Short answer
For most small and mid-sized businesses, the right website chatbot in 2026 is a RAG chatbot: a language model that answers only from your own pages and documents, captures the lead, and hands off to a person when it should. Off-the-shelf tools cost from about US$25 to US$150 a month at small volumes, or US$0.99 per resolved conversation on Intercom Fin; the language-model usage behind a chatbot is often only tens of dollars a month. A custom build makes sense when the bot must act in your own systems, such as booking, quoting or looking up an order. Below: the four types of chatbot, sourced prices, a token-cost calculation, the legal risks and how to measure results.
Four kinds of website chatbot
Website chatbots fall into four types: rule-based bots that follow a script, LLM chatbots that write free-form answers, RAG chatbots that answer from your own documents, and agents that also take actions in your systems. Cost, usefulness and risk all rise in that order, so pick the simplest type that does the job.
A rule-based bot is a decision tree: buttons and keywords lead to prepared answers. It never invents anything, but it frustrates visitors whose question is not on the menu. A plain LLM chatbot is a general language model with a prompt; it sounds fluent but answers from its general training, so it can confidently state prices or policies that are not yours. RAG (retrieval-augmented generation) fixes most of that: before answering, the system searches your website, FAQ, catalogue or policies and gives the model only the relevant passages, with an instruction to answer from them or say it does not know. An agent adds tools: it can check a calendar, create a CRM contact, look up an order or open a ticket.
| Type | How it answers | Good for | Main risk |
|---|---|---|---|
| Rule-based | Scripted buttons and keywords | Very narrow FAQs, routing to the right form | Dead ends; visitors give up |
| LLM chatbot | General model plus a prompt | Drafting, general questions | Invents facts about your business |
| RAG chatbot | Retrieves your content, then answers from it | Support, pre-sales questions, lead capture | Wrong or outdated source documents |
| Agent | RAG plus tools that act (book, look up, create) | Booking, order status, quoting, qualification | Wrong action taken on the business’s behalf |
Rule of thumb: if the answers live in documents you already have, you need RAG. If the visitor needs something done, not just answered, you need an agent with limited, logged permissions.
Is it worth it? What the adoption data says
Chatbots are now one of the most common business uses of AI, but most small businesses have not deployed one yet. Statistics Canada reports that 19.2% of Canadian businesses used AI to produce goods or deliver services in the 12 months before the second quarter of 2026, and that 28.2% of those AI users had deployed virtual agents or chatbots, the third most common use after data analytics and text analytics.
The business case is usually speed. In the lead-response research published in Harvard Business Review, an audit of 2,241 US companies found that only 37% responded to a web lead within an hour and 23% never responded at all, and firms that tried to contact a lead within an hour were nearly seven times as likely to qualify it as those that waited even an hour longer. A chatbot that answers at 11 p.m. and books the call is the cheapest way to close that gap.
Off-the-shelf chatbot tools and their real prices
Off-the-shelf tools are the fastest route: you connect your website content, adjust the tone and install a script. They price in three different ways, per resolution, per conversation or per message credit, so compare them on your own monthly volume rather than on the headline price.
Intercom charges US$0.99 per outcome for its Fin AI Agent, counted once per conversation when the customer confirms the issue is resolved, does not ask for more help, or Fin completes a workflow, including a handoff. Intercom seats start at US$29 per seat a month (Essential), with a one-seat minimum. Tidio’s Lyro AI Agent add-on starts at US$32.50 a month for 50 AI conversations and is sold standalone or with Tidio plans, which start at US$24.17 a month; Tidio also sells a Premium tier with a guaranteed 50% Lyro resolution rate on custom pricing. Chatbase sells message credits: US$40 a month for 700 credits (Hobby), US$150 for 4,000 (Standard) and US$500 for 15,000 (Pro). Zendesk includes AI agents in every Suite and Support plan and bills them on successful outcomes, but its pricing page does not publish a per-resolution rate; Suite Team is listed at US$55 per agent a month on annual billing.
| Tool | Pricing unit | Published entry price (USD) | Best fit |
|---|---|---|---|
| Intercom Fin | Per outcome (resolved conversation) | US$0.99 per outcome + seats from US$29/seat/month | Support teams already on Intercom or another help desk |
| Tidio Lyro | Per AI conversation (quota) | Add-on from US$32.50/month for 50 conversations; plans from US$24.17/month | Small e-commerce and service sites |
| Chatbase | Message credits | US$40/month (700 credits) to US$500/month (15,000 credits) | Simple RAG bot built on your site and files |
| Zendesk AI agents | Successful outcomes (rate not published) | Suite Team US$55/agent/month, annual billing | Teams already running Zendesk |
Prices from each vendor’s pricing page as viewed on September 28, 2026, before taxes. Tidio and Chatbase advertise discounts for annual billing. Per-outcome pricing is predictable per conversation but grows linearly with success: 1,000 resolved conversations on Fin cost US$990 a month before seats.
What the AI itself costs: token pricing
The language model behind a chatbot is billed per token, and for a typical small-business website the bill is usually in the tens of dollars a month, not hundreds. Anthropic’s documentation estimates one token at about 0.75 English words, and its own worked example puts 10,000 support conversations of about 3,700 tokens each at roughly US$37 on Claude Haiku 4.5.
Tokens come in two kinds: input (your instructions, the retrieved passages and the conversation so far) and output (the answer). RAG chatbots are input-heavy because every turn resends context, so input price matters more than it looks. The table below prices an illustrative month of 1,000 conversations, each using 15,000 input tokens and 1,000 output tokens across several turns. This is an example calculation, not a client measurement.
| Model (provider) | Input / output per million tokens | Illustrative month (1,000 conversations) |
|---|---|---|
| GPT-5.4-Nano (OpenAI) | US$0.20 / US$1.25 | about US$4 |
| Gemini 3.5 Flash-Lite (Google) | US$0.30 / US$2.50 | about US$7 |
| GPT-5.4-Mini (OpenAI) | US$0.75 / US$4.50 | about US$16 |
| Claude Haiku 4.5 (Anthropic) | US$1 / US$5 | about US$20 |
| Claude Sonnet 5.5 (Anthropic) | US$2 / US$10 | about US$40 |
| Gemini 3.1 Pro Preview (Google) | US$2 / US$12 (prompts up to 200k tokens) | about US$42 |
Standard (paid-tier) API prices from the OpenAI, Google Gemini and Anthropic pricing pages as viewed on September 28, 2026. Prompt caching lowers repeated input costs (Anthropic bills cache hits at 10% of the base input price on most models). Embedding your documents for search is cheap: OpenAI lists text-embedding-3-small at US$0.02 per million tokens.
What a custom AI chatbot costs to build
A custom chatbot costs more upfront than a SaaS subscription, typically a few thousand dollars for a scoped RAG bot and more for an agent connected to several systems, but its running cost is mostly tokens and hosting rather than per-resolution fees. Custom is worth it when the bot has to act in your own tools, when you need control over where data is processed, or when per-outcome fees would exceed the build cost within a year or two.
For market context, Clutch’s September 2026 AI development pricing guide reports that the most common AI project size on its platform is US$10,000 to US$49,999, with an average of US$120,594 skewed upward by large enterprise projects, and a most common hourly rate of US$25 to US$49 among the firms it lists. A website chatbot sits at the small end of that range when it is well scoped.
- Knowledge base: collecting and cleaning the pages, PDFs and policies the bot will answer from, and deciding who keeps them current.
- Retrieval and prompts: search over your content, answer rules (cite the source, say “I don’t know”, never quote a price that is not in the documents).
- Integrations: CRM, calendar, e-commerce or order system, help desk, email and SMS.
- Guardrails: topics the bot refuses, escalation to a human, logging, and a disclosure that the visitor is talking to an AI.
- Evaluation: a test set of real questions with expected answers, run before launch and after each content change.
- Operation: monitoring conversations, fixing gaps in the content, and adjusting as your offer changes.
Risks: hallucination, liability and disclosure rules
The main risk is a confident wrong answer, and the business, not the software vendor, is responsible for it. In Moffatt v. Air Canada (2024 BCCRT 149), Air Canada’s website chatbot told a customer he could apply for a bereavement fare retroactively, which contradicted the airline’s own policy page. The British Columbia Civil Resolution Tribunal called the airline’s argument that the chatbot was a separate legal entity “a remarkable submission”, held the airline liable for negligent misrepresentation and ordered it to pay $812.02: $650.88 in damages, $36.14 in interest and $125 in tribunal fees.
In the European Union, Article 50 of the AI Act applies from 2 August 2026: providers must ensure that people are informed they are interacting with an AI system, unless that is obvious to a reasonably well-informed person. The European Commission notes fines of up to €15 million or 3% of worldwide annual turnover for these obligations, with proportionality for small businesses. Showing “You are chatting with an AI assistant” in the chat window costs nothing and also helps with trust in Canada and the US.
- Answer only from approved content, and make “I don’t know, let me connect you” an acceptable answer.
- Never let the bot state prices, discounts, refunds or legal terms that are not written in its source documents.
- Give agents the minimum permissions: read availability and create bookings, not cancel or refund.
- Log every conversation and review a sample weekly, especially the ones that ended without resolution.
- Disclose that it is an AI and offer a human hand-off at any point.
- Know where conversations are stored; personal information collected by the bot falls under Québec Law 25, PIPEDA, GDPR or US state privacy laws depending on who the visitor is.
Using a chatbot for lead capture
A chatbot captures more leads when it answers the question first and asks for contact details second. Visitors who get a useful answer are far more willing to leave an email or phone number to receive a quote, a booking confirmation or a follow-up.
- Answer, then qualify: one or two questions that matter to your sales team (project type, timeline, location, budget range).
- Ask for contact details only when there is a reason: sending the quote, confirming the booking, having a specialist call back.
- Push the lead straight into your CRM with the full conversation attached, so the salesperson does not ask the same questions again.
- Trigger an immediate human notification for high-value leads, and an automated follow-up for the rest.
- Offer to book a call directly in the calendar instead of promising that “someone will contact you”.
- Respect consent rules for follow-up: in Canada, commercial emails and texts sent afterwards fall under CASL.
How to measure whether your chatbot works
Measure a chatbot on outcomes, not on the number of conversations. A bot that talks to many visitors but never resolves anything or captures a lead is a cost, not an asset. Set a baseline before launch, then compare after 30 and 90 days.
| Metric | What it tells you | How to measure |
|---|---|---|
| Resolution rate | Share of conversations solved without a human | Conversations with no hand-off and no repeat contact within 7 days |
| Hand-off rate | Where the bot’s knowledge stops | Escalations divided by conversations; read the transcripts |
| Leads captured | Direct revenue contribution | Contacts created from chat in the CRM, and their conversion to sales |
| Answer accuracy | Hallucination risk | Weekly sample of transcripts scored against your source documents |
| Cost per resolved conversation | Whether SaaS or custom is cheaper for you | Monthly tool and token cost divided by resolved conversations |
| Customer satisfaction | Whether visitors like the experience | Thumbs up or down, or a one-question survey at the end of the chat |
Illustrative break-even, not a client case: at US$0.99 per outcome, 800 resolved conversations a month cost US$792 on Fin. A custom RAG bot whose token cost is under US$50 for that volume recovers a one-time build cost faster as volume grows; below a few hundred conversations a month, SaaS is usually cheaper.
What ZeniTech charges
ZeniTech is a web, app, automation and AI agency based in Québec City, working remotely with clients across North America and Europe on Eastern Time, in English or French. We build RAG chatbots and AI agents that answer from your content, capture leads into your CRM and hand off to your team. Our AI agents are powered by Orvel AI, ZeniTech’s agent layer operated in Québec: the orchestration, context and integrations that sit on top of leading language models.
Every project starts with a written scope and a fixed price. The code repository is in the client’s name from the first commit. Prices are in Canadian dollars, before taxes; USD figures are approximate.
| Service | Price (CAD, before taxes) | Approx. USD |
|---|---|---|
| AI agent or chatbot setup (powered by Orvel AI) | $3,000 to $7,500 | approx. $2,120 to $5,290 |
| AI agent operation and improvement | $750 to $1,500/month | approx. $530 to $1,060/month |
| Simple automation workflow (e.g. chat lead to CRM and follow-up) | $750 | approx. $530 |
| Complex workflow (several systems, branching, error handling) | $2,500 | approx. $1,765 |
| Monitoring and maintenance of automations | $250/month | approx. $175/month |
| Website, if you need a new one | From $3,500 | approx. $2,470 |
| Work outside the agreed scope | $125/hour | approx. $88/hour |
Conversions use the Bank of Canada rate of 1.4168 CAD per USD (September 28, 2026) and are rounded. Language-model usage, SMS and third-party subscriptions are billed at cost or directly to you, in your name.
Frequently asked questions
How much does an AI chatbot for a website cost?
Off-the-shelf tools start around US$25 to US$150 a month at small volumes: Chatbase is US$40 a month for 700 message credits and Tidio’s Lyro add-on starts at US$32.50 for 50 AI conversations. Intercom Fin charges US$0.99 per resolved conversation plus seats. A custom RAG chatbot or agent costs more upfront; at ZeniTech setup is CA$3,000 to CA$7,500, then CA$750 to CA$1,500 a month.
What is the best AI chatbot for a small business?
The best choice depends on volume and what the bot must do. For answering questions from your website and files, a RAG tool like Chatbase or Tidio Lyro is quick to launch. If you already use Intercom or Zendesk, their AI agents fit your help desk. If the bot must book appointments, look up orders or feed your CRM with qualified leads, a custom agent usually fits better.
What is a RAG chatbot?
RAG stands for retrieval-augmented generation. Before answering, the chatbot searches your own content, such as web pages, FAQs, catalogues or policies, and passes only the relevant passages to the language model with instructions to answer from them. This greatly reduces invented answers compared with a general chatbot, and updating the bot means updating your documents rather than retraining anything.
How much do the AI tokens behind a chatbot cost?
Usually little. Anthropic estimates about US$37 for 10,000 support conversations on Claude Haiku 4.5. In our illustrative example of 1,000 conversations using 15,000 input and 1,000 output tokens each, the monthly model cost ranges from about US$4 on GPT-5.4-Nano to about US$40 on Claude Sonnet 5.5, based on each provider’s published prices.
Can my business be held liable for what its chatbot says?
Yes. In Moffatt v. Air Canada (2024), a British Columbia tribunal rejected the airline’s argument that its chatbot was a separate legal entity and ordered it to pay $812.02 after the bot gave wrong bereavement-fare information. Limit the bot to approved content, forbid it from inventing prices or policies, log conversations, and offer a human hand-off.
Do I have to tell visitors they are talking to an AI?
In the European Union, yes: Article 50 of the AI Act, applicable from 2 August 2026, requires that people be informed they are interacting with an AI system unless it is obvious. The European Commission notes fines of up to €15 million or 3% of worldwide turnover. Elsewhere it is good practice anyway, and a simple label in the chat window is enough.
Should I choose a SaaS chatbot or a custom build?
Choose SaaS when you mainly need answers from your content and volume is modest. Choose custom when the bot must act in your systems, when you want control over data location, or when per-resolution fees add up: 800 resolved conversations a month cost US$792 on Intercom Fin, while the model tokens for that volume are often under US$50.
How do I know if my chatbot is working?
Track outcomes, not chat volume: resolution rate, hand-off rate, leads created in the CRM and their conversion, answer accuracy on a weekly transcript sample, cost per resolved conversation and visitor satisfaction. Measure a baseline before launch and compare after 30 and 90 days. Read the unresolved conversations; they show exactly which content to add.
Related
Free consultation
Get a fixed price for your project in 30 minutes.
We look at what the project has to achieve, tell you what it costs, and tell you when you do not need it.
Book a call →+1 581-748-7017Sources
- Statistics Canada: Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026
- Oldroyd, McElheran and Elkington: The Short Life of Online Sales Leads, Harvard Business Review (2011)
- Intercom: Pricing (viewed September 2026)
- Tidio: Pricing (viewed September 2026)
- Chatbase: Pricing (viewed September 2026)
- Zendesk: Pricing (viewed September 2026)
- Anthropic: Claude API pricing (viewed September 2026)
- OpenAI: API pricing (viewed September 2026)
- Google: Gemini API pricing (viewed September 2026)
- Clutch: AI Development Pricing Guide (updated September 2026)
- Dentons: Airline ordered to compensate a B.C. man because its chatbot provided inaccurate information (Moffatt v. Air Canada, 2024 BCCRT 149)
- McCarthy Tétrault: Moffatt v. Air Canada, a misrepresentation by an AI chatbot
- European Commission: transparency obligations under Article 50 of the AI Act
- Bank of Canada: Daily exchange rates (September 28, 2026)