How a Website Chatbot Works for Your Business: One 9:47 PM Enquiry, Start to Finish

What happens when a visitor opens your website chat: one 9:47 PM enquiry, start to finish

It's 9:47 PM on a Tuesday. Sara is planning her parents' 40th anniversary: a dinner for about 60 people on a Saturday in March, with rooms for the relatives flying in. She has three hotel websites open in three tabs.

The events team at all three hotels went home at six.

Most articles about website chatbots stop at "it answers questions". This one follows Sara's enquiry from her first message to the moment someone at the hotel picks up the phone the next morning, so you can see each step: what she sees, what the hotel sees, and what the hotel set up beforehand.

The hotel is an example. Further down, you'll see the same chat at a dental practice, a plumbing company and an online store.

Step 1: Sara opens the chat

She clicks the chat button in the corner of the page. It's on the side the hotel chose, so it doesn't cover the "Book now" button.

The assistant greets her with the hotel's own welcome message, in the hotel's own voice. This hotel chose a warm, welcoming personality, short-to-medium replies and no emoji. A law firm or a dental practice would choose differently.

Sara types:

Hi, do you do private dinners for around 60 people? It'd be a Saturday in March.

Step 2: It answers from the hotel's own information

The assistant replies with what the hotel actually offers: the private dining room that seats up to 70, the set menus on the events page, and the line from the hotel's events brochure about minimum spend on Saturdays.

That's the most important thing to understand about how a website assistant like this works. It answers from your own material: your website, the documents you upload, and anything you add by hand. It doesn't make up a price or a policy. When the answer isn't in your material, it says so and offers to have the team follow up, rather than guessing.

Sara asks whether there's parking and whether rooms come at a group rate. Both answers are on the hotel's site, so she gets them in seconds.

Then she asks something the hotel's material doesn't cover: "Can we bring our own cake?" The assistant says it will check with the events team and makes sure that question is in the notes they'll see in the morning.

If Sara had written in Arabic or Spanish, it would have replied in Arabic or Spanish. The hotel picks which languages its assistant uses, up to 15.

Step 3: It asks what the events team would ask

A good receptionist doesn't just answer. They ask the two or three questions that let the right person call back prepared.

The assistant does the same. Because the hotel chose Hospitality & Hotels as its industry, it knows what matters for an event enquiry and asks naturally, one question at a time:

• Which Saturday, and is the date flexible?
• Roughly how many guests, and how many rooms?
• Any dietary needs worth flagging early?

It doesn't interrogate. If Sara only wants to browse, it lets her browse.

What it doesn't do: confirm the booking. It doesn't promise a date or a final price. That's the events team's job, and the assistant says so plainly: "I've noted Saturday 14 March for around 60 guests and 12 rooms. The events team will confirm availability with you tomorrow."

Step 4: It takes her details during the conversation, not after

Sara mentions her name early on. Later she gives her email so the team can send the menu pack.

Both are saved the moment she gives them, not at the end. Sara's phone buzzes and she closes the tab halfway through a sentence, but the hotel still has her name, her email and everything she asked.

That's the practical difference between this and a contact form. A form gets nothing until someone presses "Submit". A conversation leaves something behind at every step.

Step 5: The next morning, the hotel already knows who to call first

Here's what the hotel's team finds before nine.

An email for every new lead. Team members who've opted in get an email with Sara's contact details and the full conversation, so they can read exactly what she asked, word for word.

A scored lead, not a transcript. In the dashboard, Sara's enquiry is a lead with:

• a tier: Hot, Warm, Nurture or Cold. Sara, with a date, a headcount and 12 rooms, is Hot;
• a short summary of what she wants;
• the details that matter for this industry, such as date, party size and rooms, pulled out so nobody has to re-read the chat to find them;
• the full transcript underneath, including the cake question.

Already assigned to someone. The lead goes automatically to a member of the events team. If nobody picks it up within the time window the hotel set, it moves to someone else, so a Hot lead doesn't sit waiting while the person it went to is on leave.

On the pipeline board. Sara's lead sits in the first column of a five-stage board. The team drags it along as it progresses, adds notes, and moves it to Won when the deposit is paid.

With the campaign that brought her. Sara came from an ad the hotel was running for spring events, and the lead says so. Over a month, the hotel can see how many leads and how much revenue each campaign produced, and what each lead cost.

Step 6: Over the following weeks, it tells you what to fix

After a few weeks of conversations, the dashboard starts making suggestions based on what's actually happening. For example:

• Visitors keep asking whether they can bring their own cake, and your website doesn't say. (Add a line to the events page, and the assistant can answer it from then on.)
• Several leads have been sitting in New for days. (Someone should follow them up.)

The questions your assistant can't answer turn out to be one of the most useful things it produces. They're your website's content gaps, listed by the people who actually hit them.

The same chat, in three other businesses

The steps don't change between industries. What changes is what the assistant knows, what it asks, and what it must never say.

Dental practice

Plumbing / HVAC

Online store

10 PM question

"Are you taking new patients, and do you take my insurance?"

"My boiler's making a banging noise. Can someone come this week?"

"Will this arrive before Friday if I order tonight?"

What it asks

New or existing patient, what they need, a good time to call

Postcode or ZIP, how urgent, a photo or description of the problem

Order size, delivery location, wholesale or retail

What it must never do

Diagnose, advise on medication, or say a symptom can wait

Give a fixed price for a job nobody has seen, or tell someone a gas smell is safe

Promise stock or delivery dates the store hasn't published

What lands in the morning

A lead with the treatment, insurance question and urgency

A quote request with the location and urgency, emergencies first

A sales lead, with bulk enquiries separated from everyday questions

We've written longer guides for some of these: what a dental website chatbot should answer and never say, how plumbers, HVAC firms, electricians and roofers lose quote requests, and law firm website intake. For a side-by-side of five industries, see one website assistant, five industries.

What a person still has to do

A website assistant answers when nobody else can, and it hands your team a ranked list of people to call. It doesn't replace the call.

That matters to visitors. In a SurveyMonkey survey of 2,017 US adults (December 2025), 89% said they want the option of a human, and only 15% said they'd had a seamless handoff from AI to a person (SurveyMonkey, December 2025). The assistant's job is to make that handoff the easy part: the right person, with the full context, the next morning.

What you set up to make all of this happen

The whole list:

1. Describe your business and point it at your website. Setup is a guided conversation. It reads your site and drafts your assistant.
2. Add what you already send people. Brochures, menus, price lists, policies, the FAQ someone keeps pasting into emails.
3. Choose how it sounds and what it must never say. Personality, tone, reply length, how much it leads the conversation, emoji, and your own rules in plain words ("never quote a final price", "always mention free parking").
4. Paste one line of code. The install assistant shows you exactly where it goes on WordPress, Shopify, Wix, Squarespace, Webflow and other platforms.
5. Invite your team and decide who gets which leads.

Before you go live, type the ten questions your customers ask most and check every answer yourself. That's good practice with any chatbot, including ours.

Try it

You can create your ConGreeto assistant and have it answering on your site the same day, or book a 30-minute demo and we'll set it up against your own website with you. Plans and prices are on the pricing section.

ConGreeto is designed, built and run by Brillnex Systems, an AI engineering company. You can read the story of why they built ConGreeto, and the engineers have written up what running ConGreeto in production taught them about building AI products.

Frequently asked questions

How does a website chatbot work?

A visitor opens the chat and asks a question. A modern AI website assistant answers from the business's own information (its website, uploaded documents and anything added by hand), asks a few questions to understand what the visitor needs, and saves their contact details as they're given. When the conversation ends, the business gets a lead with a summary, a priority tier and the full transcript.

Does the chatbot make up answers?

A well-set-up assistant answers from your material and says when it doesn't know, offering to have your team follow up instead of guessing. No AI is perfect, so test your most common questions before going live and read the first week of conversations.

Can a website chatbot take bookings?

ConGreeto takes the details of the booking request (date, numbers, what the visitor wants) and passes them to your team, who confirm it. It doesn't confirm bookings or availability itself.

What happens if the visitor leaves halfway through?

Anything they've already shared, such as a name, email or phone number, is saved as they share it. Your team still gets the lead, with the conversation up to the point they left.

Who on my team gets the lead?

Leads are assigned automatically to a team member, and move to someone else if they aren't picked up in time. Opted-in team members also get an email for each new lead with the full conversation.

Do I need a developer to install it?

No. It's one line of code, and the install assistant gives step-by-step instructions for your website platform.

Sources

• SurveyMonkey, Customer service statistics, December 2025; 2,017 US adults, margin of error ±2.5 points.
• The hotel enquiry in this post is an illustration of how the product behaves, not a real customer's conversation.

From us

• One website assistant, five industries
• AI chatbot vs contact form
• How to get more customers from your website: the five leaks

ConGreeto is built by Brillnex Systems, a software and AI engineering company.