Chatbots & AI Assistants

Chatbot or AI assistant? Why the difference matters

Today the word chatbot is used to describe very different systems: rule-based bots that move through pre-written buttons, AI models that understand a question and write the answer themselves, and assistants that can take actions in your systems. Which one you need directly shapes the scope of the project and the experience your customer will have. That is why we start not by picking a technology but by defining the job the assistant will take on.

Rule-based chatbot

A rule-based chatbot moves through a decision tree drawn in advance: the user taps an option and the bot shows the next step. Its behavior is predictable and setup is relatively simple, but it gets stuck on a question that falls outside the tree. It is still a good option for narrow, clear jobs such as having someone pick an appointment time or asking for an order number.

AI chatbot

An AI chatbot uses a large language model such as GPT, Claude or Gemini in the background. It understands the question the user writes in their own words and builds its answer to fit that conversation. In customer communication, where the same question can be asked in ten different ways, this flexibility makes a big difference; on the other hand, the model has to be fed the right information and its limits have to be written out clearly.

An AI assistant that gets things done

One step further are assistants that do not stop at answering but take actions. An AI assistant like this looks for open slots in the calendar, checks order status, opens a record in the CRM or sends a notification to your team. When these permissions are granted, we define clearly during setup which actions the assistant can take on its own and for which ones it will ask the user or your team for approval.

In practice, most projects are a mix of these three approaches. For example, a clinic assistant can answer open-ended questions with AI and switch to a controlled, button-based flow at the appointment step. The factors that decide the mix are the cost of making a mistake, the variety of the questions and the systems to be connected.

An AI assistant for your customers and your team

When people say chatbot, they mostly think of a chat window that talks with a customer. The same technology also makes it easier for your employees to reach information inside the company. The two uses share a similar infrastructure, but their goals, risks and success measures are different.

Customer-facing use

• Pre-sales questions: product features, price range, delivery areas, which service suits whom
• After-sales support: order and shipping status, setup or usage questions, warranty and return terms
• Appointments and reservations: showing available times, taking bookings, passing on change requests
• Lead collection: gathering the information needed for a quote in the chat and passing it to the sales team

For a customer-facing assistant, the measure is that the customer gets the right answer without waiting and can easily reach a person when needed. Brand voice matters here too: the form of address, the length of answers and the tone used on each topic are set to fit your customer relationship.

Internal use

An internal AI assistant takes over the questions your employees ask each other every day. A new sales representative can learn a product’s technical specs, a support employee the return procedure and a branch employee the campaign rules without calling their manager. The information is often already in your documents; the assistant makes it searchable and findable.

• Answering from procedure, handbook and policy documents
• Comparing features from product catalogs and technical datasheets
• Suggesting a draft reply to customer representatives for an incoming message
• Summarizing long documents, meeting notes or customer correspondence

The most critical issue in an internal assistant is permissions. Accounting documents should be open only to the relevant team and personnel information only to authorized people. That is why we design which source the assistant answers from for which user around the access setup you already have in your company.

One knowledge base, several channels

If your customers reach you through your website, WhatsApp, Instagram or by phone, building a separate chatbot for each channel may look logical at first. Over time, though, this leads to the same question being answered differently on each channel: the price gets updated in one place and stays old in another. We prefer to connect the channels to a single knowledge base and shared rules.

In this structure the knowledge base, conversation rules and integrations are shared; what changes from channel to channel is only the presentation. On WhatsApp, short messages and option buttons stand out, on the website longer explanations and links, and on the phone short, spoken sentences. For channel-level details, see our WhatsApp AI Sales Assistant and website AI chatbot pages.

How do we choose the first channel?

Instead of opening all channels at once, we recommend starting with a single channel. We look at three questions: where most messages come from, where unanswered messages pile up the most and on which channel a wrong answer does the least harm. The knowledge base and rules that settle in on the first channel carry over to later channels with much less effort.

For example, for a boutique whose messages mostly come from Instagram DMs, the first channel could be Instagram; for a clinic that takes appointments by phone, a voice assistant makes more sense. A company that works with corporate customers often starts with quote questions on its website or with an internal document assistant.

Context across channels

Sometimes a customer starts a chat on the website, asks for details on WhatsApp and makes the final decision by phone. When the channels are connected to the same CRM, this path can be gathered in a single customer record and your team sees where the conversation left off. We decide with you which information will carry across channels, in a way that does not collect more personal data than necessary.

Where does an AI chatbot get the right information?

The phrase “brand-specific training” often gives the impression that a model is trained from scratch. For the vast majority of businesses, the work is different: the language model is used as it is, and it is given a knowledge base made of your documents and clear instructions. For every question, the AI chatbot first finds the relevant sections in this source and writes its answer based on them.

The technical name for this approach is “retrieval-augmented generation”, or RAG for short. Its biggest advantage is that there is no need to retrain the model when information changes; updating the relevant document is enough. In internal use, the assistant can also show which document an answer is based on, so the employee can verify the information at its source.

The state of your documents shapes the answer

An AI chatbot cannot resolve contradictions in its source on its own. If the same product has two different prices in two separate price lists, or an old return policy is still sitting in a folder, the assistant may rely on the wrong one. This is why we review the documents together before setup.

• Conflicting versions: if the old and new versions of the same topic sit side by side, it has to be clear which one is valid.
• Unwritten knowledge: rules that “everyone knows” are often written down nowhere; they need to be put in writing for the assistant.
• Scanned documents: tables that exist as photos or scanned PDFs cannot be read reliably without being converted to text.
• Frequently changing data: information such as stock, price and availability should be read live from the relevant system instead of being written into a document.

Choosing a model

We do not choose between GPT, Claude and Gemini by a single criterion. We weigh together how natural the answers are in Turkish, response speed, cost per message, the capacity to process long documents and the need to interpret content such as photos or audio. On channels like the phone, where delay is noticed immediately, speed stands out, while in an internal document assistant reading long texts correctly may matter more.

We take care to build the assistant so that it does not depend on a specific model. When the knowledge base, instructions and integrations are kept separate from the model, switching to a better-suited model later does not mean redoing the setup; it is enough to retest the answers with the new model.

What does it do when it does not know?

Language models can build a fluent and convincing answer even to a question that is not in their source. To limit this behavior, we instruct the assistant to answer only from the knowledge base and, when it is not sure, to say so openly and route the question to your team. Before launch, we test that the instruction really works with questions whose answers are not in the source.

Privacy and data security

An AI assistant is only as secure as the information you open to it. On the customer side, personal data such as name, phone number and order information is processed; inside the company, confidential documents such as contracts, pricing structures and personnel information may be involved. The rules on processing personal data are set by Law No. 6698 (KVKK, the Turkish Personal Data Protection Law); you can find the regulator’s guides on the Turkish Personal Data Protection Authority (KVKK) site.

• What stays outside the knowledge base: customer lists, identity information and internal correspondence that should never appear in answers are set aside from the start.
• Provider terms: we check where the language model provider processes data, how long it keeps it and whether it uses it for model training.
• Transparency: telling the user clearly that they are chatting with an AI assistant builds trust.

We recommend doing the legal assessment with your own legal counsel; we shape the technical design around those decisions.

Where to start: suitable jobs and scope

An AI assistant is not set up to take over every job but to do certain jobs well. To understand which job suits the assistant, we look at three criteria: whether the answer to the question exists in a written source, whether the same question is repeated often and whether the consequence of a wrong answer can easily be made right. If the answer to all three is yes, that job is a good candidate for automation.

Jobs suited to an assistant

• Repeating information questions such as opening hours, address and delivery areas
• Guidance and comparison when choosing a product or service
• Checking order, shipping or application status
• Appointment and reservation flows
• Collecting information before a quote and passing the request to the right person
• Employees’ questions about procedures and documents

What should be left to human judgment

• Questions that require medical, legal or financial assessment
• Price negotiation, special discounts and contract terms
• Decisions that require authority, such as approving a refund, deferring a payment or granting an exception
• Sensitive personal situations and serious complaints

This line falls in a different place in every sector. On an e-commerce site, explaining return terms can be left to the assistant, while in a clinic every question about treatment should be routed directly to a specialist. We set the boundary together, knowing your business and your risks.

What determines the scope

In chatbot projects, the question “how long will it take and how much will it cost” has no single answer, because two projects with the same name can do completely different jobs. When we pin down the scope, we look at these points:

• Number of channels: a single website window and a setup with web, WhatsApp and phone together do not take the same effort.
• Integrations: there is a clear difference between an assistant that only reads information and one that writes records to the calendar, the CRM or the order system.
• State of the knowledge base: a tidy FAQ document and a pile of scattered PDFs, emails and spreadsheets need different preparation.
• Number of languages: each language is tested separately; product names and terms are checked in every language.
• Conversation volume: language model providers generally charge by usage; as volume grows, this item grows too.
• Permission structure: in internal assistants, opening different sources to different teams requires extra design.

Language model usage is a cost item that continues as long as the assistant runs. That is why, when choosing a model, we look not only at answer quality but also at the usage cost at the expected message volume, and we talk through the options with you side by side.

Common wrong expectations

• “It learns on its own once it is set up”: An assistant that works from a knowledge base does not learn from conversations by itself; the source and the rules have to be updated regularly.
• “We won’t need a team”: A good AI assistant reduces repetitive work, but someone is always needed to take over the handed-off conversations.
• “It has to answer every question”: An assistant that says it does not know is more reliable than one that makes up an answer.
• “The newer the model, the better”: The result is often decided more by the quality of the knowledge base and the clarity of the rules than by the model.

After setup: maintaining the assistant

A chatbot is not finished on the day it goes live; the real learning period is the first weeks. Real users ask questions nobody thought of during testing, and the gaps in the knowledge base show up through those questions. In this period, you need to review conversation logs, list the questions that went unanswered and decide which topics to add to the source.

Part of maintenance is also your own business changing. When a new product, a changed campaign, an updated procedure or a newly opened branch is not reflected in the knowledge base, the assistant keeps giving the old information. That is why we settle during setup who will update which part of the source and how changes will reach the assistant.

• Unanswered questions: show which topics are missing from the knowledge base.
• Conversations handed off to the team: show whether the boundaries were drawn correctly.
• User feedback: shows whether the answers are clear and sufficient.
• New frequently asked topics: point to information missing from your website or product pages.

We report these findings regularly. The reports can be used not only to improve the assistant but also to see what your business hears most often from customers, because an AI assistant becomes a source that keeps a record of your customers’ questions.

Frequently asked questions

Is a chatbot the same as an AI assistant?

In everyday language they are used interchangeably. A rule-based chatbot moves through pre-drawn options; an AI-powered AI assistant understands a freely typed question, writes its answer based on your knowledge base and, if needed, takes actions in your systems.

Does an AI chatbot understand Turkish well?

Current models such as GPT, Claude and Gemini generally understand Turkish questions well and answer fluently. Even so, it is necessary to test with real example questions before launch for sector-specific terms, abbreviations and product names.

Can an AI assistant be used internally for employees?

Yes. An AI assistant fed with procedure, handbook and product documents can answer the questions employees ask often. In this use, we separately design it so that each employee gets answers only from the documents they are authorized to see.

Can I use the same assistant on more than one channel?

Yes. Your website, WhatsApp, Instagram and phone can be connected to the same knowledge base, so the customer gets the same information whichever channel they write from. Even so, it is healthier to start with one channel and add the others one by one.

Do my existing systems need to change to set up a chatbot?

Usually not. The assistant connects to your existing website, messaging accounts and systems that offer an API. For older systems without an API, the connection method is assessed separately.

Does 24/7 support mean my team will never need to step in?

No. An AI chatbot answers outside business hours and completes simple requests. For topics it does not know the answer to or that need a human decision, it records them and leaves them to your team.

Let’s define the right assistant for you together

Tell us the questions your customers and employees ask most, the channels you use and the systems you have connected; let’s work out together which job can be handled by a chatbot and which by an AI assistant that gets things done. You can see our other automation services in the AI and automation services section, and you can write to us through our contact page.