AI chatbot: AnswerHatch
AnswerHatch is an AI chatbot for websites. It answers only from the pages you publish, shows the page behind every answer, and declines and says why when it has no source.
At a glance
- To install
- 1 script tag
- Nothing to install on your server
- Questions kept for
- 30 days
- Then deleted
- Free trial
- 14 days
- Up to 250 pages, no card
The problem
Visitors can't find answers your site already gives
It cites, or it declines.
AnswerHatch
Your site already answers most questions. People still ask things a website doesn't say directly: which form, which deadline, which address, which fee. They give up and email or call, and your team repeats what the site already publishes.
The usual fix is an AI assistant. General-purpose assistants can invent an answer, and an invented answer about your organisation is a liability you will be asked to explain.
What it does
Answers from your pages, or a reasoned refusal
AnswerHatch reads the pages you publish and answers visitors' questions from them in plain language, at your site's reading level and in your site's words. Answers can link to your forms and help pages.
A reader can check any answer in one click, and every response takes one of two shapes.
See the two shapes
What happens when AnswerHatch has no source for a question?
AnsweredAn answer ships only with registered sources attached. With no source, the widget declines and gives the reason.
Source shown with the answer: answerhatch.com
What is the best pizza in Chicago?
DeclinedI do not have a source for that. My sources are AnswerHatch's own published pages, and that question sits outside them.
Reason shown: no registered source
Questions and wording are taken from answerhatch.com. The live demo (opens in a new tab) gives real answers.
See it in the live demo
Captured from answerhatch.com on 9 November 2026.
How it works
Controls sit outside the model
AnswerHatch puts its controls in code that runs before and after the model, so a prompt cannot talk the system out of a rule.
Crawl and register
With your approval, and once you've proved control of the domain, your public pages are crawled. Each is recorded with its URL, fetch date and hash.
Screen the question
A scope screen and a prompt-injection screen run before the model is called, and are built to turn away off-topic and adversarial requests.
Retrieve
The question is matched only against the pages crawled from your site and held in the registry.
Cite or refuse
An answer ships only with registered sources attached. With no source, it declines and gives the reason.
Test, then go live
A test run on your own pages checks what it must answer and what it must refuse. You promote to production when you accept the results.
What the AI does
- Writes the answer in plain language from your registered pages.
- Matches your site's reading level, from plain English to statutory language.
- Answers spoken questions through the same pipeline as typed ones.
What it is built not to do
- Answer without a source: no registered page means no answer.
- Fill a gap: nothing plausible is composed to cover a question your site doesn't answer.
- Pass screened requests to the model: requests the screens turn away are refused before the model is called.
- Train on your content: models are not fine-tuned on your content or your visitors' questions.
As stated on answerhatch.com.
Inside the product
What you get, piece by piece
There is no feature gate between a small site and a large one. Every plan gets the whole product.
The product
The widgetOne script tag, keys bound to your domain
The widget loads from AnswerHatch's content delivery network and renders on your pages. There is nothing to install on your server and no rebuild of your site.
- Each key is issued against the domain you nominate. A key copied out of your page source does not serve answers anywhere else.
- Keys are stored hashed, not in plain text, and can be rotated from the portal.
- A consent and disclosure line appears when a chat starts, on every plan. It can't be switched off by accident.
The test runEvidence for acceptance, on your own pages
Every trial includes a test run on your own pages. The results are yours to read in the portal, page by page, and to hand to procurement.
- Pass list: the questions it must answer, each with the page it used. For example, the fees for a development application, where to report a missed bin, or opening hours.
- Refusal list: the questions it must refuse, with the reason shown. For example, jailbreak attempts, off-topic steering and requests for figures your site doesn't publish.
Preprod and prodTest before your visitors see it
Every subscription runs a pre-production and a production environment, each with its own corpus, keys and bindings. You test a crawl in pre-production and promote it when you accept it.
A crawl that fails verification produces a diagnostic. It never quietly becomes the live corpus.
The provenance registryA record a compliance officer can ask for
Crawled pages go into a registry of sources, derivations and decisions. The citations a visitor sees are a view of that registry, and your compliance officer can request the registry itself to audit exactly what the widget is allowed to say.
VoiceAsk out loud, hear the answer
A visitor can speak the question and hear the answer read back, sources included. For a visitor who can't read the screen, the whole exchange can run by ear.
- The Ask by voice button transcribes the question in the visitor's browser. Chrome and Edge use their vendor's speech service, and Safari transcribes on the device. AnswerHatch receives text, never audio.
- Every answer has a Listen button. A hands-free mode opens listening, reads the answer aloud and listens again for the next question.
AccessibilityKeyboard, screen reader and voice
- AnswerHatch states that the widget is built against the Web Content Accessibility Guidelines (WCAG) 2.1 AA.
- It can be operated by keyboard from end to end, with visible focus at every stop and no focus trap.
- Answers arrive in an ARIA live region, so screen readers announce them as they appear.
- It reads its own answers and works beside the screen reader a visitor already uses. It complements an accessible site rather than replacing the work of making one.
For government and larger sitesOptions quoted for your estate
- Hosting in a nominated region, with data residency.
- Identity federation and named user provisioning.
- A negotiated service level: 99.5% in business hours as standard, with round-the-clock cover as a priced option.
- An authenticated workspace, separate from the public widget, for derived products that were never published, such as forecasts, risk ratings and generated narrative.
What it processes
What AnswerHatch handles, and for how long
AnswerHatch is built for public content. Pages behind a login, case files and internal systems are out of scope for the public widget.
| Item | What is processed |
|---|---|
| Your published pages | The public pages of the domain you nominate, crawled with your approval and stored with their URL, fetch date and hash. |
| Visitors' questions | The text of the question and the page it was asked from. Nothing else about the visitor is needed. Questions are kept for 30 days, so a fault can be diagnosed and a refusal reviewed, then deleted. |
| Voice | Speech is turned into text by the visitor's own browser. Only the text reaches AnswerHatch. |
| Your account | The contact email, the domains on the subscription and the usage counts the plan is measured against. |
| Operational logs | Request records used to run the service, enforce rate limits and diagnose faults. |
Questions can be found and deleted on request: give AnswerHatch the domain and roughly when the question was asked.
Trust and security
Where the data goes, stated plainly
These statements come from AnswerHatch's published security and privacy pages.
Common questions
What if it says something wrong?
Every answer carries the page it came from, so a reader can check it before acting on it. With no registered source, the widget declines and gives the reason. Try it on the demo: ask something in scope and it answers with links; ask something out of scope and it declines.
Will our content end up in someone else's model?
No. Questions are processed on machines AnswerHatch runs, no third-party model API is in the path, and models are not fine-tuned on your content or your visitors' questions.
Where is the data processed?
AnswerHatch states that data is processed in a single hosting region, Australia, and that government and regulated buyers can nominate a region.
Our pages use formal or technical language. Will that work?
Yes. The crawl measures how each page reads, from plain English for a council to statutory language for an agency, and answers are written to match. You run the widget against a pre-production copy of your site and accept it only when the tests pass on your own pages.
Our IT and procurement teams will want evidence. What can we give them?
The test-run results from your own domain, the provenance registry of every page the widget may answer from, and written answers to your security questionnaire.
How is it priced?
After the free trial, pricing is quoted for your site, based on the pages you publish and the domains the widget runs on. You get a yearly figure in writing. Answers are fair use, with no per-answer meter and no credit balance to run down.
Get started
From trial to live in four steps
Start the free trial
Give an email address and the domain you will answer from. 14 days, up to 250 pages, no card.
Add your site
Paste one line of code into your pages.
Run the test run
Check the pass and refusal lists against your own pages in pre-production.
Go live
Promote to production when the tests pass.
Talk to Bluebird
Talk to us about AnswerHatch for your organisation.
contact@bluebirdadvisory.com.au
Related service: Digital and Cloud Transformation
Sources
Last reviewed: 9 November 2026.
- answerhatch.com: home, security, privacy, accessibility, pricing and government pages, as at November 2026.
- Bluebird AI Solutions page, bluebirdadvisory.com.au/bluebird-ai-solutions, as at November 2026.