A chatbot on your business website? In most cases, no. Here are the three where the answer is yes.

I get the question almost every week, usually after a sales call: "Someone's offering me an AI chatbot for the website — is it worth it?" The pitch is always the same: more bookings, fewer phone calls, an answer at any hour. The studies say something else. 62% of French consumers prefer a fully human exchange, and only 6% would accept being handled by a chatbot alone. That is not a reason to reject the tool on principle. It is a reason to reserve it for the three situations where it genuinely helps, and to solve the rest another way — often for free.

AI chatbots: what vendors promise, what customers say

Almost every chatbot statistic in circulation comes from someone selling chatbots: "+40% bookings", "80% of questions handled automatically". I am not saying they are false. I am saying they are unverifiable, and that they rarely measure what matters to you — what the customer who lands on the thing actually feels.

On that point, we have independent, recent data. In September 2025, Tersea had Ipsos BVA survey a representative sample of 1,000 French adults aged 18 to 75 about what they will no longer put up with in customer service. The result is worth reading in full, because it describes a badly installed chatbot with some precision.

Read the three irritants together: customers hate the reply that misses the point, hate waiting, and hate not being able to reach anyone. A well-designed chatbot attacks the second problem. A badly designed one creates the first and the third. The Observatoire des services clients 2025 (Ipsos, 5,000 European consumers) confirms the order of magnitude: 90% of French consumers would rather wait for a human adviser than talk immediately to a virtual agent, 31% already use chatbots and only half of them say they are satisfied.

There is an important nuance, and it works in your favour. These studies cover customer service at banks, telecoms and insurers — situations where the person is calling because something has gone wrong. A local business mostly gets pre-purchase questions: are you open, do you have, do you deliver. Those are simple, factual questions, exactly the kind consumers are willing to ask a machine. So the problem is not the AI. The problem is that nobody has yet looked at what you are actually being asked.

Before installing a chatbot: what are people asking you, and why?

Try it for a week. Write down every question that comes in by phone, Instagram message, WhatsApp, at the counter. At most of the businesses I work with, the list fits in five lines.

• "Are you open today?" / "Until what time?"
• "Do you deliver?" / "Where exactly are you?"
• "How much for…?"
• "Do you still have model X / today's special / a table tonight?"
• "Can we book for eight? With a pushchair? Gluten-free?"

Now the question that changes everything: why did the person have to ask? In the first three cases the answer is nearly always the same: because the information was not on your website, or could not be found in under ten seconds on a phone. Opening hours in an image inside an expired story. The menu as a PDF. "Price on request". The delivery area nowhere.

A chatbot that answers "we're open until 7 pm" to that question does not fix the problem, it papers over it. And it papers over it with a risk: the day you close at 5 pm for stocktaking and nobody told the robot, you have just manufactured irritant number one. A website that answers before the question beats a robot that answers after it, and it never falls behind on updates, because it is the same page you already edit.

The three cases where a chatbot earns its place

What is left is what the website cannot solve alone: questions whose answer changes all the time, and requests that end in an action. That is where — and only where — an automated conversation earns its place. I see three cases.

1. Booking or ordering out of hours, connected to a real system

A customer looking for a table at 11 pm or a cake for Saturday does not want to chat; they want a slot and a confirmation. In most cases, an online booking or ordering module does that better than a chatbot, with no conversation at all. The chatbot becomes useful for what the module cannot take: a party of twelve, a private hire, a severe allergy, a custom cake. It collects the request cleanly — date, headcount, constraint — and hands it to a human with a stated response time. The non-negotiable condition: it is connected to your real schedule. A robot that "takes note" of a booking you never see in your diary is a promise you will not keep.

2. A wide catalogue, with availability or compatibility questions

A bike shop, a hardware store, a spare-parts dealer, a wine shop with four hundred references. Here the questions are not "are you open?" but "do you have this inner tube in 700×25?", "does this filter fit my cooker hood?", "what light reds do you have under fifteen euros?". No FAQ page can cover that. An assistant plugged into your stock and product sheets can, and this is the one case where generative AI brings something nothing replaced before. The condition: the stock is current and the robot knows how to say "I don't know". An assistant that invents a compatibility sends the customer back with the wrong part, and you are the one handling the return.

3. Multi-service appointment booking

Salons, beauty institutes, garages, repair workshops, practices. The appointment depends on three variables that intersect: the service, its duration, the person available. "A colour and a blow-dry Saturday morning with Léa" is a sentence that conventional online diaries turn into four screens of drop-down menus. A conversational assistant connected to that diary understands the sentence, offers two slots, confirms. Same condition as case 1: connected to the real diary, or nothing.

You will notice what the three cases have in common: the chatbot is never "the assistant that answers everything". It is a conversational interface laid over live data — a schedule, a stock list, a diary. Without that data behind it, the conversation has nothing to say that your website doesn't, and it says it less well.

What this does not mean

It does not mean a general-purpose "ask me anything" chatbot, connected to nothing but a paragraph of company blurb, has any value. That is the model you get pitched most, because it is the easiest to install, and it is the one that produces the 58% of off-topic replies. If you install one anyway, three rules are not up for discussion.

1. It says it is an AI, from the first line. That is no longer a courtesy: since 2 August 2026, Article 50 of the EU AI Act requires it. And your customers care: 72% of French consumers say they would feel deceived if they were not told.
2. A human is reachable at any time, one visible click away. Not after three attempts. A "talk to someone" button always on screen, with the channel and the delay: "reply by WhatsApp within two hours during the day". Being unable to reach a human is the third irritant in the study; 71% refuse full automation as soon as the situation becomes sensitive, and a complaint is always sensitive.
3. A closed scope. The robot answers what it knows, and for everything else it says "I don't know, I'll pass this on". An assistant that would rather answer something than nothing is the exact definition of the off-topic reply. It is the same principle I argue for with replies to Google reviews: the AI prepares, the owner keeps control.

What a chatbot really costs, and the free alternative

The subscription is the visible part: from a few tens to a few hundred euros a month, depending on whether it is connected to your tools. The invisible part is maintenance. A chatbot only knows what it has been given, and a local business changes all the time. Summer hours, a new menu, a product out of stock, a holiday closure, a price change. Every change not passed on is a wrong answer waiting to happen. Budget ten minutes a week to read the conversations and fix them, and ask yourself honestly: who on your team is going to do that? If the answer is "nobody", install nothing. An up-to-date website with no chatbot beats a chatbot that is out of date.

Because the alternative exists, and it costs nothing more than care.

• Exact opening hours on every page, as text, with exceptional closures announced in advance, and the same hours on your Google Business Profile. That is question number one, and it disappears.
• A frequently asked questions page written from your real questions of the week, not invented ones. Delivery, area, parking, pets, pushchairs, allergens, payment methods, returns.
• Prices on display. Menu as text, service rates, ranges if you must. "On request" generates a request — it is mechanical.
• A human channel with a stated delay: a call button, a WhatsApp Business link with an automatic greeting and the response time shown. Customers accept a two-hour wait when they know it is a two-hour wait.
• Online booking or ordering on your own site when your business lends itself to it, which absorbs "have you got room?" without a word of conversation. Google is in fact starting to book directly on your customers' behalf, which assumes exactly that foundation.

That combination covers the large majority of a neighbourhood business's requests. And all of that text, unlike a conversation in a widget, is indexed, citable and works for your visibility. It is the heart of what I build into a shop or local business website: answering before anyone asks.

The decision checklist

1. List the last ten questions you received, across every channel. Not from memory — from notes.
2. For each one, is the answer on your website one tap away on a phone? If not, fix the page. It is free, and it counts for Google and the AIs too.
3. Is what remains an action? Booking, ordering, making an appointment. Then you need the matching module, not a chatbot.
4. Does what still remains depend on data that changes daily? Stock, availability, compatibility, intersecting slots. That is the chatbot's only territory, and only when connected to that data.
5. Before installing: disclosure, human, scope. It introduces itself as an AI, a human button is always visible, it knows how to say "I don't know".
6. Name the person who reads the conversations ten minutes a week. No name, no chatbot.
7. Measure two numbers: the share of conversations handed to a human, and the share left unanswered. If the first is over half, the robot is pointless. If the second is zero, it is making things up.

Frequently asked questions

Does a chatbot really bring more bookings?

Not the chatbot itself: the booking system behind it. A customer who wants a table at 11 pm needs a widget that shows available slots and confirms, not a conversation. If your site already offers online booking, a chatbot adds almost nothing. If it doesn't, the module is what to install first. The only real conversational value shows up for requests the module can't take — the party of twelve, the private hire, the allergy — and those must end up with a human. I have laid out the layers in what AI can automate in restaurant bookings.

Do customers accept talking to a chatbot?

For simple, factual questions, yes, as long as they can get out. The Tersea × Ipsos BVA study of September 2025, on 1,000 French adults, found that 62% prefer a fully human exchange and only 6% would accept being handled exclusively by a chatbot. The number one irritant is the off-topic automated reply (58%); being unable to reach a human comes third (53%). A chatbot is tolerated as a shortcut, never as a wall.

Do I have to tell customers they are talking to an AI?

Yes, and it is no longer just a courtesy. Since 2 August 2026, Article 50 of the EU AI Act requires that a person be informed they are interacting with an AI system, unless it is obvious. On the customer side, 72% of French consumers say they would feel deceived if they were not told. One sentence when the window opens is enough: "I'm an automated assistant. To talk to a person, click here." What the law actually asks of a local business is in the AI Act article.

How much does it cost for a small business?

The subscription is the small part: from a few tens to a few hundred euros a month. The real cost is maintenance time. A chatbot only knows what it has been given, and a local business changes all the time. A knowledge base that isn't kept current produces exactly the irritant customers hate most. If nobody has ten minutes a week to read and fix, you are better off not installing one.

What is the alternative?

A website that answers before the question, plus a human channel with a stated response time. Exact hours on every page including exceptional closures, a FAQ page in readable text, prices on display, a call button and a WhatsApp Business link with the response time shown. That combination covers most of a local business's requests with no subscription, and the same text is what ChatGPT, Gemini and Google read when deciding whether to recommend you. If your Google listing is the weak spot, start with the Google Business Profile and website duo.

What about a chatbot on WhatsApp or Instagram instead of the website?

Same logic, same scope. The channel does not change the question: what live data is it connected to, and who do you get out to? An automatic WhatsApp responder that states your hours and the human response time is useful and free. A robot that claims to take bookings on Instagram without seeing your diary is the same problem as a widget on the site, with one extra customer who thinks they have a table.

If you are unsure, write to me with your last ten customer questions: I will tell you which belong on the website, which need a module, and whether a single one is left that justifies a conversational assistant. Nine times out of ten, the answer fits inside a website that answers before the question. The tenth is the kind of tool I set up as part of AI for local businesses, connected to your real data, with you keeping control.