AI Chatbots on Kenyan Business Websites: Answering From Your Own Content

Your website already contains the answers to almost every question a customer asks. The prices, the delivery areas, the opening hours, the specifications, the process. The problem is that visitors do not read websites. They scan, fail to find the one thing they came for, and either message you or leave.

A website chatbot closes that gap. Not the scripted menu of numbered options that everyone learned to hate, but an assistant trained on your actual content that answers in sentences.

What a Modern Website Chatbot Is

What We Build, and How It Knows Things

The assistant is given a defined body of knowledge: your website pages, your price list, your frequently asked questions, product specifications, delivery areas, policies, and any document you want it to use, such as a service manual or an admissions prospectus.

When a visitor asks something, it answers from that material rather than from the open internet. That is the whole design point, and it is what separates a business assistant from a general chatbot: it should know your delivery charge to Kitengela and have no opinion about anything else.

Three things we build in as standard:

• Boundaries. It answers from your content, and when something falls outside it, it says so and offers a person rather than guessing.
• Handover. Any conversation can move to WhatsApp or your inbox, carrying what the customer already said so they do not repeat themselves.
• Capture. Where a conversation is clearly a lead, it collects a name and a contact so the enquiry is not lost when the tab closes.

About the AI Behind It

Clients ask which model we use. The honest answer is that it depends on the job, and the model matters less than what the assistant is allowed to know and allowed to say.

There is a genuine choice to make, though, and it is worth understanding in plain terms:

• Hosted commercial models are the simplest to run and generally the strongest at conversation. Your content is sent to that provider.
• Open models, including the DeepSeek family, can be run on infrastructure you or we control. That matters when the content is sensitive and you would rather it did not leave a defined environment.

Website Chatbot or WhatsApp Agent?

These are two different products solving two different problems, and most businesses eventually want both.

What Makes One Good

1. It knows your real prices. An assistant that says "please contact us for pricing" has replaced one dead end with another.
2. It admits ignorance. "I do not have that detail, let me put you to the team" builds more trust than a confident wrong answer.
3. It is fast and light. A chat widget that adds seconds to your page load costs you more visitors than it converts. It should load after the page, never before.
4. It works on a phone. Most of your visitors are on one. The widget must not cover the content or block the call button.
5. It is easy to escape. A clear route to a human, always visible.
6. Someone reads the transcripts. The questions people ask are the best list you will ever get of what your website fails to explain.

What Goes Wrong

• It invents things. Prevented by keeping it to your content and instructing it to decline rather than improvise. This is a configuration decision, and it is the one that matters most.
• Its knowledge goes stale. Prices change on the website and the assistant keeps quoting last year's. Updating the source has to be part of someone's job.
• It becomes a wall. If customers cannot reach a person, the assistant is not saving you work, it is losing you sales.
• It collects data it should not. Do not let a public chat widget gather ID numbers, medical details or payment information. Kenya's Data Protection Act applies to anything personal you collect, however friendly the interface.
• Nobody measures it. Without reading transcripts you cannot tell whether it is deflecting work or deflecting customers.

Who It Suits

A website assistant earns its keep when your site gets steady traffic and your team answers the same questions repeatedly. Typical cases we see: schools answering fee and admission questions, clinics answering hours and insurance questions, shops answering stock and delivery questions, and service businesses explaining process and pricing.

It does not suit a site with very few visitors. If ten people a week reach your website, the honest advice is to spend the money on being found rather than on answering the few who arrive.

Getting One

Website assistants are quoted per business, because the work depends on how much content there is, how sensitive it is, and what it needs to connect to. There is no sensible list price, and we will not invent one.

What we would ask you for at the start: your website, the five questions your team answers most, your current price list, and a decision on what the assistant may and may not discuss.

Our published AI pricing today covers the WhatsApp agent: from KES 10,000 per month for monthly management, and setup is free. See the WhatsApp AI agent page, and if the real gap is that your website cannot be found in the first place, start with SEO instead.

Frequently Asked Questions

What is the difference between a chatbot and an AI assistant on a website?

An old-style chatbot follows a decision tree you build in advance and fails on anything off-script. A modern assistant reads the question and answers from your own content: your pages, prices, policies and documents. When something falls outside what it knows, it should say so and offer a person.

How does the chatbot know about my business?

We train it on a defined body of your own material: website pages, price list, frequently asked questions, product specifications, delivery areas, policies and any document you choose. It answers from that rather than from the open internet, which is exactly the point.

Will it make things up?

That is the main risk and it is a configuration decision. We keep the assistant to your content and instruct it to decline rather than improvise, so it says it does not know and offers a human instead of inventing an answer. Nobody should sell you a site assistant without explaining how that boundary is set.

Which AI model do you use, and can it run on DeepSeek?

It depends on the job, and the data question matters more than the model. Hosted commercial models are simplest and strongest at conversation, but your content goes to that provider. Open models, including the DeepSeek family, can run on infrastructure you or we control, which matters when the content is sensitive.

Should I get a website chatbot or a WhatsApp AI agent?

They solve different problems. The website assistant catches visitors who are reading your site and cannot find something. The WhatsApp agent answers people who message your business number, qualifies them and books appointments. If you can only have one in Kenya, it is usually the WhatsApp agent, because that is where buying conversations happen.

How much does a website chatbot cost?

It is quoted per business, because the work depends on how much content there is, how sensitive it is and what it connects to. We will not quote a list price we would not honour. Our published AI pricing covers the WhatsApp agent at from KES 10,000 per month, with setup quoted.