Why the chatbot on your website doesn't bring appointments (and what to use instead of endless FAQs)
The bot that says “hi, how can I help you?” and kills the lead
The practice redid its website last year. Among the new features, a chatbot in the bottom right corner: a bubble that opens after three seconds with “Hi! 👋 I’m the virtual assistant. How can I help you?”. The provider had shown promising statistics: hundreds of conversations a month. After a year, the owner does the maths and notices something: of those hundreds of conversations, the appointments booked were a handful. Phone calls and form requests haven’t gone up. Some clients have even complained: “I tried asking your bot, it answered things that had nothing to do with it, in the end I called”.
If the chatbot on your website doesn’t convert, you’re not an isolated case. It’s the most common pattern I see on professionals’ and SMEs’ websites: a tool that generates “conversations” and almost no useful contacts. The reason is rarely the technology. It’s that the bot was put there to do something vague (“help”) instead of something precise (bring the right person to book an appointment, or to request a quote). And the question “how can I help you?” is the worst possible opening: it forces visitors to explain their problem in writing, to a machine, before knowing whether it’s worth it.
This piece is for anyone with a website with a chatbot and zero calls, or thinking of adding one. Let’s look at what visitors really want, why FAQs, booking and quotes are different jobs, which numbers to watch, how to build a four-question flow that leads to the calendar, when WhatsApp, a form or the phone beat any widget, and how to use artificial intelligence without letting it do damage. At the end, a 14-day test to find out what works in your case, without trusting my opinions or the provider’s.
Is this for you?
It is for you if:
- you’re a professional (medical, legal, technical, sales practice), a service centre, an appointment-based business or an SME selling services on quotation;
- you have a chatbot or chat widget on your website, and the useful contacts it brings are few or you don’t measure them;
- you’re considering adding one because “everyone has one” or because a provider pitched it to you with artificial intelligence.
It is not for you if your site is an e-commerce store with thousands of visits a day and a structured customer service: there, support chat follows different logic (order status, returns, tracking). And it is not for you if you’re looking for a comparison of chatbot platforms: the problem, in most cases, isn’t which platform, it’s what you make it do.
What visitors to a professional’s or SME’s site want
Before deciding what to put on the site, it’s worth asking who arrives and what they want. On professionals’ and service SMEs’ sites, people almost always arrive with one of these four intentions:
- “Do you do what I need?” They have a problem and want to understand whether you’re the right person: the service, the area, the specialisation.
- “Roughly how much does it cost?” They want an order of magnitude before committing. Not necessarily the exact price, but whether it’s within reach.
- “When can you do it?” They want to know whether there’s availability in a reasonable time, and possibly to book.
- “Can I trust you?” They look for signs of reliability: who you are, experience, reviews, where you are, how you work.
Note what is not on this list: “I want to chat with a virtual assistant”. Nobody arrives on an accountant’s or physiotherapist’s site wanting to converse. They arrive with a practical question and little patience. If the answer is one click away, they stay; if they have to write it, explain it and wait, many leave, and some go to the competitor with a clearly visible “Book” button.
Then there’s a behavioural fact anyone who’s watched session recordings knows: most people don’t read. They scroll, look for a phone number, a price, a button. A bubble that opens by itself and covers half the phone screen is, for many, an obstacle to close, not help.
FAQ vs booking vs quote: three different jobs
The underlying mistake of many chatbots is trying to do three different jobs with the same tool, and doing all of them badly.
The FAQ job: answering recurring questions. “Are you open on Saturdays?”, “Do you accept my insurance?”, “Where can I park?”. These are questions with short, fixed answers. The best place for them isn’t a chatbot: it’s a clear page, well organised, findable by Google, and short answers in the right spots on the site (opening hours next to the address, agreements on the services page). A chatbot answering FAQs is a slower FAQ page.
The booking job: setting an appointment. You need to know who the person is, what they need (to direct them to the right professional or service), and when they can come. The best place is a short flow to a calendar, with very few questions and immediate or quick confirmation.
The quote job: collecting information for a proposal. You need to understand the problem in some detail (sizes, quantities, timing, budget), often with an attachment (a photo, a document). The best place is a guided form, or a conversation with a person, followed by an answer within a set time.
| Job | What the person wants | Best tool | The generic chatbot… |
|---|---|---|---|
| FAQ | a short answer, now | clear page + answers in the right spots | …makes them wait and sometimes answers wrongly |
| Booking | to know if there’s a slot and book it | 3–4-question flow + calendar | …makes them write the problem before knowing if there’s a slot |
| Quote | an order of magnitude and a proposal | guided form or human contact, answer within a set time | …collects scattered information nobody reads |
When a bot tries to do all three jobs, it ends up asking “how can I help you?”, because it doesn’t know which of the three you need. It’s exactly the question that makes people leave.
Metrics: chat started ≠ appointment in the calendar
The chatbot provider usually shows numbers like “conversations started” or “messages exchanged”. They make the tool look useful, but say nothing about what you care about. A “conversation” can be someone who closed the bubble by typing “no thanks”, or someone who asked for opening hours and left.
The metrics that matter are those that end up in a calendar or a quote. Here’s how to set up the funnel, i.e. the sequence of steps from visitor to appointment.
The funnel, in a table
| Step | What you measure | Example (practice with 2,000 visits/month) | Question to ask |
|---|---|---|---|
| Site visits | people arriving | 2,000 | Are the right people arriving (area, service)? |
| Service page visits | people interested in a service | 700 | Do service pages answer the 4 questions? |
| Contact action started | clicks on “Book”, phone, WhatsApp, form, chat | 120 | Is contact visible and easy on a phone? |
| Contact completed | booking sent, call made, form submitted | 60 | Where are people lost between start and completion? |
| Appointment confirmed | in the calendar, with confirmation | 40 | How long between request and confirmation? |
| Appointment attended | the person shows up | 34 | How many don’t show? Do you send a reminder? |
(The numbers are an example to show the structure, not a benchmark.)
Three things to note:
- chat is only one of the contact channels, and must be compared with the others at the same step: how many completed bookings come from chat, from the calendar button, from the phone, from WhatsApp;
- the critical step is often between “contact completed” and “appointment confirmed”: requests that arrive and stay unanswered for a day or two. The best chatbot in the world is useless if nobody calls back. It’s the problem described in the piece on leads that die;
- the last step, the appointment attended, depends on reminders and confirmations: an appointment booked and missed is a cost, not a success.
Measuring all this doesn’t require a complicated tool: site events (button clicks, form submissions) in a privacy-friendly analytics tool are enough, plus a weekly count of appointments by source channel, which reception can note in a spreadsheet. If you don’t ask for the channel, you’ll never know: “how did you hear about us / where did you contact us from?” must be recorded.
Flows that work: 4 questions, slot, human confirmation
When the person is ready to book, the best flow is short, guided, and ends with a time. No free-text fields at the start, no “describe your problem”, no questions you don’t need in order to book.
4-question script
Here’s the scheme I use as a starting point, to adapt to your work. It works as buttons in a widget, as a stepped form on the site or as a guided message on WhatsApp.
Question 1 — What do you need?
[ First visit ] [ Check-up / follow-up visit ] [ Request a quote ] [ Other ]
Question 2 — Where would you prefer? (only if you have several locations or online services)
[ Bologna office ] [ Modena office ] [ Video call ]
Question 3 — When suits you?
Show 3–6 free slots in the next few days + [ See more times ]
Question 4 — What's your name and how can we reach you?
Name · Phone · Email (optional) · [ ] I've read the privacy notice
→ Confirmation: "Thanks Marta, we've reserved Thursday the 21st at 10:30 in Bologna.
We'll confirm within 2 working hours on 3xx… If it doesn't suit you, reply to this message."
Line by line, the why:
- Question 1 sorts: first visits and check-ups go to the calendar, quotes to a dedicated form, “Other” to a person. Four choices at most, written as the client would say them, not with the service’s internal name.
- Question 2 is asked only if needed. Every extra question loses people.
- Question 3 shows real times, taken from the calendar. It’s the moment the person decides, and a concrete time (“Thursday at 10:30”) converts much better than “we’ll get back to you to book”.
- Question 4 asks for contact details at the end, when the person has already chosen a time and has a reason to leave them. Asking at the start, before giving anything in return, is one of the most common mistakes.
- The confirmation says what happens now and by when, with a time you’re able to keep.
Human confirmation
In professional services, it often pays for the appointment to be reserved by the flow and confirmed by a person within a short time (two working hours, for example). Human confirmation serves to check that the chosen service is the right one, to ask for missing information, to prevent fake appointments. And it has a useful side effect: a personal confirmation call or message reduces no-shows.
The calendar side — handling several professionals, several locations, gaps between appointments, services of different lengths — is a topic of its own, and I cover it in the piece on the multi-operator professional calendar. Here one principle is enough: the flow must read real availability, not a fake calendar that reception then has to fix by hand.
And the quote?
For quotes the scheme is similar, but question 3 becomes “the information we need” (three or four fields specific to your work: sizes, quantities, timing, a photo) and the confirmation says when the answer will arrive: “We’ll send you a proposal by Wednesday”. The promised time is part of the offer: a quote arriving in 24 hours often beats a cheaper one arriving in a week.
WhatsApp, form, phone: when they beat the widget
The chat widget isn’t the only way to be contacted, and often not the best. It depends on who your clients are and how you work.
WhatsApp wins when:
- your clients are individuals, not companies, and already use it for everything;
- the conversation can continue over time (confirmations, reminders, questions after the appointment);
- someone at the practice actually answers, at declared times.
Watch two things: use a business number (with the business app or official business services), not someone’s personal mobile, and remember that privacy rules apply on WhatsApp too, especially for health data.
A well-made form wins when:
- you need structured information (the quote);
- you want the request to arrive complete, without back-and-forth;
- you want precise channel tracking.
The phone wins when:
- clients are older or the service is delicate (health, legal problems, emergencies);
- the value of each contact is high and a human voice makes the difference;
- someone answers. A number that rings out is worse than no number.
The chat widget makes sense when:
- a person answers in real time during declared hours (then it’s a channel like WhatsApp, but on the site);
- or it’s a guided button flow like the one above, not a free conversation.
On most professionals’ sites, the combination that works best is: an always-visible “Book” button (opening the four-question flow), a phone number clickable on mobile, WhatsApp if someone answers, a form for quotes. The bubble that opens by itself can, in most cases, be removed.
If you use AI: boundaries, fallback, hours
I’m not against artificial intelligence on the site: I’m against using it to replace a button. There are cases where an assistant based on a language model really helps, for example when services are many and people don’t know which to choose, or when pre-booking questions are many and varied. But you need precise rules, because an AI assistant on the site speaks on your behalf.
Boundaries: what it can and can’t do.
- It can answer about services, hours, locations, indicative prices taken from content you wrote and approved, not from what the model “knows”.
- It can help choose the service and lead to the booking flow.
- It must not give professional opinions (diagnoses, legal or tax advice, technical assessments), not even “indicative” ones.
- It must not promise prices, discounts, timing or availability that don’t come from your calendar or price list.
- It must not collect sensitive data (health data, for example) in free conversation.
Fallback: what it does when it doesn’t know. The rule is simple: when the question goes beyond the boundaries or the assistant isn’t sure, the answer is a hand-over to a person, not an attempt. “On this it’s better for the doctor to answer you directly: would you like to book a first visit or be called back?” is a good answer. Inventing a plausible answer is the worst, because the client takes it as true and attributes it to you.
Hours: what it promises and when. If nobody answers outside office hours, the assistant must say so: “We’ll call you back tomorrow morning from 9”. An assistant giving the impression of a 24-hour human presence, when nobody’s behind it, creates expectations you then disappoint.
And a note on data: an AI assistant on the site processes the data of whoever writes. You must choose a provider that processes it in Europe with an adequate contract, or a solution under your control, and write a clear privacy notice. The picture on when a language model is really needed and when a chatbot is just a cost is in the piece on LLM product vs chatbot.
Bot copy mistakes
Many chatbots fail not because of the technology, but because of how they’re written. The mistakes I see most often:
- “Hi! How can I help you?” — offloads the work of explaining onto the person. Better: “Would you like to book a visit or do you have a question?” with two buttons.
- The bubble that opens by itself after three seconds, before the person has read anything. On mobile it covers the content. Better: a visible button the person opens when they want.
- The fake name and stock photo (“I’m Giulia, your assistant!”) when it’s a bot. People notice, and feel fooled. Better to say clearly that it’s an automated assistant and who’s behind it.
- Asking for name, email and phone at the start, before giving any answer. It’s a wall. Details are asked at the end, when there’s a reason.
- Long, generic answers (“Our practice offers a wide range of personalised services…”). Whoever asks for opening hours wants opening hours.
- No way out to a person. If there’s no obvious way to talk to someone, the frustrated person closes the site.
- Tone too informal for the sector. Emojis and “Hey!” are fine for some businesses, much less for a law firm or medical practice.
A poor but honest A/B test in 14 days
You don’t need an expensive tool to understand what works on your site. You need an honest comparison, measured on the right things, for long enough. Here’s a test you can do in two weeks with what you have.
Preparation (day 0):
- Make sure you measure, per channel, completed contacts and confirmed appointments (even with a spreadsheet filled in by reception: “where did this appointment come from?”).
- Note the current situation for the previous two weeks: visits, completed contacts, appointments, per channel.
Week 1 — Current version: leave the site as it is, chatbot included, and measure.
Week 2 — New version: remove the automatic bubble and replace it with an always-visible “Book” button opening the four-question flow, plus clickable phone and, if someone answers, WhatsApp. Measure the same things.
Comparison (day 14):
- compare confirmed appointments per 100 visits in the two weeks, not conversations;
- look at which channel appointments come from in the new version;
- ask reception whether the quality of requests changed (more complete? more relevant?).
Two caveats to be honest with yourself:
- two weeks is little if visits are few: with 300 visits a week and a handful of appointments, a difference of two or three may be chance. If numbers are small, repeat the comparison for another two weeks, or run it longer;
- watch out for unusual weeks: holidays, ad campaigns, seasonality. Compare similar weeks, or alternate the versions over several cycles.
If the new version brings more appointments, keep it. If nothing changes, you’ve still removed a cost (the chatbot subscription) without losing anything. If the old version wins — it happens, for example, when behind the chat there’s a good person answering in real time — you have proof your chatbot works, and can invest in it with more conviction.
How much it costs: estimates
Indicative estimates for a professional or a service SME.
| Item | Estimate | Notes |
|---|---|---|
| Generic chatbot on subscription | €30–300/month | Often the cost you remove |
| AI assistant with approved content and hand-over to a person | €100–500/month + setup | Makes sense with many services and many pre-booking questions |
| 4-question flow with online calendar | €0–60/month for the tool + 1–3 days of configuration | Many scheduling tools include it |
| Guided quote form | 1–2 days to build | One-off |
| Business WhatsApp with quick replies | free or a few €/month | The real cost is the time of whoever answers |
| Per-channel measurement | a few hours of configuration | Plus 10 minutes a week of reception’s time |
The comparison that matters isn’t between tool costs, but between cost and the value of an appointment. If a first visit is worth €80 and on average brings a client worth €600 a year, even two extra appointments a month repay any line in this table.
3 beginner mistakes that cost clients
1. Measuring conversations instead of appointments. It’s the mistake everything else stems from. As long as the number on the dashboard is “conversations started”, the chatbot will always look useful. Always ask: how many calendar appointments came from here?
2. Adding AI before having a calendar. A smart assistant that ends by saying “we’ll get back to you to book” has wasted all its value at the decisive moment. First the flow to real times, then, if needed, the intelligence.
3. Collecting contacts nobody calls back. The chatbot collects name and phone, sends them to an inbox nobody checks, and the person waits. After a day they’ve already booked elsewhere. Every contact channel must have a responsible person and a declared response time.
30/60/90-day path
First 30 days — measure and simplify
- set up per-channel measurement (site events + reception’s log);
- write answers to frequent questions on a clear page and in the right spots on the site;
- configure the four-question flow connected to the real calendar;
- run the 14-day test.
Days 30 to 60 — consolidate
- keep the version that won the test;
- define who answers each channel and in how long (including outside hours: an honest automatic message);
- add automatic reminders before appointments, to reduce no-shows.
Days 60 to 90 — evaluate AI, if needed
- if clients’ questions show that many don’t know which service to choose, consider an AI assistant with approved content, written boundaries and hand-over to a person;
- measure it with the same yardstick: confirmed appointments per 100 visits, not conversations.
When NOT to do it
- If you have few visits, the problem isn’t the chatbot but traffic: optimising the step from 50 visits to 1 appointment is of little use. People need to arrive first.
- If nobody can answer or confirm in reasonable time, no flow will save the contacts. Organisation first, then the tool.
- If your work isn’t appointment-based (selling products, ongoing services without booking), the four-question flow needs rethinking: it’s a different job.
- If the current chat works because a person answers in real time and the numbers prove it: don’t touch it. At most, add the calendar button.
- If a provider pitches an AI assistant “that does everything” without talking about calendar, boundaries and hand-over to a person, now isn’t the time.
Practical checklist
- I know how many confirmed appointments come each week from each channel.
- Answers to frequent questions are on a clear page and in the right spots on the site.
- The “Book” button is visible on every page, including on mobile.
- The booking flow has at most four questions and shows real times.
- Name and contacts are asked at the end, not at the start.
- The confirmation says what happens now and by when.
- Every channel has a responsible person and a declared response time.
- No bubble that opens by itself and covers the content on mobile.
- If there’s an AI assistant: approved content, written boundaries, hand-over to a person, honest hours, data processed in the EU.
- 14-day test done, comparing appointments per 100 visits.
The verdict
If your website chatbot doesn’t convert, the problem is almost never the technology: it’s that you asked it to “help” instead of leading people to an appointment. “Hi, how can I help you?” makes visitors work, has them write a problem before knowing whether there’s a slot, and measures success in conversations that never end up in a calendar.
In its place, on most professionals’ and SMEs’ sites, something much simpler works: frequent answers written well on the site, an always-visible “Book” button, four questions ending with a real time, a person confirming within a set time, and the channels your clients already use, from the phone to WhatsApp. Artificial intelligence can add value later, with clear boundaries and a way out to a person. But the calendar comes first, and a two-week test will tell you, with your own numbers, whether I’m right.
If you want to review the path from site to appointment for your business, you’ll find the projects I’ve worked on in the portfolio and my background in the biography. The other pieces on setting up useful automation in an SME are in the guide on SME automation: where to start.
FAQ
Why doesn’t the chatbot on my website bring appointments?
In most cases because it was designed to “help” generically instead of leading to a booking. It opens with “how can I help you?”, forces visitors to write their problem, asks for details before giving anything in return and doesn’t end with a calendar slot. Often, too, success is measured in conversations instead of appointments.
Does a professional’s website really need a chatbot?
Rarely in its generic form. People look for quick answers on services, indicative costs and availability, and a simple way to book. A clear page, a visible “Book” button and a short flow to the calendar almost always work better. A chat makes sense if a person answers in real time or if it’s a guided button flow.
Which metrics should be used to evaluate a chatbot?
Confirmed appointments, or quote requests, per 100 site visits, compared with the other contact channels. Conversations started and messages exchanged say nothing about the result. It pays to follow the whole path: visits, service pages, contact started, contact completed, appointment confirmed, appointment attended.
How many questions should a booking flow have?
As few as possible, usually four: what you need, where (only if there are several locations or online services), when (showing real times), and name and contact at the end. Every extra question loses some of the people, and asking for details at the start is one of the most common mistakes.
Better WhatsApp or the chatbot on the site?
It depends on who the clients are and who answers. WhatsApp works very well with private clients and when someone answers at declared times, including for confirmations and reminders. It must be used with a business number and in line with privacy rules. The chatbot on the site makes sense if it’s a guided flow to the calendar or if a person answers in real time.
Can artificial intelligence be used to answer clients on the site?
Yes, with precise rules: answers based on approved content, a ban on giving professional opinions or promising prices and availability not taken from the calendar, hand-over to a person when the question goes beyond the boundaries, honestly declared hours and data processed by a European provider or a solution under your control.
What should the chatbot’s opening message say?
A question offering concrete choices instead of “how can I help you?”: for example “Would you like to book a visit or do you have a question?” with two buttons. Avoid the bubble that opens by itself, fake names and stock photos for an automated assistant, and long, generic answers.
How do you find out whether removing the chatbot loses contacts?
With a two-week test: one week with the current site, one with the “Book” button and the four-question flow in place of the bubble, measuring confirmed appointments per 100 visits and the source channel. With few visits it pays to repeat the comparison several times and avoid unusual weeks.
How much does it cost to replace the chatbot with a booking flow?
Many online scheduling tools include booking with real times for a few tens of euros a month or less. Configuration takes one to three days of work. Often the overall cost is lower than the subscription of the generic chatbot it replaces.
Why do contact requests arrive but few appointments?
Often because requests aren’t handled in time: they arrive in an inbox nobody checks, or are called back after a day or two, when the person has already booked elsewhere. Every contact channel must have a responsible person and a declared response time, and the flow should propose a concrete time instead of “we’ll get back to you”.
Antonio Trento — System Architect & AI Integrator
Is this your problem?
I design and build data, backend, interface and AI agents end-to-end. No slides: systems that run and stay yours.