AI Chatbot and WhatsApp Assistant Development

What is an AI chatbot, and how is it different from a rule-based bot?

An AI chatbot is a conversational assistant that uses a large language model to understand questions users type in their own words and generates answers from the organisation's approved information. Rule-based bots follow a pre-drawn decision tree instead: the user taps buttons or types specific keywords, and the bot shows a prepared reply.

The two approaches are not rivals. Steps that must run the same way every time, such as payment, appointment confirmation or identity verification, are handled by a rule-based flow, while open questions are handled by the language model. Most of the chatbots we build use this hybrid structure.

Which channels can a chatbot work on?

We can connect a single assistant, using the same knowledge base and the same rules, to several channels. Customers then get the same information on your website and on WhatsApp, and the information is updated in one place.

• Website: a chat window that opens in the corner of the page, or a question box embedded in support, product and contact pages.
• WhatsApp: an assistant that runs on your verified business number through Meta's official WhatsApp Business Platform API.
• Instagram messages: a bot that answers messages sent to your business account through Meta's official messaging API.
• Apps and internal tools: an assistant added to your mobile app, customer portal or an internal portal for employees.

WhatsApp has its own rules. Free-form replies are allowed for a set period after the customer writes to you; outside that window, or when the business sends the first message, only templates approved by Meta can be used. Marketing templates require the user's explicit opt-in, and messages are subject to Meta's pricing. Meta's business policies also restrict general-purpose AI conversations unrelated to the business's own services, so we keep the WhatsApp assistant's scope limited to your business topics. We do not use unofficial tools that automate personal accounts, because they can get the number banned.

How does a chatbot answer with your company knowledge?

The bot's answers rest on a knowledge base made up of sources you choose and approve: frequently asked questions, product and service descriptions, pricing and campaign terms, delivery and return rules, branch and opening hours information. For every question, the model finds the relevant sections in these sources and writes its answer based on them. We explain the technical details of this method (RAG), along with model and hosting options, on our AI integration page.

1. 1.We choose the sources together and remove content that is outdated, contradictory or should not be shared.
2. 2.Changing information such as order status, stock or appointment availability is read live from the relevant system rather than copied into the knowledge base.
3. 3.We tie the bot's tone of voice, form of address and answer length to written rules that match your brand's language.
4. 4.We agree from the start who owns the knowledge base and how often it will be updated.

How is a conversation handed over to a human agent?

A good chatbot does not try to solve everything itself. Complaints, return requests, payment problems, sensitive topics such as health or legal matters, and questions the bot cannot answer are handed over to a human agent. The handover rules are agreed in writing before setup.

• Users can reach a person at any moment by writing that they want to talk to an agent.
• The agent takes over with the conversation history and a short summary prepared by the bot, so the customer does not have to explain everything again.
• Requests that arrive outside working hours are logged, and the customer is told clearly when to expect a reply.
• The handover goes to the right team and the right queue in the support or CRM tool you use.

How does a chatbot collect sales, appointment and quote requests?

A chatbot does more than answer questions; it can also move the visitor to the next step. In flows such as product recommendations, quotes, demos or appointment requests, the bot asks for the necessary details within the conversation, prompts for anything missing and writes the request directly into your CRM or booking system.

• Qualification: questions about needs, location, timing or budget range determine which team the request goes to.
• Appointments: available slots are read from your calendar, the user picks one and a confirmation message is sent.
• Quote and contact requests: the request is passed to the sales team with its source and a summary of the conversation.
• Follow-up: if the user has opted in, notifications such as appointment reminders are sent with approved templates.

For the steps that run behind the conversation, such as logging the request in the CRM, notifying the team or updating records, see our AI automation page.

Can a chatbot work in several languages?

Yes. Language models recognise the language the user writes in and can reply in the same language. To get the same answer quality in every language, however, the knowledge base must be accurate in that language too, and testing must be done separately for each one. Before launch, we test with real questions in every language that terms and forms of address are used correctly.

How is the risk of wrong answers reduced?

Language models can produce a fluent answer even on topics they know nothing about. In a bot that talks to customers this risk never disappears entirely, but design reduces it significantly:

• Scope limit: the bot answers only on defined topics and politely declines off-topic questions.
• The I-don't-know rule: if the knowledge base has no answer, the bot does not guess; it says so and directs the user to an agent.
• Showing sources: the page or document behind the answer is shown to the user with a link.
• Exact facts come from systems: prices, stock and order status are read from the relevant system, not guessed by the model.
• Protection against misuse: messages that try to push the bot outside its rules (prompt injection) are tested for and logged.
• Test set: a set of real customer questions is rerun before every change.

How are data protection, privacy notices and record retention handled?

Chatbot conversations can contain personal data such as names, phone numbers, orders or health information. Setup is therefore planned under the applicable data protection law, such as Turkey's KVKK or the GDPR in Europe. At the start of the chat, users are told clearly that they are talking to an AI assistant and how their data is processed, with a link to the privacy notice.

• The bot asks only for the information the task needs; fields such as ID numbers and card details are not requested in the chat, or are masked.
• Where conversation logs are stored, how long they are kept and who can access them is documented.
• Data processing terms with the model provider are clarified, including whether data is transferred abroad and whether it is used for training.
• Separate, explicit consent is collected for marketing messages.

We carry out the legal assessment together with your legal adviser, while we put the technical safeguards in place and document them.

Does a chatbot affect site speed and accessibility?

A carelessly installed chat window can be a heavy script that slows down page loading. To prevent this, we load the chat component after the page's main content has loaded, or when the user clicks the chat button, so Core Web Vitals scores and search visibility are protected.

Accessibility matters just as much: the chat window should open and close with the keyboard, announce new messages to screen readers, have sufficient colour contrast and not cover page content on mobile. We run these checks before launch.

How is a chatbot's success measured?

A chatbot's value shows in the work it resolves, not in the number of conversations. With the setup, we connect conversation events to your analytics and track these indicators:

• Conversations resolved by the bot, and the user's positive or negative feedback.
• Conversations handed over to agents, and the reasons for handover.
• Questions the bot could not answer; these form the list of content to add to the knowledge base.
• Quote, appointment and contact requests generated in the chat.
• The change in the time agents spend on repetitive questions.

This measurement is set up with the same approach as our CRO and analytics work, so the chatbot is tracked in the same report as your site's other conversion points.

When do you not need a chatbot?

Not every business needs a chatbot. If incoming questions are few and similar, a well-written FAQ page, a simple contact form and a visible phone number do the same job at lower cost. If the content to build a knowledge base does not exist yet, or there is no team to pick up handed-over conversations, it is better to fix those gaps first.

If you are not sure which channel and which scenario will add value for you, our AI consulting service can help clarify the need first. If the bot needs not only to answer but also to take steps on its own across several systems, the right solution may be an AI agent.

Why UNIT İstanbul for chatbot development?

UNIT İstanbul is an agency that has managed brands' digital growth through measurement since 2009, and it runs chatbot projects under its software brand, Unit Software. We build the bot not just as a piece of software but as a conversion point that welcomes visitors from your ads, answers their questions and gets their requests to your sales team. Because strategy, development and measurement sit in the same team, you work with a single point of contact.

To discuss the use case you have in mind, reach us through the contact page; in a free initial call we will assess together whether a chatbot is right for you and which channel makes sense to start with.