Old website chatbots followed rigid decision trees and frustrated visitors. Chatbots based on large language models (LLMs) understand natural language, answer follow-up questions and adapt to the conversation.
What they do well
- Answer common questions about products, services, prices and opening hours
- Qualify leads by asking the right questions and summarising requirements
- Guide visitors to the right page or offer
- Work 24/7 and in many languages
- Hand over to a human with a summary of the conversation
The project assistant on this website is an example: it interviews visitors about their project, creates a structured brief and sends it to our team.
Limitations
- They can make mistakes or invent answers if not grounded in your information
- They should not make binding commitments (prices, contracts) on their own
- Sensitive topics need clear rules and escalation to humans
How to make them reliable
- Ground answers in your own verified content (retrieval-augmented generation).
- Define scope: what the bot answers and what it hands over.
- Write clear instructions about tone, language and boundaries.
- Use tools for structured actions – for example saving a lead or checking availability – instead of letting the bot improvise.
- Test with real questions before launch and keep improving.
- Monitor conversations (with privacy in mind) to find gaps.
Privacy
Inform users that they are talking to an AI, avoid collecting unnecessary personal data, use providers with business terms that don't train on your data, and document data flows for GDPR.
Measuring success
Resolution rate, leads generated, response time, customer satisfaction and the number of support tickets avoided.
A well-built AI assistant feels like a helpful colleague available around the clock. We design, build and integrate them into websites, apps and internal tools.