Chats and virtual assistants answer questions in real time and can recover undecided visitors. But done badly, they interrupt, frustrate and harm conversion. Their effect depends on timing, tone, quality of answers and the handover to a person when needed.
The essentials in a definition
Chat and assistants on the site can help convert ad visitors by answering doubts on the spot, or get in the way if they interrupt, make mistakes or promise what cannot be delivered. Design matters: when they appear, what they say and how they hand over to a person. Usability research from the Nielsen Norman Group, the GDPR (data in conversations) and the transparency rules of the AI Act (informing when talking to an AI system) are relevant.
When they help
When there are frequent doubts that delay the decision (prices, lead times, compatibility, delivery), when the purchase is complex and when there is traffic outside service hours. A well-trained assistant replies immediately and routes the request.
They also collect real questions, useful for improving pages and ads.
When they get in the way
When they open intrusively right at the entrance, when they give wrong or generic answers, when they trap the person in a loop with no exit, or when they take up the screen on mobile. An assistant that invents information creates legal and trust problems.
On simple purchase pages it may be unnecessary.
Using chat and assistants sensibly
- Open on request or when someone hesitates
- Answer only with verified information
- Hand over easily to a person
- Test with and without the assistant
Good practice
Open only on request or after signs of hesitation, with a useful proposal. Short, verified, current answers. A clear option to talk to a person, with hours and timeframes. Limit the scope to what it knows, and say when it does not know. Say it is an automated assistant and handle data transparently.
Integrate with the CRM so as not to lose the contact.
An assistant that does not know and makes things up is worse than no assistant.
Measure and review
Compare conversion of those who use it and those who do not (carefully, since those who use it are already different). Test with and without the assistant. Review conversations, unanswered questions and satisfaction. Update the knowledge base regularly.
If it does not improve results, simplify or remove it.
Decision table
| Aspect | Good practice | Common mistake |
|---|---|---|
| When it appears | After a sign of interest or on request | Opening immediately over the content |
| Answers | Correct, short and with sources | Inventing prices or deadlines |
| Handover to human | Easy and quick | Dead end |
| Data and transparency | State it is an assistant and how data are processed | Collecting data without informing |
Common mistakes to avoid
- An assistant that invents answers. Limit it to verified content and say when it does not know.
- A chat that blocks the page. On mobile, it interrupts the purchase.
- Not measuring. Compare conversion with and without chat.
How to measure the result
Compare conversion and order value with and without the assistant, resolution rate, handovers to humans and satisfaction.
Limits and exceptions
AI assistants can make mistakes, and the consequences (incorrect prices or deadlines, undue promises) are the company’s. In regulated sectors, risk is higher. They require continuous maintenance and review of conversations and content. If the team lacks supervision capacity, a simpler assistant, with fixed answers and human contact, may be safer.
Illustrative example
Illustrative example: a shop enables an assistant that answers only from the catalogue and published policies, with the option to talk to a person. It only appears after 30 seconds on a product page or when the visitor opens it. Conversion rises, support questions fall, and the shop reviews conversations to improve pages.
Frequently asked questions
Does a chat always raise conversion?
No. It depends on quality and timing. Test.
Should I say it is automated?
Yes, for transparency and trust.
How do I avoid wrong answers?
With a verified knowledge base and clear limits.
Should I say it is an AI?
Yes, transparency is good practice and, in certain contexts, an obligation; confirm applicable rules.
What data may the assistant store?
Only what is necessary and with a legal basis; avoid asking for sensitive data and inform in the privacy policy.
A site assistant helps when it is transparent, limited to verified content, easy to bypass and with quick handover to a person; always measure the effect on conversion.
Sources and recommended reading
- Nielsen Norman Group: usability articles
- General Data Protection Regulation (GDPR)
- Artificial Intelligence Act (AI Act)
Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.