AI chatbot for your business: ChatGPT wrapper or custom?

Your customer service gets the same ten questions every day. Where is my order. What are your delivery times. Can I still return this. At some point you think: surely an AI chatbot for business could handle this? And then you notice two camps. One says: just get a subscription to a chatbot tool, it runs on ChatGPT and can be live tomorrow. The other says: have something custom built, otherwise the bot knows nothing about your company.

Both are right some of the time. In this piece we lay out when an off-the-shelf solution is enough, when you really need your own data and logic, and what you need to sort out on privacy before a single customer talks to your bot.

What is a "ChatGPT wrapper", really?

The term sounds a little dismissive, but it simply describes how most chatbot tools work. Underneath sits an existing language model, from OpenAI, Anthropic, Google or someone else. Around it, the vendor builds a layer: a chat window for your site, a place to upload some text or an FAQ, a few settings for tone and language. You pay per month.

There is nothing wrong with that. The model underneath is the same capable model you would use in a custom build. The difference is in everything around it: how much control you have over the information the bot uses, what it is allowed to do, and where your data ends up.

When an off-the-shelf chatbot is enough

We will be honest: for a good share of small and mid-sized businesses, a standard tool is the smartest start. That holds when:

• The questions are mostly general. Opening hours, delivery times, how returns work. Things that are on your site or fit in a handful of documents.
• The bot only has to answer, not act. No looking up orders, no booking appointments, no changing data.
• A wrong answer is annoying, but not harmful. If the bot misses a question now and then, it hands over to a person and no harm is done.
• You want to learn first. Which questions actually come in? How do customers respond to a bot? You find that out faster with a tool you switch on this week than with a build project.

For a chatbot platform like that, count on roughly €50 to €500 per month, depending on the number of conversations and features. For internal assistants, like a business ChatGPT or Claude plan for your team, you are looking at more like €20 to €100 per user per month. Easy to oversee, and easy to stop again.

Even then there is work to do. The bot is only as good as the information you give it. A messy FAQ gives you a messy bot. Set aside a few days to sharpen your answers, and watch what people ask during the first weeks.

When you need your own data and logic

At some point you hit the edges. You usually recognise it in one of these situations.

The answers are not in one place

Your knowledge is spread across manuals, product sheets, contracts, old emails and an internal wiki. A standard tool often lets you upload a few files, but it cannot search properly through thousands of documents that also keep changing. Then you need a setup that looks up the right pieces of information before the model answers. That is called RAG, and we explain it calmly in RAG explained: AI that answers with your own business data.

The bot has to do something in your systems

"Where is my order?" can only be answered if the bot can look into your order system. "Can I come by on Tuesday?" needs access to your calendar. The moment a bot has to connect to your own software, and certainly if it is allowed to change things there, you want to decide how that happens. Which data it can see, what it can change, and when a person needs to check first.

The rules are yours

Maybe business customers get different terms than consumers. Or the bot should deliberately not answer certain questions, such as legal or medical advice. You want that logic set down in code, not hope that a setting in a tool handles it well.

The conversation is your product

If the chatbot is not support but the core of what you sell, you do not want to depend on a vendor's roadmap. Then you build it yourself, or have it built, in a way you can change and take with you.

Custom work usually starts with a first working version in 2 to 4 weeks, so you quickly see whether it works. A full bot that runs on your own data and talks to a few systems soon lands in AI agent territory. What that costs, you can read in what it costs to build an AI agent. We often build such a bot as part of an existing portal or tool, the way we describe it under web application development.

Privacy and GDPR: what you need to sort out

This is the part people tend to skip, and it is exactly where things go wrong. A chatbot almost always processes personal data. A name, an order number, an email address, sometimes more than you think, because customers type anything into that window. So GDPR simply applies. A few things you want to know in any case, whether you buy a tool or have one built.

Do you have a data processing agreement? The vendor of your chatbot, and the company behind the language model, process data on your behalf. You put that in writing. With a serious vendor that agreement is simply available. Ask for it.

Where is the data stored? Are the conversations stored and processed in the EU, or do they go to the US? That affects what you need to arrange. More and more providers offer EU storage. Check it, and choose deliberately.

What may the vendor do with your input? Business API terms usually rule out training on your data, but check it. Free or consumer versions are often a different story. So do not let your team paste customer data into a free chat tool.

What goes into the prompt? Only send what is needed to answer the question. A bot that mentions a delivery time does not need to see a full customer profile. Sending less data is the simplest privacy measure there is.

How long do you keep conversations? Logs are useful for improving the bot. But do not keep them forever, and make sure you can delete them when a customer asks.

With custom work you have more grip on all of this. You decide which model you use, in which region, what gets logged and what gets filtered out before anything reaches the model. With a standard tool you depend on what the vendor offers. That is no reason not to do it, but it is a reason to actually read the terms.

And apart from GDPR: since 2 August 2026, the EU AI Act requires you to make clear that someone is talking to AI and not to a colleague. A short notice in the chat window is enough. What else to sort out for the AI Act is in our checklist for SMBs.

How to choose

A simple way to think about it:

• General questions, no integrations, and a wrong answer is not a big deal? Start with an off-the-shelf tool.
• Does the bot need to search through lots of your own documents that change often? Then look at a setup with RAG, as a tool or custom.
• Does the bot need to do things in your systems, or do your own rules apply? Then custom becomes interesting.
• Is the conversation itself your product? Then you want it in your own hands.

And it does not have to be all or nothing. Plenty of businesses start with a tool, learn which questions really come in, and then build the custom piece the tool cannot do. That way you only pay for custom work where it actually earns its keep.

Want a second opinion?

Not sure whether a standard tool is enough, or already running into its limits? Feel free to run it by us. We will tell you honestly when a subscription of a few tens of euros a month does the job. And if it does need to be custom, we will work out together what that means. Book a call.

Frequently asked questions

Is a ChatGPT wrapper worse than a custom build?

Not necessarily. Most tools run on the same kind of language model a custom build would use. The difference is control: which data the bot uses, what it can do in your systems, and where your data ends up. For general questions, a tool is often fine.

Can I use customer data in an AI chatbot?

Yes, if you set it up properly. You need a data processing agreement, you know where the data is processed, and you only send what is needed. Also check that the vendor does not use your input to train models.

Do I have to tell customers they are talking to a bot?

Yes. Under the AI Act it must be clear that someone is talking to AI and not to a person. A short, visible notice in the chat window is enough.

Can I move from a tool to a custom build later?

Yes, and that is often a smart route. A tool teaches you which questions really come in. After that, you build exactly the part the tool cannot do, such as connections to your own systems.