AI chatbots for online stores: what they can and can't do in 2026

Open any app store for Shopify or WooCommerce and count the products with "AI" in the name. Most promise the same thing: switch it on and the bot handles most of your tickets. Store owners who tried a chatbot a few years ago remember something else, a widget that answered "I didn't understand that" to half the questions and annoyed everyone else.

Both pictures are out of date. AI chatbots in 2026 are genuinely good at a specific kind of store question, and still bad at others. This guide draws that line clearly, so you can decide what to hand to AI before the holiday season, what to keep for people, and how much it will really cost.

Why store owners are confused about AI chatbots

The confusion is understandable:

• Every vendor quotes a different number. "Resolves 50%", "answers up to 80%", "automates 60%". Each is measured differently, on different stores.
• Demos use perfect content. The bot in the demo has a tidy help center and a catalogue with clean product data. Yours may not.
• "Chatbot" means three different things. A decision tree you build by hand, an AI agent that writes its own answers, and a mix of both are all sold under the same word.
• Pricing hides in the details. Per conversation, per resolution, per ticket plus per AI interaction. The cheapest headline price is not always the cheapest bill in December.

The way through is to look at what actually happens in a store conversation.

What an AI chatbot does well in a store

AI chatbots shine when a question has one correct answer that is written down somewhere, or that sits in your store data.

Shopper question	How well AI handles it	What it needs from you
"How long does shipping to Germany take?"	Very well	Shipping times by region in your help content
"Does this run small?" / "What size should I get?"	Well	A size guide and fit notes on product pages
"Is this jacket waterproof?"	Well	Complete product descriptions and specifications
"Can I return a sale item?"	Very well	A clear returns policy with the exceptions written out
"Where is my order #10482?"	Well, if connected	Store integration that looks up orders and tracking
"What would you recommend for a 6-year-old?"	Fairly well	Product catalogue access and some buying-guide content
"Can I cancel or change my address?"	Partly	Store integration plus rules for what is allowed and until when
"My parcel arrived damaged"	Collect details only	A flow that gathers photos and order number, then a person
"This is the third time I'm writing, I want a refund"	No	Immediate handoff to a person

Notice the pattern. The first rows depend on content, the middle rows on integration, and the last rows on human judgement.

What AI chatbots still can't do

Be honest with yourself about the limits, because your customers will find them:

• Judgement and exceptions. Whether to refund outside policy, accept a late return or send a free replacement is a business decision. AI can gather the facts; a person should decide.
• Emotion. A frustrated customer wants to feel heard by someone who can act. A polite bot reply often makes it worse.
• Information you never wrote down. If the answer lives only in your head ("we can usually ship that same day if you order before noon"), the AI doesn't know it.
• Live situations. Carrier strikes, a warehouse delay, a broken discount code. Until you update the content, the AI keeps giving yesterday's answer.
• Promises it can't keep. A good AI agent won't invent a delivery date, but it also can't guarantee one. Policies about what it may promise must come from you.

Use AI for answers and simple, reversible actions. Keep money, exceptions and anger with people.

Step 1: Find out which questions you actually get

Before choosing a tool, look at the last two or three months of messages. Tag a few hundred by reason: shipping, order status, returns, sizing, product question, discount problem, complaint, other.

Then mark each reason as content (answerable from public information), data (needs the order or account) or judgement (needs a person). The first group is what an AI chatbot will handle from day one. The second needs integration. The third tells you how much human time you'll still need.

Step 2: Prepare your content and product data

Your AI chatbot is only as good as what it reads. Before launch:

1. Write or rewrite the policy pages for shipping, returns, exchanges and payment, with the numbers stated plainly (days, countries, costs).
2. Fill in product data. Materials, dimensions, compatibility, care instructions and fit notes. Missing data means "I'm not sure" answers on product pages, exactly where sales are made.
3. Add Q&A pairs for things only your team knows, such as gift wrapping, same-day dispatch cut-off or bulk discounts.
4. Remove anything outdated: last year's holiday dates, old promotions, discontinued products.

For a full method, see our guide on writing help content your AI agent can use.

Step 3: Connect it to your store

Without a store connection, the chatbot can only describe your order policy. With one, it can:

• look up an order by number or email and show status and tracking;
• see what is in the shopper's cart and answer in context;
• recommend in-stock products from your catalogue;
• in some tools, cancel or edit orders within rules you define.

Check how the integration works in practice: which platforms are supported natively, whether customers must verify their email before seeing order details, and what actions are possible without a developer.

Step 4: Set the boundaries and the handoff

Decide in advance what the chatbot must never do on its own, for example offer discounts, approve refunds or discuss legal claims. Then design the handoff:

• Always offer a person, in one step, in every conversation.
• Hand over automatically when the shopper is upset, repeats a question or mentions words like "refund", "damaged" or "charged twice".
• Pass the full context so the agent never asks the customer to start again.
• Outside your hours, collect the email and question, say when you'll reply, and create a ticket.

Step 5: Launch before the holiday rush

Black Friday is on November 27, and summer is the best time to start, while traffic is calm and every weak answer is easy to spot. Go live on website chat during business hours first, so your team can watch every conversation. Read transcripts daily, fix the content behind every weak answer, then extend to evenings, weekends and social channels. A chatbot launched in August has three months to improve before peak traffic; one launched in late October still has a month. One launched on Thanksgiving week learns on your best customers.

How much an AI chatbot costs at store volumes

Prices vary by vendor and change often, so compare models using your own numbers:

Pricing model	You pay for	Peak-season effect
Per AI conversation	Every conversation the AI takes part in	Rises with traffic, including conversations it hands over
Per resolution	Only conversations the AI resolves	Rises with success; definitions of "resolved" differ
Per ticket plus per AI interaction	Help desk tickets and AI usage separately	Two meters rise together in November and December
Seats plus AI usage with a minimum	Agent licences, AI conversations and a monthly floor	Predictable floor, but expensive for small volumes

A simple way to compare: take your expected monthly conversations, multiply by your vendor's AI price, and add what you pay for seats or tickets. Do it twice, for a normal month and for your busiest month. If you pay per resolution, use a conservative resolution rate for the first months, because rates usually improve as you fix content.

Red flags when buying

• The vendor can't show which page or product an answer came from.
• "Automation rate" counts every conversation the bot touched, including abandoned ones.
• There's no way to test with your own content before paying.
• Order actions require weeks of custom development for a standard Shopify or WooCommerce store.
• The person is hidden behind several menus.

Which one should you choose?

• Small or mid-size store, one to ten people on support: start with Tidio. Train Lyro on your help center, add an order status Flow and test it on the free plan before the holidays.
• Established Shopify brand already on Gorgias: try its AI Agent first, and model the extra interaction costs for peak months.
• High-volume brand with subscriptions and returns apps: compare Richpanel and Gorgias on your real ticket mix.
• Established brand that wants the AI agent run for it: Tidio Premium. Tidio's services team sets up and keeps improving Lyro and its knowledge, with a guaranteed 50% resolution rate and pay-per-resolution billing, so your CX team isn't maintaining the bot.
• Multi-brand retailer with a large team: evaluate Zendesk's AI agents with your current help desk.

Once your chatbot answers the basics, put it to work on structured tasks with these chatbot flows for online stores. If you're on Shopify, see our roundup of customer service apps for Shopify.