't quoted US stock.

It works across WhatsApp, Instagram DMs, Facebook Messenger, Telegram, your website chat, and email forwarded from your existing mailbox. It replies in the shopper's own language (more than 100 are supported), it can read a photo a customer sends ("do you have this one?"), and anything it shouldn't handle goes to your team's shared inbox with the conversation attached. Shopify data is treated as untrusted data, so text inside a product page can't redirect the agent.

What does an AI assistant not fix?

Three limits worth stating plainly:

• It can explain a shipping cost, not lower it. If 40% of your abandonment is price shock, the fix is a pricing or shipping decision. Clear answers only stop the surprise.
• It is only as good as your catalog. Missing sizes, vague descriptions, and an outdated returns page produce confident answers built on bad data. Fix the data first.
• The headline numbers are correlations. Salesforce's sales-growth gap compares retailers with and without shopper agents; it doesn't prove the agent caused the whole difference. Measure your own before and after.

FAQ

Can an AI assistant answer product questions on Instagram and WhatsApp as well as my website?

Yes, if it connects to those channels directly. Entagl's Receptionist answers on WhatsApp, Instagram DMs, Facebook Messenger, Telegram, website chat, and forwarded email, using the same catalog and policy data on every channel.

Will an AI chatbot tell customers something is in stock when it isn't?

It can if it answers from memory or a stale copy of your catalog. A well-built assistant checks live inventory at the moment of the question and says so when it can't reach the data, rather than guessing.

Can an AI assistant take payment for a Shopify order in the chat?

It shouldn't. The safer pattern is for the assistant to create a checkout link for the exact product and variant, and the customer pays on your Shopify checkout page with the payment methods your store already offers.

Does an AI shopping assistant replace good product pages?

No. Shopify's Q2 2026 data shows half of AI-referred visits land straight on a product page, so the page still has to do its job. The assistant handles the question the page didn't answer, at the moment the shopper asks it.

How is my own store assistant different from ChatGPT recommending my products?

ChatGPT and other AI search tools recommend products before a shopper reaches you. Your own assistant answers after they arrive or message you, with access to live stock, your policy pages and the customer's order history. Shoppers use both, and Salesforce found they prefer brand-owned agents for returns and order questions.

Where to start this week

Export last month's DMs and chat logs and sort the questions into the six rows of the table above. Count how many went unanswered for more than an hour. That number tells you what an assistant is worth to your store before you spend anything.

Want to see the Receptionist answer real product, stock, and returns questions from your own Shopify catalog? Book a 30-minute demo and bring a few of your trickiest customer messages.

---

Sources: Salesforce, "2026 Holiday Predictions" (July 2026); Baymard Institute, cart abandonment statistics and 2026 abandonment-reasons survey; Shopify, "AI and organic search are doing different jobs" (August 2026); Entagl Response Velocity Study (2026). Figures as of October 2026.