Ask it something it can’t answer
This is not a general-purpose assistant
It knows your business and nothing else. Ask it about the weather, a competitor or last night’s football and it will decline. That is the design, not a limitation we are working on.
Trained on your content
Trained on your content.
Nothing else. And it
says so when it’s stuck.
A concierge should know one business properly, not the whole internet vaguely. This one reads your pages, your documents and your approved answers, and when the question falls outside them it says so and takes a callback.
20 minutes, on your own website, no slide deck. First month free.
- Your pagesCrawled and kept current
- Your documentsPrice lists, policies, service descriptions
- Approved answersWritten by a person, always win
Out of scope, so it declines
- The weather
- A competitor
- Last night’s football
YDO is the concierge for your business
Move faster
Answered in seconds, alerted in minutes, signed before dinner.
Better conversations
Answers from your own content, remembers people, and hands over to a person when it matters.
Straight into sales and marketing
Every conversation becomes a scored lead, credited to the ad that produced it and reported every month.
Simple outcomes
One line of code, one record per person, and a person always clicks send.
- production deployments
- 8
- sectors
- 7
- line of code
- 1
- built, replies in Sydney hours
- AU
Running inside
Why this page exists
Twelve competitors. Not one of them says where the answer came from.
We audited the twelve chat products an Australian buyer is most likely to shortlist. Every one of them sells accuracy. None of them publishes a page explaining what the assistant does when it doesn’t know.
That silence is the whole gap. A generic assistant generates a plausible answer about your delivery fees, your stock or your refund policy, and you find out from the customer who acted on it.
We know what’s stopping you.
Here’s each one, answered properly.
“What if it gives a customer the wrong answer?”
It answers only from your content and says so when nothing matches. Real questions are re-run after every change and a slip flags the same day. It is designed to refuse rather than guess.
“Will it sound like a robot?”
It speaks in your voice, set up during onboarding, and hands to a person the moment a conversation needs one. It never pretends to be human and it never performs being an AI.
“Who owns the data and the answers?”
You do. The approved answer library, the lead records and the signed documents are yours, and they leave with you.
“How long until it’s live?”
Weeks, not quarters. One line of code, and it is answering from your content the afternoon you hand it over.
How the grounding works
Three mechanisms, not a promise.
Your content is the only source
Your website, your documents and the answers a person has approved. Nothing else is in scope, deliberately. It knows one business and nothing else.
A refusal is an outcome, not a failure
“I don’t have the delivery fees for Berwick. I can have someone call you with them.” It names what it doesn’t know and offers the next step in the same breath.
Drift is caught the same day
Real questions are re-run after every content change. A slip flags red that day, rather than at the next quarterly review nobody schedules.
Try the refusal
The seeded prompt that makes it admit it doesn’t know.
Most vendors would hide this. It is the most persuasive thing on the page, so it stays in the demo.
Harvest Lane Grocers
Concierge, answering from your pages
Visitor: Do you deliver to Berwick on a Saturday?
Concierge: We deliver to Berwick Monday to Friday. Saturday runs cover the inner east only. The full list is on our delivery page.
From your Delivery page
Visitor: What would it cost?
Concierge: I don’t have the delivery fees for Berwick. I can have someone call you with them. What’s the best number?
That refusal is seeded on purpose. It is the behaviour you are buying.
Made-up store. Harvest Lane Grocers is fictional and labelled as such wherever it appears.
What buyers actually complain about
Four independent review bases. The same failure, in the customers’ own words.
Hallucination is not a theoretical risk people worry about before buying. It is what paying customers write about afterwards, by name, on the platform where it happened.
The problem, in their words
“The bot WILL generate a totally wrong answer with greatest eloquence and confidence.” · “It sometimes makes up URLs.”
Capterra, paying customer of a document-trained chatbot builder. Reproduced verbatim. How YDO resolves it
YDOanswers only from what you gave it, and says so when nothing matches. Eloquence was never the problem.
The problem, in their words
“How much it hallucinates and gives people incorrect answers.”
G2, 30 June 2026, on a major support suite’s AI agent. Corroborated by a second reviewer on 8 June 2026 and by customers of two further products. How YDO resolves it
YDOis the only product in the set that publishes what it does when it does not know. Zero of twelve competitor sites have a page on answer provenance.
The problem, in their words
“You can’t just turn them on.”
G2, 29 July 2025, on an AI agent needing monitoring and fine-tuning before it can be trusted in front of customers. How YDO resolves it
YDOre-runs real questions after every content change and flags a slip red that day. The monitoring is the product’s job, not yours.
The problem, in their words
The weekend build is genuinely possible. The nine months after are the problem.
Category finding from the same analysis. Public, dated incidents exist where an unsupervised assistant invented policy and it cost the business. How YDO resolves it
YDOis the version still working in March, maintained by someone whose job it is. And where you do need something bespoke, we build that too.
Source: YDO competitor sentiment analysis, 18 September 2026, from public review platforms and vendor community forums. Every row here describes the category, not a named vendor.
What you drag in
Setup is content, not configuration.
Your existing pages, crawled and kept current.
Your documents: price lists, policies, service descriptions. Drag in a new price list and it is answering that afternoon.
Approved answers written by a person, which always win over anything inferred.
What you get back
The library is the asset.
Every question it couldn’t answer comes back as a card with a draft answer for a human to approve.
Nothing reaches a live customer without someone clicking. The AI drafts. People decide.
After a year you own an approved answer library covering what your market actually asks, and it is yours, not ours.
Before you go further
Two things worth knowing now rather than later.
“Trained on your content” means your content, not custom software
If you need an agent that reads and changes records inside your own platform, that is a different conversation and a different build. This page is about a concierge that answers.
Why it holds up
Grounded, re-tested, and owned by you.
Re-tested after every change
Real questions are re-run after every content change; a slip flags red that day.
One line of code
Six ways to install, no “powered by” badge, and no internal tech team required.
A supplier outage costs one feature
Not a dead chat box on your home page.
Questions we get asked
Straight answers.
Can it be trained on our PDFs and price lists?
Yes. Drag them in and it is answering from them that afternoon. When a document changes, the answers change with it.
What happens to a question it can’t answer?
It says so, offers a callback, and the question comes back to you as a card with a draft answer for a person to approve.
Does it ever make something up?
It is built to refuse rather than guess, and accuracy is re-run after every content change. We will not claim it is always right. That is a different promise and it would not be true.
We could build this ourselves.
You could. The nine months after are the problem: accuracy drift when the price list changes, the handover that silently stops working, the reporting nobody rebuilt. And if you need something genuinely custom, we build that too.
See it on your own website.
Twenty minutes, on your own pages, no slide deck. Bring the question your customers ask that you dread being answered wrongly.
20 minutes, on your own website, no slide deck. First month free.