Table of Contents
- Key takeaways
- What is a custom AI chatbot?
- What makes an AI chatbot custom?
- Knowledge: the content the chatbot answers from
- Behavior: the rules and tone the chatbot follows
- Brand: how the chatbot looks and what it is called
- Channels: where the chatbot runs and who takes over
- Five routes to a custom AI chatbot for your company
- How the routes are ranked
- 1. Content-trained platform (SiteGPT): best for a chatbot trained on your own website and documents
- 2. Agent platform with actions (Chatbase): best for a choice of AI models and built-in actions
- 3. Visual flow builder (Botpress): best for scripted, multi-step conversations
- 4. Custom GPT inside ChatGPT: best for internal staff use in a ChatGPT workspace
- 5. Your own code on a model API: best for a chatbot inside your own product
- How to build a custom AI chatbot in seven steps
- Example behavior rules
- Worked example: a custom AI chatbot for a 40-person software company
- Which route should you pick?
- Common mistakes when building a custom AI chatbot
- Custom AI chatbot FAQ
- Basics
- What is a custom AI chatbot?
- What is the difference between a custom chatbot and a customizable chatbot?
- Do you need a custom chatbot, or can you use a template?
- Building
- Does a custom AI chatbot require coding?
- How do you create a custom AI chatbot without coding?
- Can you build a custom AI chatbot for free?
- How long does it take to build a custom AI chatbot?
- Cost and privacy
- How much does a custom AI chatbot cost?
- Is a custom AI chatbot on a company website private?
- Can a custom AI chatbot meet GDPR or HIPAA requirements?
- Sources
Created by
Do not index
Created time
Sep 2, 2023 10:33 PM
A custom AI chatbot is a chatbot set up to answer from your company's own content, follow your rules, carry your brand, and run on the channels your customers already use. Most companies get one by configuring a platform, and only a few need to write code.
TL;DR
- "Custom" has four layers: knowledge, behavior, brand, and channels. A chatbot is custom when you control all four.
- There are five routes: a content-trained platform, an agent platform with actions, a visual flow builder, a custom GPT inside ChatGPT, and your own code on a model API.
- A content-trained platform fits most company websites. SiteGPT Starter costs $59 a month billed monthly, or $39 a month billed yearly, with a 7-day trial that requires a credit card.
- The build takes seven steps: define the job, choose the route, load the knowledge, write the rules, brand the widget, connect channels and handoff, then test.
- Write code only when the chatbot must live inside your own product or run logic no platform offers.
Key takeaways
Layer | What you customize | How to check it |
Knowledge | The pages, files, and documents the chatbot answers from | Ask a question only your content can answer |
Behavior | Instructions, tone, scope, and what the chatbot says when it does not know | Ask an off-topic question and read the reply |
Brand | Name, colors, icon, welcome message, and the vendor's badge | Compare the widget with your website side by side |
Channels | Where the chatbot runs and how it hands a conversation to a person | Ask for a person and see who gets notified |
What is a custom AI chatbot?
A custom AI chatbot is an AI chatbot that a company has configured for its own content, rules, brand, and channels. A general assistant such as ChatGPT answers from what the model learned in training. A custom AI chatbot answers from what the company gave it.
Three terms get mixed up in searches. They describe different amounts of work.
Term | What it means | Who builds it |
Template chatbot | A ready-made script or prompt that you fill in | Anyone, in an afternoon |
Customizable chatbot | A platform chatbot whose knowledge, behavior, brand, and channels you set in a dashboard | A non-developer on a no-code platform |
Custom-built chatbot | A chatbot written in code on top of a model API | A developer or an agency |
A template is a starting point for a custom chatbot. If a template is all you need, the copyable prompts and flows in chatbot templates are the faster path.
"Custom" does not mean a custom-trained model. Almost every business chatbot uses a general model and retrieves the company's content at answer time. This method is called retrieval-augmented generation, and what is RAG explains it in full.
What makes an AI chatbot custom?
An AI chatbot is custom when the company controls four layers: knowledge, behavior, brand, and channels. A chatbot that changes only the logo and the welcome message is a generic chatbot with a new label.
Knowledge: the content the chatbot answers from
Knowledge is the set of pages, files, and documents the chatbot retrieves answers from. This layer decides whether the chatbot can answer a question about your refund policy or your product's setup steps.
Typical sources are website pages, a sitemap, uploaded files, and documents in cloud storage. The content has to be accurate before it goes in. The cleanup work is covered step by step in how to train an AI chatbot on your own data.
Behavior: the rules and tone the chatbot follows
Behavior is the set of written instructions that tell the chatbot what to answer, what to decline, and how to sound. Three settings matter most.
- Scope. The topics the chatbot answers, and the topics it turns away.
- Fallback. The exact message the chatbot gives when the content has no answer.
- Tone. Short or detailed, formal or friendly.
Instructions guide a model. They do not hard-block it. SiteGPT's own documentation says that "answer only from your content" is guidance, and that clear instructions reduce off-content answers without removing them fully.
Brand: how the chatbot looks and what it is called
Brand is everything the visitor sees: the chatbot's name, colors, icon, welcome message, and position on the page. It also covers the vendor's "Powered by" badge.
Removing the badge usually costs extra. It is an add-on at $39 a month billed yearly on SiteGPT, and $99 a month on Chatbase. Layout and wording choices for the widget are covered in chatbot design.
Channels: where the chatbot runs and who takes over
Channels are the places the chatbot answers: the website widget, a help center, Slack, WhatsApp, or a helpdesk inbox. This layer also includes the handoff to a person.
A chatbot with no handoff leaves the visitor stuck when the content has no answer. The handoff rules are covered in chatbot best practices.
Five routes to a custom AI chatbot for your company
The five routes to a custom AI chatbot are a content-trained platform, an agent platform with actions, a visual flow builder, a custom GPT inside ChatGPT, and your own code on a model API. Each route gives you a different mix of the four layers.
How the routes are ranked
The routes are ranked for one scenario: a company that wants a chatbot on its website that answers customer questions from its own content, set up without a developer. Four criteria decided the order.
- How much of the knowledge layer works without code.
- Whether the chatbot can be embedded on a public website.
- How much setup and upkeep the route needs.
- The entry price.
No route is best overall. Each one has a "best for" label, and a different scenario changes the order.
Route | Best for | Code needed | Entry point | Main limit |
1. Content-trained platform: SiteGPT | A chatbot trained on your own website and documents | None. One embed script | Starter, $59 a month billed monthly, or $39 a month billed yearly | No visual flow canvas and no free plan |
2. Agent platform with actions: Chatbase | A choice of AI models and built-in actions | None for answers | Hobby, $40 a month, with a 7-day trial | 700 message credits a month on Hobby |
3. Visual flow builder: Botpress | Scripted, multi-step conversations on a canvas | Low. Logic is built from nodes | Free, $0 with 100 conversations | Every flow has to be built and maintained |
4. Custom GPT inside ChatGPT | Internal staff use in a ChatGPT Business, Enterprise, or Edu workspace | None | A workspace plan with permission to create GPTs | Lives inside ChatGPT, and OpenAI plans to retire custom GPTs |
5. Your own code on a model API | A chatbot inside your own product | High | A developer and an API account | You build the widget, the refresh job, and the handoff |
1. Content-trained platform (SiteGPT): best for a chatbot trained on your own website and documents
A content-trained platform ranks first for this scenario because the knowledge layer needs no code. You enter a URL or upload files, and the platform indexes the content and answers from it.
- Knowledge: SiteGPT takes website URLs, sitemaps, uploaded files, and raw text. Its integrations page also lists Google Drive, Dropbox, OneDrive, Box, Notion, GitHub, and YouTube as data sources.
- Behavior: Custom instructions, a fallback message, 12 built-in personas, and custom personas are available on any plan.
- Brand: The Appearance page sets colors, icons, position, chatbot name, and welcome message.
- Channels: The widget installs with one script. Chat integrations with Slack, Crisp, Messenger, Google Chat, Freshdesk Messaging, and Zoho SalesIQ start on Growth. Zendesk starts on Scale, and WhatsApp is a paid add-on.
- Pricing: Starter costs $59 a month billed monthly, or $39 a month billed yearly ($468 a year). It includes 1 chatbot, 1,000 pages, and 4,000 GPT-4.1-mini messages or 400 GPT-4.1 messages a month.
- Limits: SiteGPT has no visual flow canvas, so it cannot run a scripted multi-step flow. There is no Intercom or HubSpot integration today. There is no free plan, and the 7-day trial requires a credit card. Starter has no chat integrations and refreshes content manually.
Yearly billing on Starter costs $468 against $708 for twelve monthly payments. That saves $240 a year, or 34%.
2. Agent platform with actions (Chatbase): best for a choice of AI models and built-in actions
An agent platform ranks second because it answers from your content and can also take actions inside a conversation. Chatbase lists models from OpenAI, Anthropic, Google, and others, plus a set number of AI Actions per agent on each plan.
- Knowledge: Hobby includes 10 MB of training content. Automatic re-sync of sources starts on Standard.
- Behavior and actions: Hobby includes 5 AI Actions per agent. Standard adds advanced integrations such as Stripe and Zendesk.
- Brand: Removing the "Powered By Chatbase" badge is an add-on at $99 a month. White-labeling is listed on Enterprise.
- Channels: A native Helpdesk for human takeover is listed from Standard.
- Pricing: Hobby costs $40 a month with 700 message credits and 2 members. Standard costs $150 a month with 4,000 message credits. Pro costs $500 a month. Each paid plan starts with a 7-day trial, and yearly billing takes 20% off.
- Limits: The pricing page shows no free plan card. Extra credits cost $40 per 1,000 through auto-recharge.
The full breakdown is in the Chatbase review.
3. Visual flow builder (Botpress): best for scripted, multi-step conversations
A visual flow builder ranks third for a content chatbot and first when the conversation must follow fixed steps. In Botpress Studio, each step is a node on a canvas, and a group of nodes forms a workflow.
- Knowledge: The Free plan includes 100 MB of vector storage for knowledge base content.
- Behavior: Standard nodes run their cards one by one. An Autonomous Node lets a language model decide what to say and which tools to use.
- Brand: White-label webchat starts on Plus.
- Channels: Webchat and WhatsApp are listed in the plan comparison. Voice is listed on Enterprise.
- Pricing: Free costs $0 with 100 conversations, 3 seats, and 3 AI agents. Plus costs $150 a month billed annually with 250 conversations a month. Extra conversations come in packs of 100 for $65.
- Limits: Someone has to design, test, and maintain every flow. The Free plan has no top-ups, so it stops at 100 conversations.
The full breakdown is in the Botpress review, and more canvas tools are compared in drag-and-drop chatbot builders.
4. Custom GPT inside ChatGPT: best for internal staff use in a ChatGPT workspace
A custom GPT ranks fourth because it lives inside ChatGPT and cannot sit on a company website as a widget. It fits a team that already works in a ChatGPT Business, Enterprise, or Edu workspace and wants a shared assistant with its own instructions and files.
- Knowledge: A GPT accepts up to 20 uploaded files, each up to 512 MB.
- Behavior: Instructions apply to every conversation. Actions can call external APIs that you define.
- Limits: OpenAI's help center says new GPTs cannot be created or published on personal accounts, including Free, Go, Plus, and Pro. It also says OpenAI plans to retire custom GPTs and recommends moving workflows to Plugins. The planned retirement date for affected Enterprise workspaces is December 11, 2026.
The other ways to give ChatGPT your files are compared in how to train ChatGPT on your own data.
5. Your own code on a model API: best for a chatbot inside your own product
Writing your own code ranks fifth for a website chatbot and first when the chatbot is a feature of your product. You control every layer, and you also build every layer.
- Knowledge: OpenAI's file search tool lets a model search files you upload to a vector store. OpenAI hosts the tool, so you do not write the retrieval code yourself.
- Behavior: You write the system instructions and any tool calls.
- Brand and channels: You build the chat interface, the content refresh job, the conversation log, and the handoff to a person.
- Limits: Developer time is the largest cost on this route, at launch and every month after. If you plan to hire an agency or a freelancer for this route, the chatbot development service guide compares their rates and timelines.
The first-year totals for this route and the platform routes are worked out in how much it costs to build a chatbot.
How to build a custom AI chatbot in seven steps
Building a custom AI chatbot takes seven steps: define the job, choose the route, load the knowledge, write the behavior rules, brand the widget, connect channels and handoff, and test before launch. The steps below assume a no-code platform.
- Define the job and the scope. Write one sentence that names the audience and the task, such as "answer pre-sales and setup questions from website visitors". List the topics that are in scope and the topics a person must handle.
- Choose the route. Use the decision matrix below. Most company websites fit route 1.
- Load the knowledge. Add the pages and files that answer the in-scope questions. Leave out drafts, internal notes, and outdated pages.
- Write the behavior rules. Put one rule on each line. Cover scope, the fallback message, tone, and when to offer a person.
- Brand the widget. Set the chatbot's name, colors, icon, and welcome message to match the website.
- Connect channels and handoff. Install the widget, connect any chat tools the team already uses, and decide who gets notified when a visitor asks for a person.
- Test, launch, and review. Run a fixed list of real customer questions before launch. After launch, read the unanswered questions every week and add content for the common ones.
Example behavior rules
These rules are a starting set for step 4. Change the company name and topics before using them.
- You are the assistant for Example Company. Answer only questions about Example Company's products, pricing, and setup.
- Answer only with information from the provided content. If the content does not answer the question, give the fallback message. Do not guess.
- Do not give legal, medical, or financial advice.
- If a visitor asks for a person, offer the handoff at once.
- Keep answers under five sentences and link the source page when there is one.
A fuller set of copyable prompts by use case is in chatbot templates.
Worked example: a custom AI chatbot for a 40-person software company
This worked example applies the seven steps to a hypothetical 40-person software company that sells an invoicing product. The company has 45 website pages, 120 help articles, and 6 PDF guides.
The job. The chatbot answers pre-sales and setup questions on the marketing site and the help center. Billing disputes and account changes go to a person.
The route. The content already exists as pages and files, and nobody on the team will maintain flows. That points to route 1.
The four layers on SiteGPT. Each layer maps to a setting named in SiteGPT's documentation.
Layer | What the company sets | Where it is set |
Knowledge | The website URL, the help center URL, and 6 uploaded PDFs | Website and file content |
Behavior | One instruction set with the five rules above, and a fallback message that includes the contact email | Chatbot Instructions |
Tone | The "Default" persona, which answers in 1 to 3 sentences when possible | Chatbot Persona |
Brand | Primary color as a hex code, a custom chat icon, the chatbot name, and a welcome message | Appearance |
Channels | The embed script on the marketing site and the help center, plus the Slack integration for the team | Installation and Integrations |
The plan arithmetic. The content size and the feature list decide the plan.
Item | Calculation | Result |
Website text | 45 pages at 3,000 cleaned characters each | 135,000 characters |
Help articles | 120 articles at 4,000 characters each | 480,000 characters |
PDF guides | 6 files at 25,000 characters each | 150,000 characters |
Total content | 135,000 + 480,000 + 150,000 | 765,000 characters |
Size in SiteGPT pages | 765,000 divided by 2,500 characters per page | 306 pages, which is 30.6% of the 1,000 pages on Starter |
Plan needed | Slack integration, 3 team members, and monthly auto refresh | Growth, because Starter has 1 member, no chat integrations, and manual refresh |
Growth price | $129 a month billed monthly, or $79 a month billed yearly | $1,548 or $948 a year. Yearly billing saves $600, or 39% |
Message allowance on Growth | As published on the pricing page | 10,000 GPT-4.1-mini messages or 1,000 GPT-4.1 messages a month |
The character counts are assumptions for the example. The content fits Starter by size. The channel and team needs move the company to Growth.
The test. Before launch, the team runs 30 real customer questions and checks each answer against its source page. Five of the 30 are questions the content does not cover, and the chatbot should give the fallback message for each.
Which route should you pick?
The route to pick depends on where the chatbot will live, where the answers come from, and who will maintain it. The matrix below maps each need to one route.
If you need... | Pick... | Why |
A website chatbot that answers from your pages, files, and cloud documents | A content-trained platform such as SiteGPT | The knowledge layer needs no code, and the widget installs with one script |
A chatbot that also takes actions, such as a Stripe or Zendesk lookup | An agent platform such as Chatbase Standard | Advanced integrations are listed from $150 a month |
A fixed sequence of questions, such as a quote or booking flow | A visual flow builder such as Botpress | Each step is a node you control |
To try a custom chatbot at no cost | Botpress Free | $0 with 100 conversations. SiteGPT and Chatbase offer 7-day trials |
A shared assistant for staff who already use ChatGPT at work | A custom GPT in a Business, Enterprise, or Edu workspace | No new tool to buy. Plan for the retirement OpenAI has announced |
A chatbot inside your own app, with your own interface | Your own code on a model API | Platforms give you a widget. Code gives you every layer |
Answers inside Intercom or HubSpot | A platform with that integration, such as Chatbase | SiteGPT has no Intercom or HubSpot integration today |
A chatbot with your brand only and no vendor badge | Any platform, with its branding add-on or white-label plan | Budget for it: $39 a month billed yearly on SiteGPT, $99 a month on Chatbase, included from Botpress Plus |
More platforms are compared in best no-code AI chatbot builders. If you picked route 1 or 2 and want to compare the tools that build that kind of chatbot, the AI chatbot creator guide ranks six by free plan and first-year cost. Widget types and install steps for a site are covered in AI chatbot for a website.
Common mistakes when building a custom AI chatbot
The most common mistakes come from customizing the visible layer and skipping the other three. Five show up often.
- Branding first. A chatbot with the right colors and the wrong answers loses trust faster than a plain one. Load and test the knowledge before styling the widget.
- No fallback message. Without one, the model fills gaps from general knowledge. The guardrails are covered in chatbot hallucinations.
- No route to a person. Visitors with a billing dispute need a human. Decide who gets the notification before launch.
- Writing code too early. A coded chatbot needs a widget, a refresh job, and a handoff before it matches a platform's first day.
- Setting it up once. Prices and policies change. A chatbot that is not refreshed repeats the old ones.
Custom AI chatbot FAQ
Basics
What is a custom AI chatbot?
A custom AI chatbot is an AI chatbot configured for one company. It answers from that company's content, follows its rules, carries its brand, and runs on its channels.
What is the difference between a custom chatbot and a customizable chatbot?
A customizable chatbot is a platform product with settings you change in a dashboard. A custom-built chatbot is written in code. Both end up as a custom chatbot when the knowledge, behavior, brand, and channels belong to the company.
Do you need a custom chatbot, or can you use a template?
Use a template when the chatbot answers a short, fixed list of questions. Build a custom chatbot when it must answer from your own pages and documents. A template can also supply the first draft of a custom chatbot's instructions.
Building
Does a custom AI chatbot require coding?
No. A platform route needs no code beyond pasting one embed script into the website. Coding is needed only when the chatbot must run inside your own product or use logic no platform offers.
How do you create a custom AI chatbot without coding?
Pick a no-code platform, add your website URL or files, write the instructions and fallback message, set the brand, and paste the embed script into your site. Then test with real customer questions before launch.
Can you build a custom AI chatbot for free?
Yes, within limits. Botpress has a Free plan at $0 with 100 conversations. SiteGPT has no free plan and offers a 7-day trial that requires a credit card. Chatbase's pricing page shows 7-day trials on its paid plans. More free options are compared in free chatbots for a website.
How long does it take to build a custom AI chatbot?
Indexing content on a platform is the short part. Most of the time goes into choosing the content, writing the rules, and testing the answers. A coded build takes longer because the interface and the handoff have to be built as well.
Cost and privacy
How much does a custom AI chatbot cost?
Entry prices read on the vendors' pricing pages are $0 for Botpress Free, $40 a month for Chatbase Hobby, and $59 a month billed monthly for SiteGPT Starter. SiteGPT Starter is $39 a month billed yearly. First-year totals for every route, including a coded build, are in how much it costs to build a chatbot.
Is a custom AI chatbot on a company website private?
The chatbot is as public as the page it sits on. Anyone who can open the chat can reach what the chatbot was given, so keep internal documents out of its content. On the vendor side, SiteGPT states on its home page that customer data is never used to train AI models.
Can a custom AI chatbot meet GDPR or HIPAA requirements?
That depends on the vendor's contracts and on the plan. On SiteGPT, a signed DPA and a HIPAA BAA are available on the Enterprise plan only. The comparison guides for GDPR-compliant chatbot platforms and HIPAA-compliant AI chatbots cover the details.
Sources
All sources were checked on October 12, 2026.
Written by
