“Can’t we just put ChatGPT on the website?” is 2026’s most common automation question — and it’s a good one, because the gap between a raw model and a business-ready assistant explains where the value (and cost) of a custom AI chatbot actually sits.
What raw ChatGPT gives you
Fluency about everything and authority about nothing. It doesn’t know your prices, your service area, this week’s availability, or that you don’t do commercial jobs — and it will cheerfully improvise all four. On your website, improvisation is a liability with a chat window.
What “custom” actually adds
- Your knowledge: services, pricing bands, FAQs, policies and tone, retrieved from a curated knowledge base (RAG) so answers cite your reality, not the internet’s average.
- Guardrails: what it must never do — quote fixed prices, promise dates, discuss competitors, wander off-topic. Tested against the awkward questions before launch, not after a screenshot goes viral.
- Actions: the difference between a talking FAQ and a lead machine — checking the calendar, booking the call, creating the CRM record, alerting a human when the conversation is worth money.
- Memory and analytics: transcripts, conversion tracking, and monthly reviews that turn real conversations into better answers.
Cost vs consequence
A DIY widget costs pocket money and behaves like it. A production build — trained, guarded, integrated with calendar and CRM — is typically £800–£2,500 for UK businesses. The difference in outcomes isn’t subtle: one deflects visitors, the other books them in.
The test that settles it
Ask any chatbot three questions: “How much for [your commonest job] in [your town]?”, “Can you fit me in Thursday?”, and something completely unrelated. A raw embed flunks all three differently; a proper build answers, books, and politely declines respectively. That’s the product gap.
I build the second kind — book a call and bring your ten most-asked customer questions; that’s day one of every good bot.
