What a custom AI assistant actually is
An LLM, or large language model, is the type of AI behind ChatGPT. On its own it knows a lot about the world and nothing about your business. A custom assistant connects that model to a small library of your own files, so when a visitor asks about your Saturday hours or your cancellation policy, it looks the answer up in your documents first.
That lookup step has a technical name, retrieval, and it is what separates a business assistant from a general chatbot. We keep the assistant focused on your business and set it up to say so politely when a question falls outside what you have given it.
Where it lives and what it can handle
Usually it sits as a chat bubble on your website. Visitors can ask questions in plain words instead of hunting through pages. It can also be linked to WhatsApp, though WhatsApp has its own rules about when businesses can message people, so we cover that on our AI chatbot page.
- Questions about services, products and what each price includes
- Opening hours, location and parking notes from your own text
- Policies such as deposits, delivery areas or allergens, taken from your documents
- Collecting a name and number when someone wants a booking or a quote
- A visible 'talk to a person' option at any point
It can still get things wrong
We want to be straight about this. Retrieval makes answers more grounded, but it does not make them perfect. A Stanford study of commercial legal research tools built this way found they still produced wrong answers in roughly 17 to 33 percent of tests (Stanford RegLab). A small business assistant has a much narrower job, yet the risk is never zero.
Your business is also responsible for what the bot tells customers. In the Moffatt v Air Canada case, a tribunal held the airline liable when its website chatbot gave a passenger wrong information about a fare (ABA summary). That is why we route prices, bookings and refunds towards a human, and why you should be able to read the conversations.
Telling people they are talking to AI
Since 2 August 2026, Article 50 of the EU AI Act has required chat systems to make clear to people that they are dealing with AI (European Commission guidelines). We build that message into the chat window from the first reply. This is general information, so if your business is regulated, check the rules that apply to you as well.
Your data and which AI powers it
Customer chats can contain names, phone numbers and personal details, so we treat them with care. Where the provider offers it, we choose European data storage and use their data processing agreement. You decide what goes into the assistant's library, and we leave out anything private.
Several companies offer the underlying model. For example, as of October 2026 OpenAI offers built-in document search with an EU data residency option (OpenAI data controls), and Anthropic's default is to delete API data within 30 days, though some newer models have different retention terms. Tools and terms change quickly, so we check them when building. We pick the provider that suits your needs, cost and privacy preferences, and explain the trade-off before building.
Is it right for you?
It makes most sense when you have plenty of written information and visitors keep asking about it: a menu with dietary notes, a long list of treatments, rental terms, course timetables. If your questions fit on one page, a clear FAQ section on your website may do the job for less.
To talk it through, fill in the brief at /start/ or message us on WhatsApp. We reply within 1 working day.