AI Chatbot for Shopify: Inbox, an App or a Custom Build?

Baymard Institute puts the average online cart abandonment rate at 70.22%, based on 50 separate studies. Look at the reasons shoppers give and a pattern appears. Extra costs came as a surprise. Delivery looked too slow. The returns policy was unclear. They couldn't work out the total. Each of those is a question, and each one can be answered before the visitor gives up. That is the honest case for an AI chatbot for Shopify. You want a shop assistant who has read your delivery page and is awake at 11pm.

The hard part is choosing which one. Shopify ships a free option. The App Store lists dozens more, most of them promising to be "trained on your store". And a custom build costs real money. This guide walks through what a store chatbot has to do, what each option can and cannot read, what it costs, and the catch that matters most if you sell in Bulgarian.

What a store chatbot actually has to answer

Before comparing tools, sort the questions your customers ask. In our experience they fall into four groups, and each group needs different access to your data.

• Product questions. Does this come in a 42? Is it machine washable? Which charger fits my model? The answers live in your product descriptions, variants and metafields, plus whatever your staff know but never wrote down.
• Policy questions. How much is delivery to an Econt office? Can I pay cash on delivery? How do returns work? These live on pages most stores update once and then forget.
• Order questions. Where is my parcel? Can I change the address? These need the customer's own order, which means the bot has to know who it is talking to.
• Reassurance questions. Is this a real shop? Will it arrive before Friday? These are rarely about facts. They are about trust, and a fast, specific answer does more than a badge in the footer.

Pull a month of emails, Instagram messages and phone notes and tag each one with a group. The split tells you which tool you need. A store where most questions are about products needs a bot that reads the catalogue well. A store drowning in "where is my order" needs a bot that can look up orders safely. Those are different jobs, and a tool that is good at one is often mediocre at the other.

Three ways to add an AI chatbot to Shopify

Shopify Inbox

Shopify Inbox is Shopify's own chat app and it is free. Its listing describes AI "instant answers" connected to your catalogue, inventory and policies, plus controls for tone and a way to rate replies so they improve. For a store that sells in one of its supported languages, it is the obvious first thing to switch on. It costs nothing, it already knows your products, and it lives inside the admin you use every day.

A chatbot app from the Shopify App Store

The App Store category is crowded. Most apps follow the same pattern: they sync your products, pages and policies into their own index, put a widget on your storefront and charge a monthly subscription, sometimes with a fee per AI-resolved conversation on top. The better ones add live chat handover, abandoned-cart prompts and helpdesk features. The weaker ones are a general language model with your product feed pasted in.

A custom RAG chatbot

The third route is a bot built for your store: a retrieval system that indexes your catalogue, policies and internal documents, answers only from what it retrieved, and connects to Shopify's APIs for anything live. We explained the mechanics in how RAG chatbots work. It costs more up front and takes weeks rather than minutes. You pay for control over what it reads, how it answers, which language it writes in and what it does when it doesn't know.

A simple rule of thumb: start with the cheapest option that can read all the data your questions need. Move up only when you hit a wall you can name.

The Bulgarian-language catch

Most comparisons of Shopify chatbots are written for American stores, so they skip the question that matters most here: does it answer properly in Bulgarian?

As of October 2026, Bulgarian is not among the languages listed on the Shopify Inbox app page. Plenty of third-party apps claim support for a hundred languages or more, but "supports" usually means the underlying model can produce Bulgarian text. It doesn't mean anyone has checked how the bot handles Bulgarian product names, Cyrillic search, or a customer who writes "наложен платеж" in Latin letters because their phone keyboard is set to English.

Three things tend to go wrong:

• Retrieval breaks before generation does. The model writes decent Bulgarian, but the search step fails to match a Cyrillic question to a product title that is half Latin brand name and half Bulgarian description. The bot then says it can't find something you sell.
• Mixed-language catalogues confuse it. Many Bulgarian stores have English product names, Bulgarian descriptions and an English supplier spec sheet pasted at the bottom. A bot that has not been tuned for that mix picks the wrong passage.
• Tone drifts. Generic bots often switch between formal and informal address within one conversation, or reply with a stiff, translated phrasing that tells the customer immediately they are talking to a machine.

None of that rules out an off-the-shelf app. It means you should test one in Bulgarian, with your real catalogue, before you trust it with customers. Ask it ten questions your customers actually sent you last month, typed the way they typed them. If it fails three, it will fail hundreds once it is live.

What the bot needs to read, beyond product pages

Every vendor promises a bot "trained on your store". In practice that usually means the product feed and your public pages. That covers fewer questions than it sounds.

The answers Bulgarian shoppers ask for most often sit somewhere else. Delivery to an Econt or Speedy office versus a home address. Whether cash on delivery is available and what it adds to the total. How the 14-day right of withdrawal works for a sale item. Which size runs small. Whether the black version is coming back in stock. Half of this is on a policy page nobody has updated since launch. The other half lives in the heads of the two people who answer the phone.

Before you install anything, do three things:

• Update the delivery, payment and returns pages so they match what actually happens today. A bot that reads an out-of-date policy will quote it, confidently and in writing.
• Write down the ten answers your staff give most often that appear nowhere on the site. Sizing notes, compatibility, restock timing. Add them as an FAQ page or as product metafields so the bot can read them.
• Remove contradictions. If the footer says free delivery over €50 and the checkout says €60, the bot will pick one and the customer will find the other.

This is the unglamorous half of any chatbot project and it is where most of the value sits. It also makes your site better for humans, so it is never wasted work. The same holds for support bots in any business, not only shops.

Order status, refunds and anything that needs a login

"Where is my order?" is the question most store owners want automated first, and it is the one that needs the most care.

To answer it, the bot has to identify the customer and read their order. Done badly, that means a bot that will tell anyone who types an order number the name, address and contents of that parcel. Done properly, the customer either signs in to their account or confirms something only they would know, such as the email on the order, before the bot reveals anything.

Ask any supplier three plain questions:

• How does the bot verify who it is talking to before showing order details?
• Can it change anything, such as cancelling an order, editing an address or issuing a refund, or can it only read?
• Where are the conversation logs stored, and for how long?

Read-only order lookup with proper verification is a reasonable first step. Letting a bot cancel orders or issue refunds is a bigger decision. It moves from answering questions to taking actions, which is automation with write access to your store, and it needs rules, limits and a log a human checks. Most stores should run the read-only version for a few months before handing over the keys.

What each option costs

• Shopify Inbox is free. The real cost is the time spent writing and maintaining the answers it draws on, plus whatever you lose if it can't serve customers in their language.
• App Store chatbots usually charge a monthly subscription, with entry plans in the tens of dollars and higher tiers priced by conversation volume or by AI-resolved ticket. Read the pricing page for the per-resolution fee. On a busy store it can end up larger than the subscription.
• A custom RAG chatbot is a project. At Ascend, RAG chatbot work starts with a two-week proof of concept on your real catalogue and policies, so you see measured answer quality before committing. A production build usually follows in four to eight weeks, depending on how much data there is and what it has to connect to. Running costs are the language model API, billed per question, plus hosting for the index; for most small and mid-sized stores that lands in the low hundreds of euros a month. Our pricing page sets out how we scope it.

Then add the cost of a wrong answer, which no pricing page shows. A bot that tells a customer delivery is free when it isn't produces an angry email, a refund or a chargeback. One that recommends the wrong charger produces a return you pay to ship twice. Multiply that by your monthly conversation count before deciding the cheap option is cheap.

When a custom build is worth it, and when it isn't

Most Shopify stores should not start with a custom chatbot. Switch on Shopify Inbox or a well-reviewed app first if:

• Your customers write in a language the tool already handles well.
• Most questions are about delivery, returns and stock, and your policy pages are clear.
• You get fewer than a few hundred chats a month, so a wrong answer is an annoyance rather than a pattern.

A custom RAG build starts to pay off when one or more of these is true:

• You sell mainly in Bulgarian, or across several languages, and the apps you tested fumble it.
• Your products need expert advice: compatibility, dosage, technical specs, fit. The answers sit in PDFs, supplier sheets or staff knowledge rather than in the product description.
• You need the bot to connect to systems outside Shopify, such as a warehouse, an ERP, a courier tracking API or a CRM.
• You want to own the conversation data and the logic instead of renting them from an app that can change its pricing or shut down.

The Houzez AI Assistant we built for a Bulgarian property portal is a useful reference even though it isn't a store. It takes a plain-language brief such as "two bedrooms in Lozenets under €250,000", returns matching listings from live data and answers in Bulgarian or English, whichever the visitor started in. Product discovery on a store with a large, technical catalogue is the same problem with prices attached. If the store itself is outgrowing Shopify, that is a separate decision, covered in Shopify vs a custom website.

Before you switch it on

Whichever option you pick, run the same checks before customers see it.

• Test it with real questions. Take 30 messages customers actually sent you, including the badly spelt ones, and grade every answer as right, wrong or "didn't know". "Didn't know" is fine. Wrong is not.
• Test what it doesn't know. Ask it something your site does not answer. A good bot says so and offers a person. A bad one invents a policy in your brand's voice.
• Say it is a bot. From 2 August 2026, Article 50 of the EU AI Act requires AI systems that talk directly to people to be designed so those people know they are dealing with AI, unless that is already obvious. A short line in the widget header covers it. Pretending the bot is "Maria from customer care" no longer does.
• Make handover obvious. Every conversation should have a visible route to a human, by chat, email or phone, and the human should see the conversation so far.
• Read the logs weekly for the first two months. The questions the bot fails on are the best content brief you will ever get. Each one is a page, a metafield or a policy line that is missing.

Then measure the right thing. The number of conversations the bot "handled" is a vanity figure, because a conversation where the customer gave up looks handled too. Track how many chats end in a purchase or a resolved question, how many get handed to a human, and how many wrong answers you found in the weekly review.

The verdict

For most Shopify stores, start with Shopify Inbox if it serves your customers' language, or a well-reviewed app if it doesn't, and spend the first month fixing your policy pages rather than tuning prompts. That alone answers most of the questions behind abandoned carts.

Go custom when the off-the-shelf tools fail your own test: when they stumble over Bulgarian, can't read the documents your answers live in, or need to reach systems outside Shopify. At that point a RAG chatbot built on your data is the only version that gives correct answers, and in e-commerce a wrong answer costs you twice: once in the refund and again in the review.

Whatever you choose, count its wrong answers every week. That number tells you more than any demo.

Frequently asked questions

Does Shopify have a built-in AI chatbot?

Yes. Shopify Inbox is Shopify's free chat app, and its listing describes AI instant answers drawn from your catalogue, inventory and policies. Check that your customers' language is on its supported list before relying on it. As of October 2026, Bulgarian is not.

Can an AI chatbot answer "where is my order" on Shopify?

Yes, if it connects to your order data and verifies the customer first, usually through account sign-in or by matching the order email. Avoid any bot that shows order details to anyone who types an order number.

How much does an AI chatbot for Shopify cost?

Shopify Inbox is free. App Store chatbots typically charge a monthly subscription, sometimes with a fee per AI-resolved conversation. A custom RAG chatbot is a project: at Ascend it starts with a two-week proof of concept, and running costs for a small or mid-sized store usually land in the low hundreds of euros a month.

Will a chatbot reduce abandoned carts?

It can help with the abandonment that comes from unanswered questions, such as unexpected costs, delivery times and returns, which are among the top reasons in Baymard's research. It won't fix prices that are too high or a checkout that is broken.

Do I have to tell customers they are talking to a bot?

In the EU, in practice yes. Article 50 of the AI Act, which applies from 2 August 2026, requires AI systems that interact with people to be built so people know they are talking to AI, unless that is already obvious. A visible label in the chat window is the simplest way to make sure.

Can the chatbot recommend products?

Most can, with varying quality. The better ones ask a follow-up question before recommending, such as size, budget or use case, and only suggest products that are in stock. Test recommendations against your own catalogue, because generic demos always look good.

Should I build a custom chatbot or move off Shopify?

Those are separate decisions. A custom chatbot works perfectly well on a Shopify store through Shopify's APIs. Only consider replatforming if the store itself, not the chat, is what limits you.