How to Build a Customer Support Chatbot

Created by

Do not index

Created time

May 28, 2026 11:16 AM

Most chatbot builders make a website owner draw a decision tree before answering a single question. Map every branch, anticipate every phrasing, write a scripted reply for each one, then watch a visitor ask something nobody charted and get a dead end. That is the model behind flow builders like Landbot, Drift, and Intercom, and it breaks the moment a real customer asks a real question. Meanwhile, 74% of consumers now expect 24/7 customer service availability, which no business-hours team and no half-finished decision tree can deliver.

There is a faster way to build a customer support chatbot for a website. Point an AI at the help center a business already has, let it train, paste one snippet, and it answers customer questions in their own words, including the long-tail ones no script anticipated. This tutorial covers how to build one with SiteGPT in about 20 minutes, with no flow charts, no coding, and a real human escalation handoff for the moments the AI does not know the answer. Customer support is one slice of the broader shift toward customer service automation, and a website chatbot is the most direct place to start.

TL;DR: Create a free SiteGPT account, train it on a help center URL (scrape the whole site or connect specific docs), set the Problem Solver persona, add support-focused instructions, turn on Human Support so unresolved chats hand off to the team with full context, customize the widget, then paste one JavaScript snippet to go live. Total setup time: about 20 minutes, no decision trees and no coding required.

What You'll Build

By the end of this tutorial, a working customer support chatbot will be live on a website that:

Answers support questions in plain language by reasoning over a help center, product pages, and policy docs, including specific long-tail questions no one ever scripted

Hands the conversation to a human agent when it cannot answer confidently, passing the full chat history along so the customer never repeats themselves

Runs 24/7 from a single JavaScript snippet, with a Chat History view that shows the team exactly what customers are asking

Time to complete: about 20 minutes

Difficulty: Beginner

What you'll need: A SiteGPT account (free trial available), and a help center or website with support content the chatbot can train on

Why a Customer Support Chatbot Beats a Scripted Flow Bot

A scripted flow bot can only answer the questions someone remembered to map. A trained AI support chatbot answers the question a customer actually typed. That difference shows up in three concrete outcomes that matter to any support team.

It Handles Questions No One Scripted

A decision tree covers the paths its builder anticipated. Real customers do not stay on those paths. They ask "if I downgrade mid-cycle, do I keep my saved templates until renewal?" and a flow bot hits a branch nobody drew. A customer support chatbot trained on the full help center reasons over the underlying content and assembles an answer from the billing and account docs, even though no one wrote that exact question down. This is what separates a conversational AI chatbot from a scripted one, and it is the entire reason to skip the decision tree.

It Resolves Most Tickets Before They Reach a Human

A support chatbot answers instantly, at any hour, in any time zone. Tidio's chatbot statistics show customers increasingly reach for chat first, and a trained bot deflects the repetitive questions, password resets, shipping timelines, return windows, that otherwise fill a queue. As one SiteGPT ecommerce customer put it, "Our team spent a disproportionate amount of time answering the same questions over and over again, instead of focusing on higher-value tasks." A trained chatbot absorbs that load so agents handle only what genuinely needs a person.

It Escalates Cleanly Instead of Dead-Ending

The failure mode of a scripted bot is the dead end: "I didn't understand that, please rephrase." A well-built support chatbot does the opposite. When it cannot answer confidently, it hands off to a human agent and carries the full conversation along, so nothing falls through the cracks and the customer never starts over. That clean handoff, covered in Step 5, is what makes an AI support chatbot safe to put in front of customers.

Why SiteGPT Works Well for Customer Support Chatbots

SiteGPT is best suited for businesses that want a customer support chatbot trained on the help center they already have, not one assembled from hand-drawn decision trees. What makes a good support chatbot for a website comes down to two things: how accurately it answers, and how gracefully it fails. SiteGPT is built around both.

On accuracy, it ingests the content a business has already invested in. Crawl a full website, or connect a Zendesk help center, a Notion workspace, a Confluence space, Google Drive files, or uploaded PDFs. The chatbot reasons over that source material rather than matching to a fixed list, which is how it answers questions no one explicitly wrote.

On graceful failure, where flow builders like Landbot make a team wire up a separate live-chat tool to hand off a conversation, SiteGPT has Human Support built in. No third-party integration needed: a customer can request a human at any point, the chat escalates with full context attached, and the team picks it up from the Chat History view. Teams comparing the two options often start with these Landbot alternatives for exactly this reason. As one verified G2 user put it:

"SiteGPT makes it easy & intuitive to get your chatbot setup & working in no time at all - anyone can do it." Verified User, G2

Step-by-Step: How to Build a Customer Support Chatbot for Your Website

The setup below takes about 20 minutes start to finish, requires no coding, and never asks anyone to draw a single decision tree.

Step 1: Create Your Account and Chatbot (~2 min)

The first step creates the chatbot that will become the support agent. This is where it gets a name and a welcome message, the first thing a customer sees when the widget opens.

Go to SiteGPT and sign up for a free account

From the dashboard, click to create a new chatbot

Give it a recognizable name and set a welcome message that matches a support context

Name it something a customer will trust on sight, like "Acme Support Assistant," and write a welcome message that invites a real question rather than offering buttons, for example "Hi, ask me anything about your order, returns, or account."

Pro tip: Write the welcome message as an open invitation, not a menu. The entire advantage of an AI support chatbot is that customers can ask in their own words, so the greeting should signal that, not push them back into a list of canned options.

Step 2: Train It on Your Help Center (~5 min)

Training is what makes the chatbot a support agent instead of a blank LLM. This is the step that replaces the decision tree: instead of mapping branches, point SiteGPT at the support content that already exists.

From the chatbot, open the training sources and click to add links

Choose "Scrape Website" to pull an entire help center automatically, or "Multiple Links" to train on specific pages

Add the URL of the help center, FAQ, or support docs and let SiteGPT crawl and index them

If support content lives outside a public website, connect it directly. SiteGPT pulls from Notion, Confluence, Google Drive, and uploaded files, so internal handbooks and policy docs become part of the chatbot's knowledge without any rewriting.

What to train a support chatbot on:

Help center and FAQ articles (the core support content)

Shipping, returns, and refund policy pages for ecommerce, or account and billing docs for SaaS

Product and pricing pages so it can answer pre-sale questions too

Any internal troubleshooting docs for technical support scenarios

The more complete the source content, the more long-tail questions the chatbot resolves on its own. For a deeper look at what to feed it and how, these chatbot training tips cover how to structure source content so answers stay accurate.

Step 3: Set the Problem Solver Persona (~2 min)

The persona shapes how the chatbot talks. For customer support, the goal is a calm, resolution-focused tone, not a salesy one.

Go to Customizations, then Chatbot Persona in the left sidebar

Select the Problem Solver persona

Save the selection

The Problem Solver persona is the right fit for a support use case because it prioritizes diagnosing and resolving the customer's issue over upselling. It asks clarifying questions, stays patient, and keeps the conversation focused on getting the customer unstuck.

Step 4: Add Custom Support Instructions (~5 min)

The persona sets the tone; instructions set the rules. This is where the chatbot learns how to behave in edge cases: what to do when it is unsure, how to handle account-specific requests, and when to escalate.

Go to Customizations, then Chatbot Instructions, then Add Instruction

Write instructions tailored to a support context

Save and test a few questions to confirm the behavior

The single most useful thing to define here is what the chatbot does when it does not know. The instruction below is ready to paste in and adapt. It tells the bot to stay accurate, never invent policy, and route anything it cannot resolve to a human.

Example instructions for a customer support chatbot:

You are a customer support assistant for [Company]. Answer questions using only
the help center and product documentation you were trained on. Be concise,
friendly, and solution-focused.

If a customer asks about their specific account, order, or billing details that
you cannot see, do not guess. Explain what you can help with and offer to connect
them to a human agent.

If you are not confident in an answer, or the customer asks to speak to a person,
escalate to human support rather than guessing. Never invent policies, prices,
refund timelines, or features that are not in your training content.

This one instruction block is what makes an AI support chatbot safe to deploy. A scripted flow bot fails silently on anything off-script; a chatbot with these guardrails knows the boundary of its own knowledge and hands off instead of hallucinating.

Step 5: Turn On Human Support and Escalation (~5 min)

This is the step that makes a customer support chatbot trustworthy, and it is where SiteGPT separates from flow builders that bolt live chat on as an afterthought. Human Support is built in, so when the AI reaches the edge of what it knows, the conversation hands off to a real agent with the full chat history attached. The customer never repeats themselves, and nothing falls through the cracks.

Open the Leads section and go to the Human Support tab

Toggle on Enable Human Support to allow customers to request a human during any conversation

Under Escalation Button Settings, turn on Show escalation buttons after responses so a "Connect to an agent" option appears after each AI reply

Set the Request Human Support Prompt text (for example, "Connect to an agent") and a positive feedback prompt for resolved chats

Save the settings

When a customer escalates, the conversation moves into the Chat History view, where the team can read everything the AI already covered and pick up exactly where it left off. No copy-pasting context, no asking the customer to explain again.

For unresolved chats where the customer left contact details, escalation can also notify the team by email so no request sits waiting. This is configured alongside the lead settings.

Pro tip: Keep escalation buttons visible after every response rather than burying the option. Customers trust an AI chatbot more, not less, when an obvious path to a human is always one click away. The point is not to trap people in the bot; it is to resolve what can be resolved instantly and route the rest cleanly.

Step 6: Customize the Appearance (~3 min)

A support widget should look like part of the brand, not a bolted-on tool. This step is quick.

Open the chat interface settings and set the primary color to match the brand

Upload a chatbot avatar or icon

Choose the widget position, bottom-right is standard for support

Step 7: Deploy to Your Website (~2 min)

The final step puts the chatbot live. It is a single snippet, no developer required.

Go to Installation in the left sidebar

Copy the JavaScript embed code

Paste it into the website's HTML before the closing `</body>` tag

SiteGPT supports WordPress, Shopify, Squarespace, Wix, Webflow, and custom HTML directly

To confirm it is working, clear the site cache and open the page in an incognito window. The chat icon should appear in the corner. Open it, ask a question only the help center would know the answer to, and watch the chatbot answer a question no one ever scripted.

Tips to Get More Resolutions from Your Support Chatbot

Setup gets the chatbot live. These tips get it resolving more tickets without a human.

Tip 1: Mine Chat History for documentation gaps. The Chat History view is the most underused asset in any support chatbot. Read what customers actually ask and where the bot hesitated or escalated. Each escalation is a signal that the help center is missing a page. Write that page, retrain, and the next customer with the same question gets an instant answer.

Tip 2: Retrain whenever the source content changes. A chatbot trained on living documentation is only as current as its last training run. After a pricing change, a new policy, or a product update, recrawl the site or reconnect the source so answers stay accurate. Stale answers erode trust faster than no answer at all.

Tip 3: Tune the escalation threshold to the stakes. For high-stakes topics like billing disputes or account security, instruct the bot to escalate early rather than attempt an answer. For low-stakes questions like store hours or shipping windows, let it resolve fully. Matching escalation aggressiveness to the topic keeps customers safe without flooding the human queue.

Tip 4: Study who is doing this well. Looking at companies using chatbots for customer service shows a clear pattern: the best deployments lead with the AI for instant resolution and reserve humans for the genuinely complex cases. Borrow that structure rather than reinventing it.

Frequently Asked Questions

How do I build a customer support chatbot for my website without coding?

Build one by training an AI chatbot on existing support content instead of programming responses. With SiteGPT, sign up, point it at a help center URL or connect docs from Notion or a knowledge base, let it train, set a support persona, add instructions, turn on human escalation, then paste one JavaScript snippet into the website. The whole process takes about 20 minutes and requires no decision trees and no developer.

How long does it take to build a customer support chatbot?

About 20 minutes for a working chatbot using SiteGPT. Account creation and chatbot setup take roughly two minutes, training on a help center takes about five minutes plus crawl time, and configuring the persona, instructions, human escalation, appearance, and deployment takes the rest. Most of the elapsed time is the chatbot indexing source content in the background, which runs while other settings are configured.

How is an AI support chatbot different from a scripted flow chatbot?

A scripted flow chatbot, like those built in Landbot or Drift, follows a decision tree where a builder maps every branch and writes a reply for each. It can only answer questions someone anticipated. An AI support chatbot trained on documentation reasons over the source content, so it answers questions in the customer's own words, including ones no one scripted. It also fails gracefully by escalating to a human rather than dead-ending.

What should I train a customer support chatbot on?

Train it on the support content customers ask about most: help center and FAQ articles, shipping and return policies for ecommerce or billing and account docs for SaaS, product and pricing pages for pre-sale questions, and any internal troubleshooting guides. SiteGPT can crawl a full website or connect directly to Notion, Confluence, Google Drive, and uploaded files, so existing documentation becomes the knowledge base without rewriting it.

What happens when the chatbot cannot answer a question?

It escalates to a human instead of guessing. With Human Support enabled in SiteGPT, a customer can request a person at any point, and the chatbot routes the conversation to the support team with the full chat history attached. The agent picks it up from the Chat History view and continues without asking the customer to repeat anything. Clear instructions also tell the bot to escalate whenever it is unsure rather than inventing an answer.

Can a customer support chatbot hand off to a human agent?

Yes. SiteGPT has Human Support built in, so no separate live-chat tool or third-party integration is needed. Enable Human Support, then show escalation buttons after responses so a "Connect to an agent" option is always available. When a customer escalates, the conversation moves to the team with full context, and email notifications can alert agents so no request waits unanswered. This is the difference between a chatbot that helps and one that traps people.

Does a website support chatbot work for both ecommerce and SaaS?

Yes, the build process is the same; only the training content differs. An ecommerce support chatbot trains on shipping, returns, and order policies, while a SaaS support chatbot trains on account, billing, and product documentation. Because the chatbot reasons over whatever content it is given, the same SiteGPT setup handles a returns question or a technical support question equally well, as long as the relevant docs are part of its training.

How much does it cost to build a customer support chatbot?

SiteGPT starts at $39 per month on the Starter plan, which includes one chatbot, 4,000 messages per month, and training on up to 1,000 pages, enough for most small business support sites. A free trial is available to build and test a chatbot before committing. Growth ($79/month) adds more chatbots, messages, and team members for businesses with higher support volume. There is no separate charge for the built-in human escalation feature.

Will the chatbot answer questions that are not in my help center?

It will answer questions it can reasonably infer from the content it was trained on, even if no page states them word for word, because it reasons over the source material rather than matching a fixed list. For anything genuinely outside its training, well-written instructions tell it to say so and escalate to a human rather than guessing. That combination, broad reasoning plus a clear handoff, is what keeps answers both helpful and accurate.

How do I keep the chatbot's answers accurate over time?

Retrain it whenever the source content changes. A chatbot reflects its last training run, so after a price change, a new policy, or a product update, recrawl the site or reconnect the source in SiteGPT. Reviewing the Chat History view regularly also surfaces where the bot escalated or hesitated, pointing to documentation gaps worth filling. Keeping both the source content and the training current is what keeps a support chatbot trustworthy month after month.

Conclusion

A customer support chatbot built this way answers questions no decision tree could anticipate, resolves the repetitive load instantly around the clock, and hands the genuinely hard cases to a human with full context intact. No flow charts, no coding, about 20 minutes from sign-up to live. SiteGPT starts at $39 per month with a free trial, so a working chatbot can be tested before any commitment.

The most effective first move is to train the chatbot on the help center page that generates the most repetitive tickets, the returns policy, the billing FAQ, the setup guide, and watch how many of those questions it resolves before they ever reach the team. For teams running customer-facing chatbots across multiple sites or channels, SiteGPT's integrations connect the support agent to where customers already are.

Get Started with SiteGPT Free →

Last updated: May 2026. All SiteGPT features and pricing verified as of May 2026.

How to Build a Customer Support Chatbot for Your Website (Step-by-Step Tutorial)