Turn your website into your receptionist's brain

Your website says what your business does. Your receptionist should say the same thing.

Obvious when stated, routinely violated in practice.

You and I have both met the chatbot that contradicts the pricing page it sits on. The fix is not discipline; it is making the website and the receptionist read from the same source.

Before the how, one distinction is worth getting right, because the whole system is organised around it.

Structured vs unstructured

Everything your business knows falls into two piles.

Unstructured data is prose: your pages, your blog posts, your policies. Things you explain. It has no fixed shape, and its value is in the wording.

Structured data is records: products, services, prices, employees. Things you list. Every entry has the same fields (a name, a description, a price), and its value is in being exact.

The two are handled differently on purpose, because they fail differently.

Prose that is slightly loose is fine.

A price that is slightly loose is a refund conversation.

So prose goes into the knowledge base, and records go through dedicated APIs, addProducts and listProducts, which get their own post next. Keep this segregation in mind as you write content; it decides where each piece of information should live.

searchMyKnowledge

The unstructured half works like this. Your website content and blog posts sync into a knowledge base: the unglamorous name for a store of your prose that an AI can search by meaning, not by keyword.

Then a tool called searchMyKnowledge is enabled on your receptionist. From that moment, when a customer asks a question, the receptionist searches your actual content and answers from what your site really says, not from the model's general guesses.

Search is the middle of a trio:

• addToMyKnowledge puts content in.
• searchMyKnowledge reads it.
• deleteMyKnowledge takes it out again, behind a deliberate type-to-confirm step, because a knowledge base you cannot prune fills up with last year's prices.

In, search, out: the full lifecycle, and all three run without touching website code.

The loop this creates is quietly powerful.

A question keeps coming up in chats. You write a blog post answering it properly. The post publishes on the website and syncs to the knowledge base, so the receptionist learns the answer the same day.

One piece of writing, teaching two surfaces.

Why markdown makes this fast

Here is the mechanical reason the loop moves quickly. All of this content lives in markdown: plain text files with the lightest possible formatting marks.

AI models are unusually comfortable in markdown. It is close to the native format of how they read and write.

Which means you can use AI end to end:

• Research a topic.
• Draft the post in markdown.
• Revise it in conversation.
• The finished file is the deliverable.

That same file then does three jobs:

• It publishes as a styled page on your website.
• It syncs into the knowledge base.
• And your receptionist starts answering from it.

One format, three outcomes, nothing out of sync, and the whole cycle runs at writing speed rather than development speed.

Knowledge changes; code does not have to

Notice what never happened in any of the above: nobody edited website code.

The content layer and the code layer are deliberately separate. Updating what your business knows, its prose and its records, never requires touching how the website works.

You bought the code once; your knowledge keeps moving without it.

That separation is exactly what makes the structured half possible, and that is where we go next.

Next up: updating your products without touching the website code.