MySiteGPT Assistant
MySiteGPT Assistant is a site assistant you install with one snippet. It reads the site's own pages, answers visitor questions from that content, and cites the page each answer came from. Retrieval-augmented rather than fine-tuned, which is the practical difference between an assistant that answers from what you actually published and one that improvises a plausible-sounding version of your business.
The actual capabilities.
Trained on your site, not the open web
It indexes the pages you point it at and answers from those. Ask it something the site does not cover and it says so, rather than filling the gap with a confident guess.
Every answer carries a citation
Each response names the page it came from and links there. That is the honest check on hallucination, and it also does the job a chat widget should do anyway: move the visitor to the page that answers them properly.
One snippet to install
Drops into Webflow, WordPress, Squarespace, Shopify, Framer or hand-written HTML. No re-platforming, no plugin ecosystem to negotiate.
Bring your own API key
Run it against your own model account so token spend, rate limits and the data path stay under your control and your contract, not somebody else's resale margin.
Multilingual by default
Answers in the language the visitor writes in, from source content that may only exist in one. Useful the moment a site sells outside its home market.
Captures the lead
A visitor who asks a buying question is a lead. The assistant collects the enquiry in the conversation rather than pushing them to a contact form they will not fill in.
Built for the questions people ask before they install anything: is this worth it, and is my content ready for it?
Work out the deflection maths before you commit.
Is there enough on the site for an assistant to answer from?
What an assistant would actually be able to read.
Valid FAQPage JSON-LD for the answers you already have.
Set the assistant's scope and tone deliberately.
Honest side-by-sides against the alternatives.
Who reaches for it, and why.
SaaS and documentation sites
Deflecting the same twenty support questions, with a link to the doc that answers each one.
Agencies
An assistant to install for clients that runs on the client's own key and content.
Support teams
Cover outside business hours without pretending a human is on the other end.
Sites selling across borders
Answering in the visitor's language without translating the whole site first.
Asked often enough to answer here.
How do I add an AI chatbot to my website?
Point the assistant at your site so it can index the pages, then paste one embed snippet before the closing body tag. It works on hosted builders and hand-written HTML alike — there is no plugin or platform requirement.
What does "trained on my site's content" actually mean?
Your pages are indexed and retrieved at question time, and the model answers from the retrieved passages. Nothing is fine-tuned into model weights, so a page you edit today changes the answers today, and removing a page removes it from the answers.
What is a RAG chatbot?
Retrieval-augmented generation: the system searches your content for passages relevant to the question, then asks the model to answer using those passages. It is what makes citation possible, and what keeps the assistant inside the boundary of what you have actually published.
Can I use my own API key?
Yes. Bring-your-own-key is supported, so the model calls run on your account, at your rates, under your provider agreement.
Will it make things up?
It answers from retrieved passages and cites them, which is the strongest practical guard available. Where the content does not cover a question, the correct behaviour is to say so and hand the visitor to a human — and that is what it is configured to do.
How do I tell whether a chatbot is worth it?
Start with the ROI calculator: current ticket volume, the share of those tickets your published content already answers, and the cost of handling one. If your content cannot answer them, the assistant will not either — which is what the content readiness checker is for.
Retrieval pipelines, embeddings and evaluation — the part of AI work where the demo is easy and the ninety-fifth answer is hard. Also the operational side: per-tenant content indexing, key handling, and keeping a widget fast enough that a visitor does not abandon it.
The same team takes this on for clients — see what we build, or start an enquiry.