What is an AI chatbot for a website?

An AI chatbot for a website is a chat widget, powered by a large language model, that answers your visitors' questions in natural language using your own business content. Unlike old rule-based bots, it understands questions it has never seen before and replies conversationally, 24/7 — so customers get instant, accurate answers without waiting for your team.

What can an AI chatbot do for my business?

It answers the repetitive questions your team types out every day, works around the clock in any language, captures leads with the visitor's email or phone, opens support tickets when someone needs a person, and shows you what visitors actually ask. In practice that means fewer inbound tickets, faster replies, and enquiries captured at 2am instead of lost.

How do I add an AI chatbot to my website?

Create a bot in the SpideyChat dashboard, teach it your business (describe it in plain language, point the crawler at your website, or upload documents), then paste one line of code before your closing tag. It works on WordPress, Shopify, Webflow, Wix, Squarespace, and custom React or Next.js sites — no plugin, and your AI chatbot is live in minutes. You can do the whole thing on the free plan first.

Can a chatbot answer from my website?

Yes — that's the default way to teach one. Point the crawler at your site and SpideyChat reads your pages, extracts the content, and keeps it as part of the bot's knowledge, so answers come from what you already published. Re-crawl whenever your site changes and the bot updates with it.

How do I train an AI chatbot — and do chatbots need training?

There's no model training on your side. The Knowledge Learner gives you four ways to teach your website chatbot: describe your business in plain language, point it at your website and let the crawler do the rest, upload documents (PDF, DOCX, TXT, CSV, or Markdown), or add manual Q&A pairs. Everything is chunked and embedded into a private vector index, and the AI answers using only the most relevant passages.

How do chatbots learn my business?

By indexing the content you give them, not by memorising conversations. Your description, crawled pages, uploaded files, and Q&A pairs are split into passages and stored in a private index for that bot. When a visitor asks something, the bot retrieves the closest passages and answers from those — which is why the answers sound like your business rather than generic AI.

Do I need to code an AI chatbot, or hire a developer?

Neither. Teaching your bot and customizing the widget all happen in the dashboard. Installing it is a single copy-paste script tag — the same kind you'd use for any analytics tool — so anyone who can edit their website's footer can ship it.

Do AI chatbots hallucinate?

General-purpose chatbots can. SpideyChat's answers only from the content you gave it and is instructed not to guess, which is what keeps it grounded. When it doesn't have the answer it says so rather than inventing one, and the visitor can open a support ticket from the same chat.

What happens when a chatbot doesn't know the answer?

It tells the visitor it doesn't have that information instead of improvising, and offers the widget's “Report an issue” option, which files a numbered support ticket right inside the chat. On the Pro plan and above those tickets land in your dashboard inbox so your team can pick the conversation up.

How accurate are AI chatbots?

Accuracy comes from the content, not the model. Every bot runs on a state-of-the-art large language model tuned for fast, high-volume support, and it answers from your own material — so it's as accurate as what you taught it. Thin or out-of-date content is the usual cause of a weak answer, and you can fix that by editing the knowledge base or re-crawling.

Do AI chatbots really work?

For the questions your content already answers — pricing, hours, policies, how-to, availability — yes, and those are the bulk of what most sites get asked. They're not a replacement for human judgement on complex or sensitive cases, which is why handoff and tickets matter: the bot clears the repetitive volume and routes the rest to you.

Can AI chatbots reduce support tickets?

That's the main reason teams install one. Repeat questions — shipping, refunds, opening hours, “how do I…” — get answered instantly from your own content instead of becoming emails. What's left reaches you as a ticket with the conversation attached, so you start with context instead of a blank thread.

What languages do AI chatbots support — can they handle multiple languages?

Your bot replies in the visitor's own language automatically, and one bot handles all of them. Because it runs on a multilingual language model, a single website chatbot can greet and answer visitors around the world in their native language, even when your knowledge base is written in just one language. There's nothing to configure.

Can AI chatbots capture leads?

Yes. Bots can collect visitor details like email and phone during a conversation (lead capture on Starter and above), so an after-hours question becomes a contactable lead instead of a closed tab. Captured leads appear in your dashboard alongside the conversation that produced them.

How do chatbots qualify leads?

By asking the same few questions a good salesperson would — what they're looking for, budget or timeline, and how to reach them — inside a normal conversation. You get the visitor's answers with the full chat transcript, so you can tell a serious enquiry from a browser before you spend time on it.

Can a chatbot hand off to a human?

Yes. Any visitor can open a numbered support ticket through the widget's “Report an issue” option, and on the Pro plan and above those tickets land in your dashboard inbox with the conversation attached. Both you and the visitor get an email, so nothing sits unnoticed.

Can a chatbot book appointments?

It captures appointment and booking requests — what the visitor needs, preferred times, and their contact details — and routes them to your team to confirm. It doesn't write directly into your calendar, so you stay in control of what actually gets booked.

Can AI chatbots replace live chat?

For most of the volume, yes — the bot answers instantly, 24/7, without anyone staffing the queue. What it doesn't do is judgement calls, so treat it as the first line: it handles the repetitive questions and hands the rest to you as a ticket rather than leaving someone waiting for an agent.

Is a chatbot better than a contact form?

A contact form takes a message and gives nothing back; a chatbot answers on the spot, and still collects the visitor's details when it can't. For anything already covered by your content, the visitor gets their answer immediately instead of waiting a day for a reply — and you get fewer “just checking” emails.

Are AI chatbots secure — is my data safe with an AI chatbot?

Your data stays yours. Everything you use to teach a bot — your description, crawled pages, uploaded documents, and Q&A pairs — is stored privately in your account and indexed separately for each bot, so information never bleeds between bots. Data is encrypted in transit and at rest, and you can review every conversation and edit the knowledge base at any time.

Which website platforms does the chatbot work with?

Any website. Installation is a single tag (or an iframe embed), so SpideyChat drops into WordPress, Shopify, WooCommerce, Webflow, Wix, Squarespace, Ghost, Framer, and custom React or Next.js sites alike — the same kind of snippet you'd paste for an analytics tool. No plugin, framework, or platform lock-in required.

How much does an AI chatbot for a website cost?

There's a Free plan at $0 forever — one chatbot and 100 messages a month, no credit card required. Paid plans are Starter at $20/month (3 bots, 2,000 messages, website crawler, and lead capture), Pro at $59/month (10 bots, 9,000 messages, the support-ticket inbox, and branding removal), and a custom-priced Enterprise tier for unlimited bots and volume. Annual billing lowers the monthly rate, and you can upgrade, downgrade, or cancel anytime.

Is there a free AI chatbot for websites?

Yes — SpideyChat's Free plan is free forever: one chatbot and 100 messages a month, no credit card required. It's the whole product at a smaller size, so you can train a bot and put it on your site before paying anything. Upgrade when you outgrow it, and downgrade or cancel anytime.

What is the best AI chatbot for a small business?

The one you can actually get live and keep current. For a small team that usually means no-code setup, a free tier to test with, one-line installation, and answers grounded in your own content — which is exactly what SpideyChat is built around. If you need deep CRM workflows and a dedicated admin, a heavier enterprise suite may suit you better.

How long does it take to set up a chatbot?

Minutes. Create a bot, teach it your business — paste a description, point it at your website, or upload a few documents — customize the widget, then copy one line of code onto your site. There's nothing to install and no model to train yourself: your AI chatbot is live the moment you paste the snippet.

Reference

AI chatbot & AI agent glossary

Plain definitions of the terms used around website AI agents. Neutral and industry-standard, not product descriptions — where a term has a wider meaning than the way we happen to use it, the wider meaning is the one given here.

What is the difference between a chatbot and an AI agent?

A chatbot holds a conversation and produces replies. An AI agent is given a set of tools it may call — recording a lead, opening a ticket, fetching a human — and the surrounding software validates and executes those calls. The difference is not the language model but whether anything is connected to the other end, so an agent completes tasks where a chatbot only describes them.

Agents

AI employee also: AI website employee: An AI employee is software placed on a website that performs front-of-house work: it answers visitor questions from the business's own content, asks qualifying follow-ups, records the enquiry, and escalates to a person when it cannot help. The distinguishing feature is that it completes tasks rather than only producing replies.; The practical test is what exists after a conversation ends. A transcript means a chatbot; a recorded lead, a ticket or a notified human means an employee.; What is an AI employee? →
AI agent: An AI agent is a system that uses a language model to decide which actions to take toward a goal, rather than only generating text. It is given a set of tools it may call, and the surrounding software validates, authorises and executes each call. Agency comes from the tools, not from the model.; AI agent vs chatbot →
AI receptionist: An AI receptionist handles the first contact a visitor has with a business: answering common questions, screening whether the enquiry is worth the team's time, and recording the details. Most operate on a website rather than a phone line, which distinguishes them from a human answering service.; What an AI receptionist does →
Chatbot: A chatbot is software that holds a text conversation with a user. The term covers everything from a fixed decision tree to a language model answering freely. In current usage it usually implies a system that replies but does not take action beyond the conversation itself.

How it works

Retrieval-augmented generation (RAG) also: RAG: Retrieval-augmented generation is a technique where relevant passages are retrieved from a document store and supplied to a language model, which is then instructed to answer only from them. It is the standard way to make a model answer accurately about a specific business rather than from general training data.
Grounding: Grounding is the practice of restricting a language model to answering from a supplied set of sources, and requiring it to say so when those sources contain nothing relevant. An ungrounded model will produce a plausible answer regardless, which is how chatbots come to quote prices a business does not charge.; Stop a chatbot inventing answers →
Hallucination: A hallucination is output that is fluent and confident but not supported by any source. It is not a malfunction: a language model generates plausible continuations, so when it has no relevant information the plausible continuation is an invention. Grounding and an explicit permission to say 'I do not know' are the practical defences.
Embedding: An embedding is a list of numbers representing the meaning of a piece of text, positioned so that texts with similar meanings sit close together. Storing embeddings of a website's content allows a search for 'do you fix boilers' to retrieve a passage about heating repairs even with no shared words.
Vector search also: semantic search: Vector search finds documents by comparing embeddings rather than matching keywords, ranking results by distance in meaning. It is what allows a chatbot to answer a question phrased in the visitor's own words when the website uses different terminology entirely.
Tool calling also: function calling: Tool calling is the mechanism by which a language model requests that a specific function be run, returning a structured call rather than prose. The application decides whether the call is permitted, validates its arguments, executes it, and returns the result. This is what separates an agent from a chatbot.; What an AI employee can do →
Knowledge base: A knowledge base is the set of content an AI system is permitted to answer from — typically a website's pages, uploaded documents and manually written question-and-answer pairs. Its coverage sets the ceiling on answer quality, which is why a thin website produces a vague assistant.; What content a chatbot needs →
Prompt injection: Prompt injection is an attack where text placed in content the model reads — a crawled web page, an uploaded file, a user message — is written to influence the model's behaviour. It matters most for agents, because a model that can call tools can be pushed toward calling them, so authorisation must be enforced outside the model.

Operating it

Handover also: escalation, handoff: A handover is the transfer of a conversation from an automated agent to a person. Good practice is to hand over immediately on request, when the knowledge base has nothing relevant, when emotion or money is involved, and when the conversation has stopped progressing.; When to hand over to a human →
Knowledge gap: A knowledge gap is a question an assistant could not answer because nothing relevant existed in its knowledge base. Logged systematically, gaps become the most actionable report a chatbot produces: each one names a specific piece of content missing from the website.
Lead qualification: Lead qualification is establishing whether an enquiry is worth pursuing, and gathering enough detail to act on it. A qualifying question is one whose answer changes what happens next — location, scope, timescale, budget — as opposed to a question that merely collects information.; Can a chatbot qualify leads? →
Containment rate also: deflection rate: Containment rate is the share of conversations resolved without a human. It is widely reported and easy to misread, because a visitor who gave up in frustration counts identically to one who was helped. Handover reasons are a more honest measure of whether an assistant is working.; Chatbot metrics that matter →
Chat widget also: embed, launcher: A chat widget is the interface a visitor sees on a website: a launcher button and the panel it opens. It is normally added with a single script tag, and it loads in an iframe so the host site's styles and the widget's cannot interfere with each other.

Commercial

Per-resolution pricing: Per-resolution pricing charges each time an AI resolves a query without human involvement. It aligns cost with delivered value, and it makes spending unpredictable: the bill rises with traffic and with the assistant's own effectiveness, which is uncomfortable for a business that needs to forecast.; Pricing models explained →
Flat pricing: Flat pricing charges a fixed monthly fee for a plan with a stated allowance, regardless of how many conversations occur within it. Its advantage is forecastability: a seasonal spike or a successful campaign costs the same as a quiet month.; SpideyChat pricing →
White label: White labelling is removing a vendor's branding so the product appears to belong to the business deploying it. For chat widgets this usually means removing a 'powered by' line, and for agencies it can extend to custom domains and sender addresses.

Want to see these ideas working rather than defined? Build an AI employee from your own website — free, no signup.

What is the difference between a chatbot and an AI agent?

Agents

AI employee

AI agent

AI receptionist

Chatbot

How it works

Retrieval-augmented generation (RAG)

Grounding

Hallucination

Embedding

Vector search

Tool calling

Knowledge base

Prompt injection

Operating it

Handover

Knowledge gap

Lead qualification

Containment rate

Chat widget

Commercial

Per-resolution pricing

Flat pricing

White label