This article was updated on November 4, 2026.
Building an AI chatbot takes seven steps: define the purpose, identify the target audience, choose a platform, design the conversation flow, connect a language model, integrate the bot with your business systems, and test and launch it. This framework still holds. However, some of the tool recommendations we made in 2025 have changed: OpenAI's Agent Builder is being shut down, and new rules are coming for AI bots on WhatsApp. Below you'll find both the steps and today's conditions.
What changed in this article?
The core idea of the article is the same. We updated three points:
- Platform information: OpenAI is deprecating Agent Builder. Existing users can keep using it during the transition period, and the product is planned to shut down on November 30, 2026. Committing to this tool for a new project would not be wise.
- WhatsApp rule: If you plan to publish your bot on WhatsApp, you need to know Meta's 2026 rule (we cover it in Step 6).
- Statistics: The 65% figure in the previous version dated from early 2024. The current data is below.
Why are AI chatbots still on the agenda?
AI chatbots can answer customer questions instantly, offer personalized recommendations and automate repetitive tasks. Their use is now widespread.
- According to McKinsey's early 2024 data, 65% of organizations regularly used generative AI. That share has risen in later surveys.
- In McKinsey's 2025 report, 88% of organizations say they regularly use AI in at least one business function, up from 78% a year earlier.
- In the 2026 survey, 44% of organizations report that AI is scaling across the company, up from 38% a year earlier.
When it comes to AI agents, the picture is more cautious. In the 2025 report, 23% of respondents say their company is scaling an agentic system, but in no single function does that share exceed 10%. According to the 2026 survey, large organizations have accelerated their use of agents, while the share of small organizations scaling agents has stayed flat at 22%. In other words, there is an opportunity for SMEs, but building a task-specific, measurable bot is more realistic than building a bot that does everything.
The 7 key steps to building an AI chatbot
Step 1: Clarify the bot's purpose
First, write down what job the bot will do. Will it answer customer questions, recommend products to increase sales, or take over routine tasks? Once the purpose is clear, it also becomes clear which data, which tools and which success metric you need.
Step 2: Define the target audience and use cases
Your bot may serve customers, employees or dealers. Each audience has different questions, language and expectations. List the 10-20 most frequent questions and design the bot for those first.
Step 3: Choose the right development platform
There are two paths:
- Custom development: With languages such as Python and JavaScript, it gives you full control to fit your needs.
- Ready-made, visual platforms: They deliver results faster without writing code.
The most important lesson of 2026 is this: don't depend on a single platform. The shutdown of Agent Builder is a concrete example. Build a structure that can keep working even if the platform changes. For example, keep your knowledge sources and business rules independent of the platform. For new work and migration planning, OpenAI also recommends using ChatKit with your own server-side agent setup. ChatKit remains available.
Step 4: Design the conversation flow
Plan how the bot will respond to each question, when it will ask questions, and when it will hand the conversation over to a human. The flow should include three things:
- A welcome message
- An exit path for situations it can't answer
- A handoff point to a customer representative
Step 5: Connect the language model
Modern chatbots run on large language models that understand the user's message and generate a response suited to the context. When choosing a model, compare speed, cost, Turkish response quality and data security. Make sure the bot answers only based on your own product, pricing and policy information. This reduces the risk of incorrect information.
Step 6: Connect the bot to external sources
If the bot can only talk, it stays limited. When it's connected to APIs, databases, e-commerce infrastructure or an ERP, it can report order status, check stock and create appointments.
If you plan to publish on WhatsApp, note the following: since January 15, 2026, WhatsApp's Business Solution Terms have prevented general-purpose assistants where AI is the primary product from being offered on the platform. This rule does not ban businesses' bots. Bots are allowed for structured business tasks such as customer service, order inquiries and appointment management. So design your bot around tasks specific to your business rather than as a broad "chat about anything" assistant.
Step 7: Test, improve and launch
Before launch, test with real customer questions. Try vague, misspelled and off-topic questions as well. After launch, review conversation logs regularly, add unanswered questions to the knowledge source and monitor the handoff-to-human rate. A bot is not a tool you set up once and then leave alone.
Checklist before you start
- Is the bot's purpose written in a single sentence?
- Have the 10-20 most frequent questions been listed?
- Are the knowledge sources up to date and in one place?
- Can the bot be moved if the platform changes?
- Have the rules of channels like WhatsApp been checked?
- Is it clear where customer data goes?
- Has the handoff point to a human been defined?
What does Medyateji recommend?
We recommend starting with a small, measurable use case, such as frequently asked questions or an order status inquiry. Once you see the results, expand the scope.
Medyateji works as a hybrid agency and on a turnkey basis: the scope, timeline and fee are clear from the start. There are no contracts and no surprise fees, and prices are published online. Depending on your topic, you can check these pages:
- Our AI Agent page for bot and workflow automation.
- Our Straiteji AI strategy work to decide which processes you should use AI in.
- Our AI Projects page for AI projects. Prices start at 10,000 TL.
- Our ERP integration service to connect the bot to your e-commerce or ERP data.
- Our website and software service for the website where the bot will live.
Frequently asked questions
Do you need to know how to code to build an AI chatbot?
No, not for simple bots; you can set them up with ready-made, visual automation platforms. However, a bot that works integrated with systems such as an ERP, an e-commerce platform or a database requires software support.
Is OpenAI Agent Builder still available?
OpenAI is deprecating Agent Builder, and the product is planned to shut down on November 30, 2026. ChatKit is still available; for new projects, using your own server-side agent setup is recommended.
Is it prohibited to use an AI chatbot on WhatsApp?
No. Meta's rule, which took effect on January 15, 2026, covers general-purpose assistants where AI is the main product. Task-specific bots for order tracking, customer service and appointments are allowed.

