Insights · globalhr.ro
How AI assistants handle website inquiries, and when a human should take over
In short: An AI assistant on a business website answers routine questions from approved content, collects the details your team needs, and routes each inquiry by type. It should hand off to a person when it has no approved answer, when the visitor asks for someone, or when the topic involves money, complaints, advice or sensitive personal data.
An AI assistant for website inquiries is a chat tool that reads a visitor's question, replies using content the business has approved, and passes the conversation to a staff member when it should not answer alone. Done well, it shortens the time between "I have a question" and "someone useful is working on it." Done badly, it traps visitors in a loop or gives confident wrong answers. The difference is almost entirely in how the handoff is designed.
What the assistant actually does
Behind the chat window, a well-configured assistant runs four jobs in sequence. Understanding them helps you decide what to automate and what to leave to people.
- Classify the request. Is this a general question, a new inquiry, an existing client, a vendor, or a complaint? Classification decides everything that follows.
- Answer from approved sources. The assistant searches your service pages, FAQ and policies, and replies only with what those sources support. It should not improvise prices, deadlines or promises.
- Collect structured details. For a new inquiry, it asks for the fields your team needs: name, contact details, service, location, timing, and anything specific to your business.
- Route and hand off. It sends a summary to the right person or queue, tells the visitor what happens next, and stops.
Notice what is missing: deciding, advising, negotiating. Those stay with your team.
The handoff rules that matter most
A handoff is the moment the assistant stops and a person takes over. Define the triggers before you switch anything on. The table below is a reasonable starting point for most service businesses; adjust it to your own risks.
| Trigger | Why it matters | What the assistant does |
|---|---|---|
| Visitor asks for a person | Ignoring the request erodes trust quickly | Confirms contact details and hands off at once |
| No approved answer found | This is where made-up answers come from | Says it will check with the team and collects details |
| Pricing beyond published ranges | Quotes create expectations and sometimes obligations | Records the scope and passes to whoever quotes |
| Complaint or frustration | Needs judgment and tone a script cannot supply | Acknowledges, records, escalates immediately |
| Advice requests (legal, tax, medical, financial) | Liability and professional rules | Declines to advise, offers a consultation |
| Sensitive personal data volunteered | Chat is the wrong place for it | Asks the visitor not to share it there, points to a secure channel |
| Repeated misunderstanding | Two failed attempts usually means a third will fail too | Offers a person instead of rephrasing again |
A good rule of thumb: if a wrong answer would cost you money, a client or your reputation, the assistant should collect details and hand off rather than answer.
Designing the handoff itself
Deciding when to hand off is half the work. The other half is making sure the handoff does not drop the ball.
Pass a summary, not a transcript
Your staff member should receive a short structured note: who the visitor is, what they asked, what the assistant already told them, and what is still unknown. Nobody wants to scroll through a chat log to find the phone number.
Tell the visitor exactly what happens next
"Someone will be in touch" is vague. "Your request has been sent to our scheduling team, who reply by email during business hours" sets an honest expectation. Never promise a response time your team cannot keep.
Route by type, not to one inbox
If every handoff lands in the same shared inbox, you have only moved the bottleneck. For example, a property management office might send maintenance requests to the maintenance coordinator, owner inquiries to the account manager, and rental applications to leasing.
Cover out-of-hours
Decide what the assistant says at 11 p.m. on a Saturday. Usually that means collecting details, confirming they were received, and stating when the office reopens. If you handle emergencies, give the emergency route clearly and do not rely on chat for it.
Preparing your content so answers are correct
An assistant is only as accurate as the material it draws on. Before launch, review these sources:
- Service pages that state what you do and do not do, in plain language
- An FAQ that answers the ten questions your front desk hears most
- Service areas, hours, and any eligibility rules
- Which documents a new client needs to bring or send
- Published prices or ranges, if you are comfortable sharing them
- A short list of topics the assistant must never discuss
If your website does not cover these, fix the website first. The same content helps human visitors and search engines, so the effort is not wasted.
Trade-offs to weigh honestly
An assistant reduces repetitive work for your team and gives visitors an immediate first reply at any hour. It also adds something new to monitor. Language models can misread a question or phrase an answer awkwardly, so plan to review conversations weekly at first and update your content when you spot gaps. Some visitors simply prefer a phone call or an email form; keep those options visible rather than forcing everyone through chat.
There is also a data question. Decide where conversations are stored, who can read them, and how long they are kept. If you work with regulated information, such as health or tax records, take advice on your obligations before letting any automated tool collect it, and design the assistant to redirect that information to a secure channel such as a client portal.
A sensible rollout
- List the request types you receive and who handles each.
- Write the handoff rules using the table above as a template.
- Run the assistant on a test page and ask it the awkward questions your staff actually get.
- Launch on the website first; add email or WhatsApp once the rules are stable.
- Review a sample of conversations every week and update content or rules.
At globalhr.ro we configure AI assistants for websites, email and WhatsApp with the handoff rules agreed in writing before launch. If you would like a proposal, describe your most common requests and we will reply by email within one business day.
Questions and answers
Can an AI assistant replace my receptionist?
It can take over repetitive first replies and information gathering, which frees your receptionist for calls, visitors and judgment work. It is not a replacement for the person who handles complaints, exceptions and relationships.
How do I stop the assistant from inventing answers?
Restrict it to approved content, tell it to hand off whenever that content does not cover a question, and test it with difficult questions before launch. Review conversations regularly and close gaps in your content.
Should the assistant say it is an AI?
Yes. Visitors should know they are talking to an automated assistant and how to reach a person. Being clear about this avoids confusion and is increasingly expected by regulators and clients alike.