How to choose the right AI chatbot — a 7-point checklist
The market for AI chatbots is a jungle in 2026. From plug-and-play widgets to enterprise solutions with six-figure price tags. How do you choose? Here is the checklist we use when advising SMBs.
1. Does it actually understand what the customer says?
Many "AI" chatbots in the market are still rule-based (decision trees). They only work if the customer picks between predefined options. A real AI chatbot built on modern LLMs (GPT-5, Gemini 3, Claude) understands free text. Always test: write a poorly worded question and see what happens.
2. Can it actually retrieve information from your site?
A chatbot without your own knowledge is useless. It must index your website, price lists, FAQs and PDFs. Demand a demo of retrieval-augmented generation (RAG). If the vendor doesn't know what that is, move on.
3. Can it actually act — not just answer?
Real value lies in action. Booking, creating a lead in the CRM, sending an SMS, alerting the right person. A chatbot that only says "Great question, we'll email you" is a waste. Ask concretely: "What actions can your bot perform in my systems?"
4. How good is the qualification?
Not all visitors are real customers. A good chatbot asks a few smart follow-up questions to separate qualified from unqualified: industry, size, budget, timeline. You shouldn't spam your phone with "do you have summer jobs?" inquiries.
5. How smooth is the handoff to a human?
AI is not always the answer. When the customer is angry, confused or has a complex case, the bot must identify it and hand off — ideally in real time via Slack/Teams/SMS alerts. Zendesk data shows 75 % of customers expect to be transferred to a human when needed.
6. Do you get your data out?
If the chatbot stores all conversations and leads in a closed system you can't access, you are trapped. Demand API/export, CRM integration, and full insight into conversation history. You own the data — not the vendor.
7. What does it actually cost?
Pricing models vary wildly. Most common:
- Per conversation (typically €0.50–1.50/conversation) — unpredictable if traffic varies
- Per month, fixed (€150–800) — predictable
- Per lead (€5–30 per qualified lead) — useful if quality is high
Always check what's included: AI model costs, updates, knowledge base retraining, support.
Bonus test: do they use it themselves?
Many chatbot vendors have… no chatbot on their own site. Or one that is obviously bad. That's a strong signal. A vendor that doesn't trust their own product shouldn't build yours.
Checklist summarized
- ✅ Understands free text (LLM-based, not rule-based)
- ✅ Retrieves info from your site (RAG)
- ✅ Can perform actions (booking, CRM, alerts)
- ✅ Qualifies leads with smart questions
- ✅ Clear handoff to human
- ✅ You own and can export the data
- ✅ Predictable pricing model
Sources
Frequently asked questions
How long does implementation take?+
A serious AI chatbot is typically live in 1-2 weeks, including integrations and training. Be skeptical of both '24 hours' (likely thin) and 'multiple months' (likely over-engineered).
Do I need technical skills to operate it?+
No, if the vendor handles operations. You should be able to update the FAQ and knowledge base from a simple UI — without writing prompts.
How do I avoid the chatbot 'making things up' (hallucinating)?+
Choose a vendor who strictly grounds the bot in your knowledge base and tests it regularly. A requirement that the bot says 'I don't know' when uncertain is critical.
Want to know what your website can do better?
Book a free review — we'll look at your website and give you concrete suggestions.
Free ReviewTakes 2 min · No commitment · Reply within 24 hours