What does a Make.com and GPT chatbot actually involve?
A chatbot built this way is a Make.com scenario that receives an incoming message through a webhook, sends that message to an OpenAI GPT model as a prompt, and sends the generated response back to wherever the message came from. No custom code is required for any of it.
Where this pays off: a simple version of this can answer common questions on a website or messaging app around the clock, without you writing a line of code or paying for dedicated chatbot software. It is also a genuinely useful way to learn how prompts, webhooks, and automation fit together.
What do you need before you start?
You need an OpenAI account with API access, which is separate from a regular ChatGPT subscription, and a Make.com account, which has a free tier suitable for building and testing this scenario. You will also want a messaging platform in mind, like Telegram, Slack, or a webhook on your own website, to send and receive messages through.
How do you build the chatbot step by step?
Building this scenario takes six steps: set up the trigger, configure the webhook, connect to GPT, craft your prompt, send the response back, then test and refine. Each step below builds on the one before it.
- Create a new scenario and choose a webhook trigger. A webhook lets your chatbot receive messages from platforms like Telegram, Slack, or your own website whenever someone sends one.
- Configure the webhook. Add a webhook module, which gives you a unique URL. Send a test message to that URL from your chosen platform so Make.com can learn the shape of the incoming data.
- Connect to GPT. Add an OpenAI module and choose the chat completion action, which sends a prompt to GPT and returns a generated response. Connect your OpenAI account using your API key.
- Craft your prompt. Build a prompt that includes the user's message, pulled from the webhook data, along with instructions describing how the chatbot should behave. For example: "You are a helpful assistant for a small bakery. The customer says: [message]."
- Send the response back. Add a module that sends the GPT-generated text back to the platform the message came from, whether that is another webhook call or a platform-specific module like Telegram's send-message action.
- Test and refine. Send a variety of test messages and watch how the chatbot responds. Adjust your prompt and any parameters, like response length, until the behavior feels right.
What makes a good prompt for a chatbot like this?
A good prompt clearly defines the chatbot's role, tone, and boundaries before it ever sees the user's message, since GPT responds to whatever framing you give it. Vague prompts produce vague, inconsistent answers.
Start by telling the model who it is, such as "You are a support assistant for [your business]," then describe what it should and should not do, like staying on topic or referring complex questions to a human. Keep early versions simple and adjust based on real test conversations rather than guessing at the perfect prompt upfront.
How do you give the chatbot memory of earlier messages?
By default, each request to GPT is independent and has no memory of earlier messages in the same conversation, so anything the chatbot needs to "remember" has to be included in the prompt each time. Make.com's built-in data store, or a connected spreadsheet, can hold recent messages so you can append them to future prompts.
For most simple use cases, like answering FAQs, this kind of memory is not strictly necessary. It becomes more valuable once you want the chatbot to handle longer, more natural conversations that reference something the user said a few messages earlier.
What should you watch out for with this setup?
The biggest things to watch for are cost, since API usage is billed separately from a ChatGPT subscription, and reliability, since GPT can occasionally produce an unexpected or off-topic response. Add error handling and a fallback message for cases where the automation fails.
It is also worth setting expectations with anyone using the chatbot that they are talking to an AI assistant, particularly if it might hand off to a human for anything complicated. Our guide to AI ethics and risks covers responsible-use considerations that apply here too.
What should you test before letting people use it?
Before pointing real users at your chatbot, test it against ordinary questions, edge cases, and a few messages meant to confuse it. This reveals most of the rough edges while the stakes are still low.
- Ordinary questions. Confirm the chatbot answers the questions you actually expect people to ask, in the tone your prompt described.
- Edge cases. Try messages with typos, very short messages, or questions completely outside the chatbot's intended scope, and see whether it handles them gracefully.
- Confusing or adversarial messages. Send a message trying to get the chatbot to ignore its instructions. Tightening the prompt is usually enough to handle most of these attempts for a simple internal or low-stakes use case.
- Failure behavior. Confirm what happens if the OpenAI module returns an error or times out. A clear fallback message is better than a broken automation and silence.
Treat the first week of real use as an extended test phase. Reading through actual conversations and adjusting your prompt based on what people really ask is far more effective than trying to anticipate everything in advance.
Where does this fit with other automations?
A chatbot is a natural extension once you understand basic Make.com scenarios, since it uses the same trigger-and-action pattern as simpler automations, just with an AI model in the middle instead of a plain data transfer. If you have not built a scenario before, our step-by-step guide to turning emails into Trello cards is a gentler starting point.
Once your chatbot is handling incoming questions, you might also want a way to summarize what it is learning from users. Our guide to AI reports with Make.com shows how to turn that kind of data into a regular digest.
Next step: for the fundamentals of how Make.com scenarios work before you add AI into the mix, visit our automation hub.
