How to Choose an AI Chatbot for Your Website: A Buyer Checklist

Adding an AI chatbot to your website sounds simple. Most website builders, help desks and CRMs now offer one, and dozens of standalone tools promise to answer customer questions around the clock. The hard part is telling them apart, because almost every demo looks impressive.

The differences show up after launch: a bot that makes up a refund policy, one that cannot pass a frustrated customer to your team, or a bill that doubles because traffic grew. This buyer checklist covers the 6 areas that matter most, with questions to ask every vendor before you sign.

Start with what you need the chatbot to do

Before comparing tools, write down the jobs. A chatbot that answers FAQs is a different purchase from one that books appointments or processes returns.

Common jobs for a website chatbot:

• Answer questions about products, services, hours, pricing and policies
• Capture leads by collecting contact details and qualifying visitors
• Book appointments or reservations through your scheduling system
• Look up orders or accounts and handle simple changes
• Route conversations to the right person or team

Pull your most common questions from email, phone notes and existing chat logs. If 80% of messages fall into 10 topics, those 10 topics are your test set. If you need a refresher on the terms vendors use, our explainer on AI agents vs chatbots vs copilots helps.

The 6-part buyer checklist

Use this as a scorecard. Rate each vendor 0 (missing), 1 (partial) or 2 (strong) on each line.

A tool that scores 2 on data sources and handoff but 1 on branding is usually a better buy than the reverse.

Key takeaway: Judge chatbots on accuracy, handoff and data handling first. A beautiful widget that gives wrong answers or traps customers costs more than it saves.

Data sources and accuracy

This is the most important line on the checklist. Modern AI chatbots usually work by searching your content and generating an answer from what they find, a pattern called retrieval-augmented generation. Our RAG explainer covers how it works.

Ask every vendor:

• Which sources can it use? Website crawl, help center, uploaded files, product catalog, databases.
• Can I restrict it to those sources only? You do not want general internet knowledge answering questions about your return policy.
• How often does it re-sync? If you change prices or hours, how quickly does the bot know?
• Does it show citations? Links to the source article build trust and help you debug wrong answers.
• What happens when it does not know? The right answer is a clear “I’m not sure, let me connect you,” not a confident guess.

Accuracy matters legally as well as practically. In Moffatt v. Air Canada (2024 BCCRT 149), a Canadian tribunal held the airline responsible when its website chatbot gave a customer wrong information about bereavement fares, rejecting the argument that the chatbot was responsible for its own statements. Treat what your bot says as what your business says.

Human handoff

Every chatbot should make it easy to reach your team. Check:

• Handoff options: live chat transfer, ticket creation, email follow-up, callback request
• Triggers: customer asks for a person, low confidence, repeated questions, negative sentiment, sensitive topics
• Context: the full transcript and a short summary travel with the handoff
• After-hours behavior: the bot tells people when to expect a reply instead of pretending someone is online

If you plan to automate more of your support, see our guide on how to automate customer service with AI, which covers handoff design in detail.

Privacy, security and disclosure

A website chatbot collects names, emails, order numbers and sometimes sensitive details customers type without being asked. Ask vendors:

• Is our data used to train your models or anyone else’s? Look for a clear contractual answer and an opt-out.
• How long are conversations stored, and can we delete them?
• Where is data processed and stored?
• What security documentation can you share? Many vendors provide a SOC 2 report, an independent audit of their security controls.
• Can we mask or block sensitive data such as card numbers or health details?

State privacy laws may give customers rights over chat data, so check your obligations in our state data privacy laws guide. California’s bot disclosure law (Business and Professions Code section 17941) also makes it unlawful to use a bot to mislead people in California about its artificial identity to influence a purchase, with clear disclosure as a safe harbor. Separately, website chat tools have been the subject of lawsuits in California under the state’s Invasion of Privacy Act over third-party vendors handling chat conversations, so make sure your privacy notice explains how chat data is processed. This is not legal advice.

Simple rule: label the bot as AI in its greeting and link to your privacy policy in the chat window.

Pricing models compared

Chatbot pricing varies widely and changes often. Understand the billable unit before comparing numbers.

2 published examples: Intercom lists its Fin AI agent at $0.99 per outcome, counting a conversation when the customer confirms it is resolved, does not ask for more help or Fin completes a workflow. Zendesk bills AI agents per automated resolution, with included allowances by plan. Always confirm current pricing on the vendor’s own page. For a full budget, including setup and training, see our guide to AI implementation cost.

Integrations, channels and accessibility

A chatbot that cannot see your systems can only answer generic questions. Check for native connections to:

• Help desk for tickets and live chat handoff
• CRM to log leads and recognize returning customers
• Booking or scheduling tools for appointments
• Ecommerce platform for order status, returns and product questions
• Other channels such as SMS, social messaging or voice, if you want one bot across channels

Ask how actions are secured. If a bot can change an order, it must verify the customer first.

Accessibility and user experience

Your chat widget is part of your website, so it should meet the same accessibility standard. WCAG 2.2, published by the W3C as a Recommendation in October 2023, is the current version of the Web Content Accessibility Guidelines. Test that the widget works with a keyboard only, reads properly with a screen reader, has sufficient color contrast and does not block content on mobile.

Also check: Can visitors close or minimize it easily? Does it pop open on every page? Aggressive pop-ups annoy more people than they help.

Where to start: a 5-step selection process

1. List your top 10 questions and 2 or 3 tasks the bot must handle.
2. Shortlist 3 tools, including any AI chat built into software you already use.
3. Run a trial with your real content. Ask each bot your 10 questions plus tricky ones: misspellings, off-topic questions, an angry message and a request for a person.
4. Score them using the checklist table above, and review privacy terms and pricing at your expected volume.
5. Launch on a few pages first, read conversation logs weekly for the first month and fix content gaps before rolling out site-wide.

The bottom line

The right AI chatbot for your website is the one that answers your customers’ real questions accurately from your own content, hands off cleanly to your team, handles data responsibly and has pricing you can predict. Test with real questions, not vendor demos, and keep a person one click away.

Sources

• California Business and Professions Code section 17941 (California Legislative Information) — bot disclosure rule and safe harbor
• California Penal Code section 631 (California Legislative Information) — Invasion of Privacy Act provision cited in chat lawsuits
• Web Content Accessibility Guidelines (WCAG) 2.2 (W3C) — current accessibility standard, Recommendation October 2023
• Intercom pricing — Fin AI agent per-outcome pricing and outcome definition
• Zendesk pricing — automated resolution billing for AI agents