# AI Assistant vs Scripted Chatbot: Which One Fits Your Shop

> A scripted chatbot follows buttons you drew in advance. An AI assistant reads your catalog and answers in plain language. Here is how they differ, and when each one makes sense for an online store.

Published: 2026-10-08 | Author: Vendigit
Canonical: https://vendigit.com/journal/ai-assistant-vs-scripted-chatbot

A customer opens the chat on your site and types: "Do you have a cordless drill under 3,000 that takes the same battery as my old one?" A scripted chatbot shows four buttons: Delivery, Payment, Catalog, Talk to a manager. An AI assistant goes to your catalog, finds the drills that match, and replies with two options and their prices. That gap is the whole debate of **AI assistant vs scripted chatbot**. Both are useful. They solve different problems, and picking the wrong one costs you either money or customers.

This article explains how each works, where each one breaks, and how to decide for your own shop.

## What a scripted chatbot is

A scripted chatbot (also called a flow bot, button bot or rule-based bot) follows a decision tree that someone built by hand. You draw the flow in a visual editor: greeting, three buttons, a branch for each button, a fixed answer at the end of each branch. Some builders let the bot react to keywords in free text, but the answers are still the ones you wrote.

Most live-chat and messenger platforms sell this. For example, as of October 2026, [SendPulse's chatbot pricing page](https://sendpulse.com/en/pricing/messengers) lists a free plan with up to 3 chatbots and 500 subscribers, and [Tidio's pricing page](https://www.tidio.com/pricing/) sells its "Flows" automation as a separate add-on.

What scripted bots do well:

- **Predictable answers.** The bot says exactly what you wrote. Legal and brand teams like that.
- **Guided actions.** "Choose a size, choose a colour, pick a delivery city" works well as buttons.
- **Broadcasts and subscriptions** in messengers, which is a different job from answering questions.

Where they struggle in a shop:

- **Customers don't press buttons, they type.** "Is the blue one back in stock?" does not match any branch.
- **Your catalog changes, the script does not.** A flow that names a price or a product goes out of date the day you change the price.
- **Every new question is new work.** Someone has to notice the gap, draw a branch, write the answer and test it.

## What an AI assistant is

An AI shopping assistant reads what the customer wrote, works out what they need, looks it up in the shop's own data and writes a reply in plain language. There is no tree to draw. The quality depends almost entirely on what the assistant is allowed to read and what rules it follows.

That is also its risk. A general language model given a pasted price list can produce a confident, wrong number. A good shop assistant is built to keep prices tied to your feed.

Here is how Vendigit, our product, handles it, as one example of the approach:

- It answers from your **product feed** (the same YML, XML, CSV or JSON export your platform already makes for marketplaces) and from your own delivery, payment and returns pages.
- The model points to a product by its ID, and Vendigit fills in the name, price and link on the product card straight from the feed. The model is instructed not to state a price or stock level it didn't find; if a lookup returned nothing, it says the shop doesn't have it.
- When it should not answer (a complaint, a refund dispute, wholesale prices, a question about an existing order) it hands the conversation to your manager in the same chat.

## Side-by-side comparison

| | Scripted chatbot | AI assistant (catalog-grounded) |
|---|---|---|
| How the customer talks to it | Mostly buttons, some keywords | Free text, any wording |
| Where answers come from | Text you wrote in the flow | Your product feed and site pages |
| Setup | Drawing and testing flows | Connecting a feed and your site |
| Upkeep when prices change | Edit every flow that mentions them | None, the feed re-syncs |
| "Do you have X under Y?" | Usually sends to catalog or manager | Searches the catalog and answers |
| Comparing two products | Only if you scripted that pair | Compares from product data |
| Risk | Outdated answers, dead ends | Wrong answers if not grounded in data |
| Best at | Guided steps, broadcasts | Product questions, store policies |

## The questions shops actually get

If you look through a month of your chat history, most messages fall into a few groups. We cover them in detail in [common customer questions for online stores](/journal/common-customer-questions-online-store), but the short version:

1. **Availability.** "Do you have it? When will it be back?"
2. **Price and options.** "How much is the 18V version? Is there something cheaper?"
3. **Fit and compatibility.** "Will this battery fit my drill?"
4. **Delivery and payment.** "How long to Lviv? Can I pay on delivery?"
5. **Problems.** "My order arrived damaged."

A scripted bot handles group 4 well, because delivery terms rarely change and fit nicely into a button. Groups 1 to 3 are where it fails: there are thousands of products and endless ways to ask. That's exactly where a catalog-grounded AI assistant earns its place. Group 5 should go to a person with either tool.

## When a scripted chatbot is the right choice

Be honest with yourself. A scripted bot is enough when:

- Your catalog is small (a few dozen products) and rarely changes.
- Most of your chats are the same three questions about delivery and payment.
- You mainly use messengers for broadcasts, promo codes and subscriptions.
- You want a fixed, pre-approved answer for every case and accept that anything else goes to a manager.

## When an AI assistant is the right choice

An AI assistant makes sense when:

- You have hundreds or thousands of products, and customers ask about specific ones.
- Prices and stock change weekly or daily.
- Questions arrive outside working hours and you lose them by morning.
- Your managers spend most of the day typing "yes, in stock, here is the link."

The condition is data. An assistant like Vendigit needs a product feed and readable policy pages on your site. If your platform (Horoshop, Prom, WooCommerce, OpenCart, Shopify, Magento or a custom build) can export a feed, you have what it needs. Our guide to the [YML product feed](/journal/what-is-yml-product-feed) explains what that file is.

## Can you use both?

Yes, and many shops will. Keep the scripted bot for what it is good at: newsletters in messengers, a guided "find your size" flow, promo campaigns. Put the AI assistant where open questions arrive: the chat on your site, and the Telegram chat where customers write to you directly. The two jobs barely overlap.

What you should avoid is the middle option: a scripted bot with a "free text" branch that passes everything to a general AI model with no access to your catalog. That's how a shop ends up with a bot that invents a discount.

## How to decide in 15 minutes

1. Export the last 100 chat messages.
2. Mark each one: could a button with a fixed answer have handled it?
3. If more than half are "yes," a scripted bot will do.
4. If most are about specific products, prices or stock, you need an assistant that reads your catalog.
5. Mark the ones that need a person (complaints, returns, orders). Whatever you pick, check how it [hands those over to a manager](/journal/when-ai-should-hand-over-to-manager).

For a wider look at the tools on the market and how to compare them, see our guide to the [best AI chatbots for online stores](/journal/best-ai-chatbots-for-online-stores).

## FAQ

### Is an AI assistant harder to set up than a scripted bot?

Usually it's less work. A scripted bot needs every flow drawn by hand. A catalog-grounded assistant needs a product feed and your site. Vendigit's setup is five steps: add the shop, verify the domain, add the widget or Telegram, add the feed, go live.

### Will an AI assistant make up prices?

A general chatbot with a pasted price list can. A grounded one is built to avoid it. In Vendigit, prices and links on product cards come straight from the feed, and the assistant is instructed not to state any price, stock level or delivery term that a lookup did not return.

### Can a scripted bot answer questions about specific products?

Only the ones you scripted. For a catalog of thousands of items, that means sending the customer to the catalog page or to a manager.

### Does an AI assistant replace my managers?

No. It takes the repetitive questions. Complaints, refunds, wholesale and order problems go to your team, with the full conversation attached.

Want to see how a catalog-grounded assistant answers your own customers' questions? [Start the 14-day free trial](https://app.vendigit.com/register). No card required.