AI & visibility

AI-powered live chat for your website: what to do, what to avoid, how to prepare

An AI assistant on your site should reply in the visitor's language, never invent prices and hand over to a person when needed. A guide to setup, KVKK and measurement.

Collobr Team 8 min read

The job of the AI assistant on your website is to answer the visitor's question in their own language, based only on the information you have provided, to hand over to one of your staff where it is not sure, and to record the conversation as a sales opportunity when the visitor gives their contact details of their own accord. Most of the success is decided before setup: a clean product catalog, up-to-date frequently asked questions, clear business hours and written handover rules. Because you collect visitor data, you also need to plan from the start how you will meet your duty to inform under KVKK (the Turkish Personal Data Protection Law), if your business falls under Turkish law.

What should the assistant do?

A good assistant takes over the first five minutes of an experienced sales rep. It greets the visitor, understands what they are looking for and points them in the right direction. The tasks below are the core of this first contact.

What should the assistant not do?

AI models write fluently, and fluency makes wrong information look convincing too. That is why the list of prohibitions is at least as important as the list of tasks.

How should you prepare before setup?

The assistant cannot answer better than the information you give it. Setting aside a day or two of desk work for preparation saves weeks you would otherwise spend fixing things afterward.

1. Clean up your product catalog

Each product should have a clear name, a short and concrete description, who it is suitable for and, if applicable, its price. Internal jargon like "Premium package" does not match visitors' questions; wording like "up to 10 users, setup included" does. Remove discontinued products from the catalog, or the assistant will keep recommending them.

2. Write down your frequently asked questions

List with your team the questions you hear most often by phone and email: shipping time, payment options, setup, returns, invoicing. Write a two- or three-sentence answer to each, in your own voice. This text becomes the assistant's main reference source.

3. Clarify business hours and handover rules

A handover rule should not be as vague as "pass to a person if needed." Which topics are handed over, who they go to during business hours and what the visitor is told outside business hours should all be written down. A sample set of rules:

4. Describe the tone

Describe in a few sentences how your brand talks to customers: formal address, short sentences, no emojis or only sparing ones, technical terms explained when needed. If you add two or three sample replies, the assistant adapts to this style much more easily.

5. Test it yourself before going live

Have a few people on your team ask tough questions like real customers: a product not in the catalog, a question in a foreign language, price haggling, an angry complaint. Every answer the assistant makes up or fails to hand over reveals a gap to add to the catalog or the rules.

What should you watch for under KVKK?

During a chat, the assistant receives personal data such as names, phone numbers and email addresses. At this point, your business must fulfill its obligations as a data controller under Law No. 6698 on the Protection of Personal Data. The main points are:

Chat content can sometimes include special categories of data too; for example, a visitor may mention their health condition. If such data comes up often in your sector, it is wise to review your privacy notice and processes with a lawyer.

How do you measure success?

Online you will find general figures like "AI increases sales by this much," but the only meaningful comparison for your site, your product and your visitors is your own data. Note the current state for a few weeks before switching the assistant on, then track the same metrics regularly.

MetricHow it is calculatedWhat it shows
Number of leadsNumber of records from chat that include contact details (weekly)The assistant's contribution to the sales pipeline
Lead qualityShare of these records that move on to the quote stageWhether incoming records are real buyers
First response timeTime between the visitor's first message and the first replyWhether visitors are kept waiting
Handover rateChats passed to a person / total chatsHow sufficient the catalog and FAQ are
Wait after handoverTime from handover to a staff member's replyWhether your team is ready for handovers
Number of appointmentsMeetings booked into the calendar from chatThe outcome closest to a sale

A high handover rate is not bad in itself. What you really need to look at is the topics of the chats that were handed over. If the same question is handed over again and again, it is time to add its answer to the FAQ.

Alongside the numbers, read about ten randomly selected chats from start to finish once a week. A made-up piece of information, unnecessary pushiness or a missed sales opportunity does not show up in the tables; it shows up immediately in the chat transcript.

How does it work in Collobr?

Collobr's AI-powered live chat is added to your website with a single line of code. The assistant answers visitors in their own language, recommends products from your catalog, creates a lead and notifies you as soon as contact details are given, suggests available times, adds the appointment to your calendar and sends an invitation. WhatsApp messages land in the same inbox too; you can take over the conversation whenever you want.

This article is for general information only and is not legal advice.

If you would like to try the assistant with your own catalog, no credit card is required for the 14-day free trial. Get started now.

Sources

Related articles

All articles →

Stay informed

We’ll email you when new articles are published. At most one email a month.