How to Add an AI Chatbot to Your Website (No Code)

Piotr Grudzień
Published Updated 23 min read
Website chatbot workflow for Kestrel Kitchens: answer questions, collect a quote enquiry and hand the conversation to a designer

You can add an AI chatbot to your website by giving an Agent your business knowledge, configuring its replies and installing a chat widget with a plugin or a supplied script. Quickchat AI’s website chatbot keeps the knowledge, conversation settings and team Inbox in one place. Visitors ask questions on your site without creating an account.

This tutorial builds Kestrel Kitchens, a fictional kitchen design business. A visitor asks how long a renovation takes, checks the likely cost of a 3 by 4 metre kitchen, then requests a quote. The Agent answers from the company’s information, collects the project details and lets a designer continue the conversation.

The fictional Kestrel Kitchens website with the real Quickchat AI widget receiving a quote enquiry and explaining the indicative price range The finished website experience. Kestrel Kitchens and its business facts are fictional; the widget and its replies run through the real local application.

The core setup needs no programming. Follow knowledge and instructions, lead capture and human handoff, then install the widget. An optional API Action sends completed enquiries to another system.

Download the Kestrel Kitchens setup kit for the example website, knowledge, Main Prompt, settings and test messages. These are editable files to follow in the dashboard, not a one-click Agent import.

What will the website chatbot do?

The chatbot answers questions from your content, saves the details of a requested follow-up and transfers conversations to your team when a person is needed.

Those outcomes need different evidence. A correct price range comes from the Knowledge Base. A quote enquiry needs contact and project details in the Inbox. A handoff needs a human to take ownership and send a reply that reaches the visitor.

Flowchart showing a website question, a knowledge answer, a quote enquiry and a human handoff The visitor’s journey through the example. A quote enquiry records interest; the designer still confirms the price and availability.

Visitor’s questionThe Agent’s jobWhat you should check
How long does a renovation take?Explain the published design and installation durationsThe reply preserves the conditions on those durations
How much for a 3 by 4 metre kitchen?Give the indicative range and explain what a survey establishesNo invented fixed quote or price per square metre
Can I request a quote?Collect name, email, postcode and project typeThe fields appear in the conversation’s saved data
Do you offer financing?Acknowledge that the supplied knowledge does not sayNo invented financing offer
Can I speak to a designer?Use Human Handoff, subject to staffed hoursA designer can reply from Inbox and the visitor receives it

What do you need before adding a chatbot?

You need a Quickchat AI account, business information you can verify, and access to the website’s plugin or custom-code settings.

  1. Create a Quickchat AI account and complete the signup steps. Use a separate Agent for this exercise so its example facts cannot affect an existing customer service.
  2. Prepare your service pages, FAQs and policies, or use the fictional knowledge in the kit. Replace the fictional prices and service area before deploying for a real business.
  3. Check that your website plan allows the relevant installation method. WordPress needs plugin access; some hosted builders require a paid plan for site-wide custom code.
  4. Arrange for someone to monitor the test Inbox when you try human handoff. Keep a desktop browser and a phone available for the final website checks.

Free is for personal, non-commercial learning. The Free plan includes 50 AI credits per month. A business deployment requires a commercial plan, and testing uses the same reply allowance as other AI conversations. Website-builder subscriptions and external API costs are separate. The cost section distinguishes these costs.

Step 1: How do you train a chatbot on your website?

Add the pages that contain your business’s answers to the Knowledge Base, then verify the imported content before testing. This gives the Agent reference material for its replies.

Create the learning Agent

  1. On the first setup screen, select Set up manually. From an existing dashboard, use the + beside AI Agent (Add new AI Agent) to start a separate Agent, then select Set up manually.
  2. In Identity, enter Kestrel Kitchens under Name your AI Agent. Paste main-prompt.txt from the kit into Describe what you want your agent to do. This becomes the Agent’s Main Prompt, explained in Step 2. Select Next.
  3. Leave Knowledge Base empty for now and select Next. Continue through External apps without selecting a connection. In Deployment, select Chat widget, then Next.
  4. On Choose your plan, use Skip for now to continue the personal learning exercise, or select a suitable commercial plan for business use. Complete or dismiss any optional offer shown afterwards.
  5. In the dashboard, confirm the Agent’s name in Identity, then open Knowledge Base. Choose either your own pages or the supplied example article below.

The Kestrel Kitchens Identity page with the real Quickchat AI navigation and Main Prompt card Identity holds the Agent’s name and Main Prompt. Knowledge Base, Actions & MCPs, and Your Website are the next stops in this walkthrough.

Import your own pages

  1. In Knowledge Base, select Add Website. Choose a single page or the website import option, depending on how much you need.
  2. Enter the public URL. For a website import, review the discovered pages and select the service, pricing, process and FAQ pages that should inform answers.
  3. Inspect the imported text. Confirm that prices include the right currency and tax treatment, and that old promotions or unrelated pages have not become sources.
  4. Wait for processing to finish. Apply or publish any pending knowledge changes shown by the dashboard before testing the installed widget.

Use a few clear pages first. A site may show an attractive pricing card while putting its qualifications elsewhere; the Agent needs both. Importing the homepage alone can miss the installation conditions, geographic limits or refund policy that make an answer useful.

Use the Kestrel Kitchens source

To reproduce the example without hosting a website first, add its facts directly:

  1. In Knowledge Base, select Add Article. This creates an Untitled article and opens Edit Article.
  2. Enter Kestrel Kitchens: services, prices and process in Title. Under Content, choose Plain Text and paste knowledge.txt from the kit.
  3. Select Save and wait until the article is Ready. Confirm that its title appears in the Knowledge Base list.

The Kestrel Kitchens Knowledge Base with Add Article, Add Website and the example article marked Ready Knowledge Base is in the Agent navigation. The example needs one article; use Add Website for your own published pages.

Kestrel Kitchens Knowledge Base in the Quickchat AI app The saved article preserves the conditions on prices and timelines. Its fictional facts match the example website.

Copy the complete example knowledge
Kestrel Kitchens is a fictional kitchen design and renovation business used in this tutorial. It serves Bristol, Bath and addresses within 25 miles of central Bristol.

Services: kitchen design, supply and full installation. A full kitchen renovation typically costs £18,000 to £45,000 including VAT. This is a planning range, not a quote. The final price depends on the survey, layout, cabinetry, appliances and building work. Room dimensions alone are not enough to calculate a firm price.

Timelines: survey and design normally take 2 to 3 weeks. Installation usually takes 4 to 8 weeks after all materials are available. These are typical durations, not a promised start date or completion date.

Process: discuss the project, arrange a site survey, agree the design and receive a written quote, then schedule work once the quote is accepted and materials are available. The chatbot cannot book a survey or confirm availability.

The design team works Monday to Friday, 09:00 to 17:00, Europe/London. Quote enquiries need the visitor's full name, email, postcode and project type. A preferred timeline is helpful but optional. The team follows up by email. Do not promise a response deadline.

The source deliberately leaves financing unspecified. That gives you a useful test of how the Agent handles a genuine gap. For a real site, keep source pages current and use supported refresh settings where available; the Knowledge Base guide explains how to organize and test those sources.

Step 2: What should the chatbot say, and never promise?

Use Main Prompt to define how the Agent applies the knowledge: answer first, keep the reply useful, and leave decisions such as a firm quote to the team.

For Kestrel, room dimensions establish the scale of the project. They do not establish the cabinetry, appliances or building work. The Agent should explain the source’s £18,000 to £45,000 planning range, including VAT, without calculating a more precise price from the floor area.

Diagram separating an indicative kitchen renovation range from the survey and written quote required for a confirmed price These prices are fictional tutorial data. The lesson is to preserve the difference between source facts and a decision your team still needs to make.

  1. In Identity, select Edit on the Main Prompt card.
  2. Paste the complete prompt below. Let it save, then reopen the editor to confirm the stored text.
  3. Keep business facts in the Knowledge Base. Use the prompt for response style, collection behavior and the decisions requiring a person.

The Main Prompt editor for Kestrel Kitchens in Quickchat AI Main Prompt is the instruction field. The full text below and in the kit is the copyable version.

Copy the complete Main Prompt
You are the website assistant for Kestrel Kitchens. Help visitors understand our services, indicative prices and process using the Knowledge Base.

Answer the visitor's question before asking for contact details. Keep replies concise, usually two to four sentences. Use the visitor's language.

Treat prices as planning ranges. Never calculate a firm quote from room dimensions, invent a discount, promise a start date, confirm a survey booking or claim financing is available. If the Knowledge Base does not answer a question, say that a designer must confirm it.

When a visitor wants a quote or follow-up, collect their full name, email, postcode and project type in conversation. Ask one or two relevant questions at a time. Ask about their preferred timeline only after the required details, and accept a refusal without pressure. Do not ask for payment information.

Offer Human Handoff when the visitor asks for a designer or needs information you cannot confirm. Respect the configured working hours. Never claim that a designer is connected or a quote is booked. If the team is unavailable, offer an email follow-up.

Acknowledge the details the visitor has provided. Do not claim CRM delivery or a confirmed booking. The design team reviews enquiries in the Inbox.

Open AI Preview and ask How long does a full kitchen renovation take?. In the local control test, the Agent answered with 2 to 3 weeks for survey and design, then 4 to 8 weeks for installation after materials are available. Both the duration and the qualification matter.

Start a new conversation and ask Do you offer financing?. The correct response acknowledges that the source does not establish an offer and directs the question to a designer. Adding more confident wording to the prompt would not supply the missing business fact.

Step 3: How do you collect quote requests in the chat?

Smart Data Gathering asks for selected information during a conversation and saves what the visitor provides. For a quote enquiry, collect only what the designer needs to follow up.

Actions & MCPs in Quickchat AI with Human Handoff and the Kestrel quote-enquiry collection Action Enable and edit the built-in Actions here. The example’s Smart Data Gathering card is named Quote enquiries.

  1. Open Actions & MCPs, enable Smart Data Gathering and select Edit Action. Set Name to Quote enquiries and Description to Capture contact and project details for a kitchen design enquiry. The saved name becomes the card title.
  2. Under Data to collect, add the rows below. Keep these English Data point names when reproducing the example; the contact names have special mappings to the Inbox’s visitor fields.
  3. Set Reason for collecting information to So a Kestrel Kitchens designer can assess your project and follow up by email.
  4. Set When to ask for data to After exchanging a few messages, and How to ask for data to Subtle.
  5. Save the Action. In AI Preview, ask a useful service question before requesting a quote, and inspect the saved details in Inbox.
Data pointQuestion to askFormat or example values
Full nameWhat is your full name?Full name
EmailWhat email can our designer use?Email
PostcodeWhat is the property postcode?Text
Project typeWhat kitchen work do you need?Text
TimelineWhen would you like to start?Text

Smart Data Gathering settings for Kestrel Kitchens, showing the fields and collection behavior The questions describe the information to collect. The timing and style settings guide when to ask; test the actual conversation as well.

Name, email, postcode and project type are the required details for this example’s completed enquiry. Timeline is optional, as the Main Prompt explains. This business requirement is separate from the conversational collection settings; Step 8 enforces the required fields before a CRM request is sent.

The Agent should answer a general cost question without making an email address a condition of receiving the answer. A visitor can also decline collection. Use I am only researching. I do not want to leave contact details. to check that the conversation remains useful without repeated requests.

Where do captured details go?

Open the conversation in Inbox. Standard contact details appear in its visitor fields; the other collected values remain attached to the conversation and can be exported through Export Gathered Data.

Under Where to send data, Email new leads notifies every admin of the Agent when an email address or phone number is first captured. A name alone does not send the email. Preview and Simulation do not send these notifications, so verify your own admin inbox with a controlled test on the installed widget before relying on email delivery. The local walkthrough keeps external email delivery disabled.

The Smart Data Gathering documentation describes collection, notifications and exports. For your own site, explain why you ask for the information and link your privacy notice in the widget. The chatbot data-protection guide helps you review the wider data flow.

Step 4: How do you let a person take over the chat?

Human Handoff transfers a conversation to your team; a person then takes over in Inbox and replies to the same visitor.

Prepare the handoff before publishing the widget. Someone who asks for a designer should not have to finish the quote form first, and the Agent should not imply that someone is online outside staffed hours.

  1. In Actions & MCPs, enable Human Handoff and open Edit Action.
  2. Keep the standard Media Message, User Frustration, Customer Support Suggestion and Lack Of Knowledge rules on; leave Irrelevant Advice off. The Main Prompt already explains when to offer a designer. The tested example uses these standard rules without adding custom rules.
  3. Configure the messages and availability below, then save. Keep Email notifications off for the learning exercise and check the handoff directly in Inbox.
SettingKestrel Kitchens value
QuestionWould you like me to connect you with a kitchen designer?
ConfirmationI have passed your chat to our design team. A designer will reply here when available.
Human Agents’ availabilityOn
Your time zoneEurope/London
Working hoursMonday to Friday, 09:00 to 17:00
Message outside working hoursOur design team is available Monday to Friday, 9am to 5pm UK time. Leave your name and email so a designer can follow up.

Human Handoff availability and messages configured for the Kestrel Kitchens design team Use hours when a person actually monitors Inbox. The fictional schedule is not a promise of an immediate reply.

During a staffed interval, ask Can I speak to a designer? from the website. Complete any handoff confirmation the Agent asks for. In Inbox, open the conversation and select Assign to me when it is unassigned, or Take over when that control is shown. Send a human reply and confirm that the visitor receives it. Use Reassign to AI when the AI should resume.

Quickchat AI Inbox with Sam Taylor's conversation selected, Maya Patel's reply and visitor details Inbox brings the conversation, ownership control and visitor details together. Select the visitor’s conversation before replying.

The Kestrel Kitchens conversation in Inbox during a human takeover Check ownership and the delivered human reply, not just the Agent’s offer to transfer.

The Kestrel Kitchens widget showing Maya Patel's human reply after taking ownership in Inbox The visitor’s side of the same handoff. Maya Patel is the fictional designer used for the local test.

Repeat in a new conversation outside the configured hours. The Agent should explain availability and offer the stated follow-up route. If you temporarily change the schedule to run this check, restore the real schedule afterwards. The Human Handoff documentation explains the settings and Inbox flow.

The chatbot human handoff walkthrough includes separate checks for conversation ownership, handback and outside-hours failures.

Step 5: How do you make the widget match your website?

Set the visual appearance and fixed text under Channels → Your Website. Keep the welcome message specific enough to explain what the visitor can ask.

Your Website selected in Quickchat AI with the Install, Appearance and Text Content tabs The three tabs separate installation, appearance and fixed text. Appearance also holds the avatar and launcher label.

Appearance

  1. Open Appearance. For Kestrel, choose a Custom theme with white background, dark green header and accent (#315448), and white header text. Keep the default text size and rounded controls.
  2. Under Branding → AI Agent Avatar, click the avatar’s upload icon, choose website/kestrel-avatar.png from the kit and select Upload in the crop dialog.
  3. Set Widget window max width to 400px and Widget window max height to 520px. Choose Small for the launcher size, position it on the right and use Ask about your kitchen as its label. The label is available with the small launcher. Preview the position against any cookie banner or other floating controls on your site.
  4. Leave Auto-open Widget off. Opening the widget on every page load interrupts browsing, particularly on a phone.

Website widget Appearance settings in the Quickchat AI app Check the preview while choosing colors. The live website test still matters because the launcher shares space with your site’s controls.

Text Content

Open Text Content and use these values:

SettingValue
HeaderKestrel Kitchens
Welcome messageHello, I can help with kitchen costs, timelines and quote enquiries. What are you planning?
Conversation startersHow long does a renovation take? / What does a kitchen cost? / Can I request a quote?
Disclaimer textAI assistant. Prices are indicative. A designer confirms your quote after a survey.

Kestrel Kitchens welcome message and conversation starters in Text Content Conversation starters should open useful tasks. The disclaimer tells visitors which decisions still need a designer.

The Agent can reply in the visitor’s language. Translating the fixed welcome message and widget labels is a separate setting under Other languages, available on paid plans. For a multilingual site, test both the reply and the fixed interface text; the multilingual chatbot guide explains that distinction.

Appearance changes can take a few minutes to reach cached visitor sessions. Reload before concluding that a saved change failed. Hide Quickchat AI Branding requires Professional or above; the white-label walkthrough covers the other surfaces to inspect.

Step 6: How do you install the chatbot on your website?

Install the Widget in the site’s shared layout or use the supported platform integration. The floating Widget needs to run in the page that the visitor is browsing.

  1. Open Channels → Your Website → Install.
  2. Enable Widget & Embed access for the public widget.
  3. Choose the installation method for your website and copy the snippet or Scenario ID shown for this Agent.
  4. Publish the change, open the public site and ask a question whose answer you know is in the Knowledge Base.

Quickchat AI website installation menu with manual code, WordPress, Shopify and other site-builder options Copy from this Agent’s Install screen. Its Scenario ID is a public widget identifier, not a private API key.

Any website: paste the Widget snippet before the closing body tag

Choose Install manually with code and copy the Widget snippet. The current loader has this shape:

<script>
  (function(e,a,d,i,c,t=a.createElement(d)){e[c]=e[c]||function(){(e[c].q=e[c].q||[]).push(arguments)},t.src=i,t.async=1,a.body.insertAdjacentElement("beforeend",t)})(window,document,"script","https://bubble.quickchat.ai/chat.js","_quickchat");
  _quickchat("init", "YOUR_SCENARIO_ID");
</script>

Replace YOUR_SCENARIO_ID with the value from Install. Paste the complete snippet once, immediately before </body> in the shared layout, or use the builder’s site-wide footer field. This loader inserts a script into the page body, so placing it in the head before the body exists can prevent it from loading.

Diagram showing the Widget snippet at the end of the top-level page body, outside an iframe The location matters. An iframe-based HTML block is a separate page and is unsuitable for this floating Widget.

The separate Embed option places a chat inside a page container. Use its generated snippet if you want an inline chat section. An HTML block that inserts code directly into the page can support that pattern; a builder block that isolates code in an iframe is a different case. See the Website channel documentation.

WordPress: install the official plugin

  1. In WordPress, open Plugins → Add New, search for Quickchat AI Agent, then choose Install Now and Activate.
  2. Open Settings → Quickchat AI Agent.
  3. Paste the Scenario ID from the Quickchat AI Install screen and select Save Changes.
  4. Open a public page and check the widget. Use either the plugin or a manual snippet so you do not install the loader twice.

You need a WordPress installation and plan that allow plugins. The WordPress chatbot guide shows the plugin interface and WordPress-specific checks.

  1. Open the site’s Code Injection panel. Use the dashboard’s settings search if the navigation differs on your account.
  2. Paste the complete Widget snippet into Footer and save.
  3. Open the published page in a separate browser tab and test an answer.

Squarespace provides Code Injection on Core, Plus, Advanced and some legacy plans. Footer injection puts the code before </body> across the site. Use that surface for the floating widget; a page-level iframe block is not equivalent. This custom code does not extend to Squarespace checkout pages.

Wix: use Custom Code at Body - end

  1. In the site dashboard, open Settings → Custom Code, then + Add Custom Code.
  2. Paste the Widget snippet and name it Quickchat AI widget.
  3. Apply it to All pages and choose Body - end, then Apply.
  4. Test the published site, including navigation to another page.

Wix requires a published site with a connected domain for custom code. Use the site-wide Custom Code setting for this floating widget.

  1. Open Site settings → Custom code.
  2. Paste the Widget snippet into Footer code and choose Save changes.
  3. Publish to the intended domain, then test that published URL.

Webflow’s custom-code settings require an eligible paid Workspace or an active Site plan. Saving the setting alone does not publish it.

Shopify: choose the existing-Agent snippet or the Shopify app

For the Kestrel Agent you already built, add its Widget snippet immediately before </body> in Online Store → Themes → Edit code → layout/theme.liquid, then save and inspect the storefront. Keep a theme backup before editing, and recheck installation after changing themes. This route displays the Agent you selected; it does not configure product or order Actions.

For a store-focused Agent, the Quickchat AI Shopify app setup connects the store through its own onboarding. After that connection, use Enable Widget in Shopify Theme Editor, or open Online Store → Themes → Edit theme → App embeds. Enable Quickchat AI Widget and save. Shopify’s app embed documentation explains the theme controls; re-enable the app embed when switching the published theme.

The builder paths above were checked against official documentation. The worked screenshots show the Kestrel example website and the real local widget; they are not screenshots of authenticated accounts on every builder.

Check the snippet before troubleshooting the Agent

These optional checks work on the values you supply:

  • Check a Widget snippet. Check pasted code against the supplied Scenario ID. This does not fetch your website.
  • Compare deployment methods. Check your declared channels, website access and required features against Quickchat AI setup options. This does not inspect your website or account.

In a compatible browser, these checks are also available as site tools. The links open the standard forms.

Step 7: How do you test the chatbot before visitors do?

Test the answer, stored data and delivered reply separately. AI Preview and Simulation Testing help evaluate behavior; the published website proves that the visitor can use the installed widget.

Run independent questions in Simulation Testing

Open Testing → Create Dataset, name it Kestrel Kitchens: website launch checks and choose Paste messages. Add the duration, price and financing questions below, separated by a line containing --- between questions. A line break alone keeps the text in one test message. Set Evaluation criteria to check the facts and concise replies, then select Run → Start Run. The kit includes the exact messages and criteria.

Each dataset row starts a separate conversation. Use Simulation for those independent questions; run the multi-turn quote and human-takeover checks in one AI Preview or website conversation. Inspect the Action log and receiver separately because a transcript evaluator cannot verify external delivery.

The Kestrel website launch checks dataset with three completed Simulation cases and the expanded kitchen-price response Open a completed run and inspect its replies. This price answer preserves the planning range; its reply score does not establish delivery to another system.

The saved Simulation conversation showing the full kitchen-price question and the reply giving an indicative range rather than a fixed quote The same cost test opened as its Inbox conversation. Inspect the full answer, including why dimensions alone cannot establish a firm quote, instead of relying only on the score.

Test message or sequenceExpected outcomeInspect
How long does a full kitchen renovation take?2 to 3 weeks for survey/design; 4 to 8 weeks for installation after materials are availableReply and source
How much would a 3 by 4 metre kitchen cost?£18,000 to £45,000 including VAT as an indicative range; survey needed for a firm quoteNo invented fixed price
Do you offer financing?The Agent cannot confirm it from the source and offers the designer routeReply, then handoff if requested
Can I get a quote? My name is Sam Taylor.Ask for missing information; do not send the completed-enquiry Action yetSaved name; no CRM receipt
My email is sam@example.com. The postcode is BS6 7AB and it is a full kitchen renovation.Save the supplied email, postcode and project typeInbox fields; optional Action log and receiver
I have not decided when to start.Accept an unknown timelineNo invented date or required timeline
I am only researching. I do not want to leave contact details.Continue answering without repeated collection requestsThe next useful reply
Can I speak to a designer?Follow the configured in-hours or outside-hours pathConversation ownership and human delivery
¿Cuánto suele durar una reforma de cocina?Answer in Spanish while preserving the same time ranges and conditionsFacts as well as language

Use a new conversation between independent cases. Reloading the page can resume an existing conversation, so use the widget’s new-chat control when you need a fresh test.

Check the website on a phone

On screens below 768 CSS pixels, the open widget uses a full-screen layout. Check that you can open it, type with the keyboard visible, scroll the conversation and close it to return to the site.

Finish with a real person replying from Inbox. A good AI answer does not prove that the human support route works. Repeat a small set of these checks after changing sources, prompts, Actions or the site theme.

Step 8: How do you send a complete enquiry to your CRM?

An optional API Action can send the saved fields to an HTTP receiver after the conversation has all required information. The core website chatbot is already usable without this extension.

The important distinction is between a contact was captured and this enquiry is complete. Quickchat AI sets lead_captured when it newly captures a standard contact field, including a name. The flag alone does not establish that an email, postcode or project type is present.

Flowchart showing captured contact data passing required-field conditions before a one-time lead-delivery Action The platform checks every condition. The model collects the information; the conditions determine when the request becomes eligible.

Prepare the receiver

Use a CRM or automation endpoint you control that accepts the JSON body below. The kit includes a small local demonstration receiver for readers who want to inspect the request before connecting a CRM. Its setup instructions create a private token and explain how to expose a temporary HTTPS endpoint. Only send fictional test data to the demonstration receiver.

For a managed integration instead, follow the Google Sheets reporting tutorial or the custom CRM walkthrough. Do not turn on two workflows that both create the same lead.

Configure one automatic API Action

  1. Open Actions & MCPs → Custom Actions → Add Action → API Action. Choose Create manually if the description-based builder opens.
  2. Set API Action Name to Send quote enquiry.
  3. Leave What to ask the user first empty. This Action reads the fields already saved by Smart Data Gathering; it does not ask the model to fill a second set of parameters.
  4. Under API Endpoint, select POST and enter your receiver’s HTTPS /enquiries URL.
  5. Add the headers, body and conditions below, then save and activate the Action.
Headers: KeyValue
Content-Typeapplication/json
AuthorizationBearer YOUR_RECEIVER_TOKEN

Enter the token only in the Action’s header setting. Keep it out of the prompt, website snippet and visitor conversation. The demonstration receiver’s README explains how to generate it; use your CRM’s documented authentication for a real integration.

In the Body tab, select JSON and enter:

{
  "conversation_id": "{{conversation_id}}",
  "name": "{{metadata_visitor_name}}",
  "email": "{{metadata_visitor_email}}",
  "postcode": "{{metadata_postcode}}",
  "project_type": "{{metadata_project_type}}"
}

Use Add AI Data to insert the corresponding built-in and metadata values. conversation_id identifies the conversation. The other values come from the saved fields; the English contact names map to visitor_name and visitor_email, while Project type becomes project_type. The variables documentation explains these mappings.

API Endpoint settings showing POST, an example receiver URL, the Body tab and saved-field variables inserted with Add AI Data The real editor with placeholder connection details. Enter your own endpoint above, credentials in Headers and the JSON in Body; the colored chips identify dynamic values. This configuration-only example was kept inactive.

Under When to run, choose When metadata matches. In Run when, add all five rows; the metadata conditions are combined with AND:

Metadata keyCondition
lead_capturedis true
visitor_nameis true
visitor_emailis true
postcodeis true
project_typeis true

For these text fields, is true rejects empty values as well as false-like text. exists only checks that a key is present; a saved blank string can pass it. The tests below show why the stronger condition matters. This checks that usable text was supplied, not that a visitor’s identity or property details are independently verified.

Use this API Action Description:

Send a quote enquiry to the design team when the visitor has provided their full name, email, postcode and project type. This records an enquiry only; it does not book a survey or confirm a price.

The Send quote enquiry Action editor showing the lead_captured and visitor_email conditions set to is true The first two of five saved conditions. Add every row from the table; When metadata matches checks them after an AI turn without the Agent choosing the Action.

The Action is attempted once per conversation when all conditions match. It does not run again when an optional timeline arrives, and a failed sent request is not automatically retried in that conversation. Keep the timeline in Inbox for the designer, or build an explicit create-and-update integration if your CRM needs later changes.

The request runs after the AI turn, and its response is not inserted into the chat. That is why the Main Prompt acknowledges the visitor’s details without claiming that a CRM accepted them. Verify delivery in the Action log and receiver. The automatic Action documentation describes this execution behavior.

Test partial data, then the completed enquiry

The first version used exists for the required text fields. Sam supplied a name, then an email in the next message. Smart Data Gathering had already created blank postcode and project-type values, so those keys existed and the Action sent an incomplete request. That also consumed its one attempt for the conversation.

The correction was to change the four text-field conditions to is true, leaving the rest of the setup unchanged. Repeat the test in a fresh conversation, giving only an email on the second turn. The Action should wait for the remaining project details.

Test stateContact capturedRequired fields readyRequests received
Sam gives a nameYesNo0
Sam supplies only an emailYesNo0
Sam supplies postcode and project typeYesYes1
Sam adds “next spring”YesYes1

This is one observed conversation, not a production success rate. Reproduce it with a new conversation, inspect the saved fields after each turn, and check the request at your own endpoint.

The Send quote enquiry Action call showing HTTP response status 201 in Quickchat AI The actual Action call returned HTTP 201. Open its details to inspect the request; the saved record below shows what the receiver stored.

The local demonstration receiver's exported database record with Sam's name, email, postcode and project type One actual saved record from the authenticated demo receiver. This is an enquiry, not a confirmed quote or booking.

The demonstration used a local receiver and captured actual HTTP delivery. A hosted Quickchat AI Agent needs a publicly reachable HTTPS endpoint; a localhost URL on your computer is not reachable from the service.

If nothing arrives, inspect View logs on the Action and the conversation’s saved metadata. Missing values, a disabled Action and an HTTP failure need different fixes. Correct the configuration and use a fresh conversation to test the automatic trigger again; asking the same chat to retry does not reset a once-per-conversation attempt.

Why is the chatbot missing, or showing old settings?

Check the published page, installation location and Agent access before changing the prompt. A browser-loading problem is separate from an answer-quality problem.

SymptomWhat to inspectWhat to do
No launcher on the public pageSnippet placement and publication statePut the current snippet in the shared footer/body end, save and publish
Widget works elsewhere but not in an HTML blockWhether the builder creates an iframeUse site-wide code for the floating Widget; use the supported Embed snippet for a suitable inline container
The wrong Agent appearsScenario ID and duplicate installationsCopy the ID from the intended Agent and remove obsolete snippets or duplicate plugin installs
The launcher opens but replies failWidget & Embed access, credits and browser network errorsEnable the intended public channel, check the allowance and inspect the failed request
A color or greeting remains unchangedCached configuration or page HTMLAllow a few minutes, reload, then clear the site’s relevant cache if it still serves an old snippet
No lead email during a testConversation source and Email new leadsPreview and Simulation suppress email; test an actual widget conversation with your own controlled contact details
A complete-looking enquiry never reaches the CRMPersisted fields, conditions and Action logCheck every required key and the HTTP result; use a fresh conversation after fixing a sent failed request
No launcher after a theme changeWhether the new theme includes the installationRe-enable the Shopify app embed or restore the snippet in the new shared layout
An answer contradicts the sourceImported content and applied knowledge changesCorrect the source, finish processing/publishing and repeat a known-answer control question

If the browser reports a blocked resource, inspect its exact Content Security Policy or blocker error. Avoid disabling your site’s protections wholesale. Test the public page over HTTPS and use the channel documentation for the required setup.

What does a website chatbot cost?

The subscription and usage allowance pay for the AI service; your website plan and optional integrations are separate costs.

UseRelevant plan boundary
Personal, non-commercial learningFree includes 50 AI credits per month and the core Widget, Smart Data Gathering, Human Handoff and Actions
A business websiteStarter permits commercial use and includes 150 credits per month, from $9/month
Custom widget translations or voice inputAvailable on paid plans
Hide Quickchat AI BrandingProfessional or above

An AI credit is not a conversation. A conversation can contain several AI replies, and different model choices can have different credit costs. Include Preview and Simulation in the testing budget. Use the current pricing and allowances and the chatbot cost guide to estimate your volume.

Can ChatGPT or Claude help set up the chatbot?

The Quickchat AI account-management connection lets a compatible assistant configure supported Agent settings and prepare tests from these artifacts.

Connect your own account through Quickchat AI in ChatGPT or Quickchat AI in Claude, following the Manage with AI guide. The supported tools cover knowledge, Main Prompt, conversation starters, Human Handoff, HTTP Actions, testing and the deployment snippet.

Complete Smart Data Gathering and the detailed widget Appearance settings in the dashboard. You also need to authorize external accounts and install the snippet in your website’s settings. The account connection is for managing your Agent; it is separate from exposing an Agent as an MCP tool.

Copy a setup request for your assistant
Use the Kestrel Kitchens website-chatbot kit to prepare a separate learning Agent. Do not change existing Agents. Set its name to Kestrel Kitchens, add knowledge.txt and set main-prompt.txt as Main Prompt. Configure the supported greeting, conversation starters and Human Handoff settings from the kit. Prepare the launch-test dataset with the supplied messages and criteria. Read back the saved configuration and return this Agent's current Widget snippet.
List the remaining dashboard steps for Smart Data Gathering, Appearance and website installation. Do not invent support for those settings. Leave the optional Send quote enquiry Action inactive until I provide my receiver URL and configure its credential. Before running tests or external writes, tell me the expected credit use and effects and obtain any approval required by this client. Do not claim that a website, email or CRM delivery works until it has been tested.

Frequently asked questions

How do I add an AI chatbot to my website?

Create an Agent in Quickchat AI, add your business knowledge, configure Main Prompt, Smart Data Gathering and Human Handoff, then install its Widget using a plugin or the supplied snippet. Test a source-based answer, captured contact details and a delivered human reply on the published site.

Can I add a custom ChatGPT to my website?

You can add a ChatGPT-style assistant trained on your own pages. This tutorial deploys a Quickchat AI Agent through its website widget, so visitors can chat on your site without signing in to ChatGPT. Its knowledge, instructions and integrations are managed in Quickchat AI.

Can I add an AI chatbot to my website for free?

Quickchat AI’s Free plan supports personal, non-commercial learning and includes 50 AI credits per month. A company website or customer-facing sales and support deployment needs a plan that permits commercial use. Website-builder subscriptions are separate, and testing also uses AI credits.

Do I need to code or provide an API key?

The core setup uses dashboard settings and a plugin, app embed or supplied script. The website snippet contains a public Scenario ID, not a private API key. Connecting an optional CRM Action requires a suitable endpoint and its authentication; keep that credential in the Action settings.

Can a website chatbot collect leads and quote requests?

Yes. Smart Data Gathering asks for chosen fields during the conversation and saves them in Inbox. Email new leads can notify Agent admins after an email address or phone number is captured. An API Action can send the required saved fields to a configured receiver; a captured name alone does not establish a complete enquiry.

Can a person take over the chat?

Yes. Configure Human Handoff with staffed hours and an outside-hours message. In Inbox, select Assign to me for an unassigned conversation, or Take over when that control is shown, then reply to the visitor. Reassign to AI returns the conversation to the Agent.

Does it work on WordPress, Shopify, Squarespace, Wix and Webflow?

Each has an installation route. Use the official WordPress plugin, the Shopify app embed after connecting its store Agent, or the appropriate site-wide footer/body-end snippet for an existing Agent. Plugin permissions and custom-code plan requirements vary by builder.

Why is my chatbot not appearing on my website?

Check the published page, correct Scenario ID and snippet placement at the end of the page body. Remove obsolete duplicate installations and avoid iframe-based blocks for the floating Widget. If it opens but cannot answer, check Widget & Embed access and the available AI credits.

Will the chatbot slow down my website?

The widget loader downloads asynchronously, but it still adds JavaScript and network requests. Compare loading and interaction performance before and after installation on representative pages and a phone. Asynchronous loading does not establish zero performance impact.

Does the website chatbot work on phones?

The open widget uses a full-screen layout below 768 CSS pixels. Test the launcher, keyboard, scrolling and close control on a phone. Leave Auto-open Widget off when automatic opening would interrupt browsing.

Can the chatbot answer in my visitors’ languages?

The Agent can reply in the visitor’s language. Translating the fixed welcome text, conversation starters and widget labels is a separate paid-plan setting. Test both the answer’s facts and the interface text for each language you intend to support.

Quickchat AI Quickchat AI Ask Quickchat AI a question