You don't train Maya by writing prompts. You correct it the way you'd correct a new hire.

Most AI tools ask you to become a part-time bot builder — write the prompt, map the intents, draw the decision tree, and keep patching it when it breaks. Maya doesn't work that way. You point it at your website, it reads your public pages, catalogue, structure and connected business data and configures itself, and your job is to review what it learned and approve it. When it later gets something wrong, you don't rewrite a prompt — you tell it the correct answer in one plain sentence, the way you'd correct a new employee, and it holds that going forward. Training Maya is verifying, not programming.

Last updated 2026-07-16

Why "just write a good prompt" is the wrong job to hand a business owner

Prompt-and-flow tools quietly move the work onto you. To make them useful you have to anticipate every question, phrase each answer, wire the branches, and then maintain all of it as your business changes — a skill most owners never signed up to learn. So the tool ships half-configured, answers stiffly, and the person who was supposed to be freed up is now debugging an intent tree at 10pm. The setup burden is the reason so many AI receptionists sit switched off.

Maya starts from the opposite end. It configures itself from the website you already have, then hands you the far smaller, far more natural job: check that it's right, and correct it in a sentence when it isn't. Replacing configuration with verification is the whole shift. It's the difference between building an employee from a blank manual and hiring one who already read the company handbook on day one — you're not writing their brain, you're just telling them the few things the handbook missed.

How you actually train Maya

It sets itself up from your site — so there's nothing to write on day one — Maya reads your public pages, catalogue, structure and connected business data and configures itself — no prompt, no FAQ document, no intent list. You don't start at a blank canvas; you start at a Maya that already knows most of your business and just needs checking.

You review what it learned — and approve it before it talks to a single customer — instead of trusting a bot you configured blind, you inspect what Maya understood about your business and sign off on it. Nothing goes live to a visitor until you've verified it's right. Setup becomes an act of reading, not authoring.

You correct it in one sentence — not by editing a prompt — when Maya answers something the wrong way, you tell it the right answer the way you'd tell a new hire — "we don't do same-day, but we do next-morning" — and it holds that. No syntax, no variables, no flow to re-wire. If you can brief a person, you can train Maya.

It flags its own gaps instead of guessing — so you know exactly what to teach — rather than you having to imagine every question in advance, Maya tells you which real questions it couldn't confidently answer. Your training list writes itself from actual customer conversations, so you only teach what's genuinely missing.

It shows you what customers kept asking that you don't offer — the same loop that trains Maya also tells you something about the business — every week it surfaces the requests your visitors made that you don't yet serve, so "training the receptionist" quietly doubles as market feedback.

It keeps reading your site as it changes — so training isn't a one-time chore — when you update a page, add a service, or change a price, Maya keeps reading your public site, so the answers track your business instead of drifting stale the way a hand-written prompt does the day after you write it.

What happens when Maya doesn't know the answer?

This is the mechanism that makes prompt-free training safe. Maya checks its answers against your approved business information before it responds. When it can't verify something, it doesn't improvise to fill the gap — it says so, offers the visitor the right next step, and flags the question for you. That flag is the training signal: it's Maya telling you, in plain terms, exactly what it doesn't yet know. You either mark it correctly handled, or teach it the answer in one sentence.

What that means for you: you never have to pre-write answers to questions you didn't know people would ask. Maya surfaces the real gaps and you fill them one sentence at a time — so it gets more useful every week without you ever being rewarded, or punished, for how well you can phrase a prompt.