FAQ chatbot for a website: what it solves, what it does not
An FAQ chatbot answers, on your website, the questions whose answer already exists in writing: opening hours, prices, lead times, terms, how to order, address and access. Properly fed, it absorbs most repetitive enquiries and turns the rest into qualified requests. It does not look up a customer file, does not commit the company contractually, does not replace professional advice, and does not rescue a website whose information is missing or wrong. Its quality depends almost entirely on the knowledge base written for it, not on the language model behind it. Expect €350 setup and €100 per month for a supervised deployment, knowledge base included.
What it actually solves
An FAQ chatbot handles one precise category of enquiries: those whose answer is stable, written down, and identical for everyone. It is unglamorous, and that is exactly where the volume sits.
- Pure information questions: opening hours, closing days, address, access, payment methods, usual lead times. They require no judgement and repeat endlessly.
- Pricing questions when the grid is public. The agent gives the amount together with the condition attached to it, which a pricing page alone does badly because nobody reads it in full.
- Routing. A vague request is reformulated into a precise question, then directed to the right page, the right form or the right person. It is the least visible gain and the most useful one.
- Qualification before contact. When a request goes beyond the FAQ, the agent collects what is needed and forwards a complete enquiry, instead of a three-word message you have to answer just to understand it.
The real benefit is therefore measured in two places: how many enquiries no longer reach a human, and how good the ones that still do have become.
What it does not solve, and must not attempt
This section is longer than the previous one, deliberately. Most disappointments come from a chatbot asked to do a job that was never its own.
- It does not read your files. "Where is my order?" assumes a connection to your management software and an authenticated visitor. Without both, the only good answer is a handover to an identified channel.
- It does not commit the company. A firm quote, an exceptional discount, a contractual exception: these are decisions, not information. Our agents are explicitly constrained never to produce them.
- It does not replace professional advice. In health, law or accounting, the answer depends on a specific case the agent knows nothing about. It points to a professional, it does not rule.
- It does not compensate for an empty website. If opening hours are written nowhere and nobody knows them, no agent will invent them correctly. A chatbot amplifies the quality of your content, in both directions.
- It does not replace the phone for urgency and emotion. An unhappy customer wants a person. The agent must recognise that and hand over quickly, not argue.
What decides quality, in order
In our experience, the factors separating a useful agent from a gadget are not the ones people expect.
| Factor | Real weight | Why |
|---|---|---|
| The knowledge base | Decisive | It decides what can be answered correctly. It is also the only factor your company fully controls. |
| The refusal instructions | Decisive | They decide what will not be invented. An agent with no written limits eventually promises something. |
| The handover | Important | An agent that knows when to stop and pass on beats an agent that insists. |
| The choice of model | Secondary | Current models all write correctly. No model knows your prices, whichever one you pick. |
Ranking drawn from our own FAQ deployments: corrections made after go-live almost always concern the knowledge base or the instructions, almost never the model.
What it costs, and where the money goes
| Service | Setup | Monthly |
|---|---|---|
| FAQ Widget | €350 | €100 / mo |
| Concierge Widget | From €500 | €150 / mo |
Amounts excluding VAT, generated from the grid that produces our quotes. The subscription covers hosting, model usage whatever the volume, supervision and corrections.
Most of the setup fee does not pay for technology. It pays for the inventory of your content, writing down what exists nowhere, calibrating the tone and running regression tests. That work is what decides whether the agent answers correctly, and it is precisely the work a self-service licence leaves on your side of the table.
How to know whether it is the right tool for you
- For two weeks, write down the questions actually received by phone and by email. Without that record, the decision is made blind.
- Count how many of those questions already have an answer written somewhere. That is the exact scope of an FAQ chatbot.
- If that share is low but the email volume is high, the right tool is not a widget but an agent handling the inbox itself.
- If the answers depend on a customer file, the prior question is not the chatbot but access to your business software.
An FAQ chatbot is a good investment when a significant share of enquiries is repetitive and documented. Otherwise it adds one more layer without taking anything off anyone, and that is worth saying before selling it.
Will an FAQ chatbot annoy my visitors?
The risk is real, and it almost always comes from three specific faults. A window that opens by itself and covers the content, an agent that refuses to let people reach a human, and evasive answers that go in circles. None of the three is inherent to the technology: they are configuration choices. So we only open the agent automatically when the client explicitly asks for it, we always leave a visible path to a human contact, and we require the agent to say it does not know rather than pad. A visitor who gets an accurate answer in ten seconds is not annoyed, they leave satisfied.
How is it different from a button-based chatbot or a decision tree?
A button-based chatbot only understands paths planned in advance: the visitor has to guess which branch holds their question, and any unforeseen wording falls into a failure message. An agent built on a language model understands the question as asked, including badly phrased or in another language, and looks the answer up in a knowledge base written in plain text. The decision tree keeps one genuine advantage: it is perfectly predictable. That is why we combine both, framing the agent with strict instructions about what it is not allowed to answer.
Do we need an existing FAQ page to deploy one?
No, and that is rarely the case anyway. An existing FAQ page saves time, but it is never enough: it contains what the company thought to write down, not what customers actually ask. So we start from your existing content, whatever it is, then complete it with the questions collected on the phone and in the inbox. That second part is what gives the knowledge base its value, and it is also the part that exists nowhere before the project starts. A company with no FAQ page can perfectly well deploy an agent, the writing work is simply longer.
What happens if the agent gives a wrong answer?
Conversations are kept and reviewed during the supervision period that follows go-live. A wrong answer therefore surfaces quickly, and the correction happens in the knowledge base, not in the code: content is versioned, the new version replaces the old one, and rolling back stays possible without any loss. If something serious happens, the widget can be switched off immediately, which removes it from the site without your developer having to act. Those two safeguards, the review and the kill switch, are the reason we charge a subscription rather than shipping a licence with no follow-up.
Agents covered on this page
Each page details the deployment, the pricing and a live demonstration.