The chatbot that says when it does not know

Last updated: 2026-08-22.

Most website chatbots will improvise an answer when they are unsure. That is how a bot ends up inventing a price, a policy, or a promise the business never made. Site Rep is built the opposite way: it answers only from your approved pages, shows the source under every answer, and asks for follow-up when source backing is not there.

The operating rule

Every answer is grounded in your own content, with the source shown.
When your pages do not cover a question, Site Rep says so and offers to take the visitor's email — it does not guess.
An unanswered question becomes a private follow-up item, and can become a lead when the visitor leaves contact details.

We test this, and you can see the result live

Site Rep is checked with a fixed set of questions against the live demo bot's real sources: questions it should answer, and questions it should hand off instead of guessing. The result keeps the behavior reproducible against the public demo.

Live right now: 19 of 19 should-answer questions passed, 7 of 7 should-hand-off questions passed, and 2 of 2 pricing answers match the live checkout. These counts come from the same shared honesty evals as the machine-readable check result, so they always match what that check reports.

It is not a marketing number. You can try the same questions yourself in the live demo and see when Site Rep cites a source or asks for follow-up.

Questions it must answer

These are covered by the demo sources, including messy real phrasings:

What does Site Rep cost?
how much is the starter plan
im on wix will this work
how do i sign up and get started
ok how do i cancel if i dont like it

Questions it should hand off

These are off-topic or adjacent-but-unsupported, so Site Rep should ask for follow-up instead of guessing:

can you file my taxes
do you integrate with Salesforce
are you SOC 2 certified
what is the capital of France

Pricing accuracy

AI assistants get SaaS prices wrong more than half the time — a 2026 audit found 54.5% of AI pricing claims were wrong against the vendor's own live pricing page (Automate to Profit, August 2026). Site Rep's pricing answers come from the live Dodo checkout preview, not from model memory. The honesty check now verifies that the demo bot's named-plan prices match the live checkout amounts.

How answers are generated

Site Rep phrases facts already written on approved pages, cites the source page, and asks for follow-up when source backing is not there.

Why this matters: your business is liable for what your chatbot says

Courts have already held businesses liable for false statements their AI chatbots make to customers. In 2024, the British Columbia Civil Resolution Tribunal found Air Canada liable for false bereavement-fare advice its chatbot hallucinated (Moffatt v. Air Canada, 2024 BCCRT 149) — the airline could not deflect responsibility by blaming the AI. In May 2026, the Higher Regional Court of Hamm (Germany) reached the same conclusion (Az. 4 UKl 3/25): a company is responsible for what its chatbot tells customers, and "the AI made a mistake" is not a defense.

This is the risk Site Rep's source-backed approach is built to reduce. Because Site Rep answers only from your approved pages and hands off when source backing is missing, it cannot invent a price, a policy, or a promise you never published. The cite-or-refuse behavior tested on this page — 19 of 19 should-answer questions answered from sources, 7 of 7 off-topic questions handed off instead of guessed — is the same guardrail that keeps the bot from making claims your business would then be liable for.

This is not legal advice. Talk to a lawyer about your specific situation and jurisdiction.

Try it yourself

Open the live demo and ask any question above, or your own. Watch it cite a source when source backing exists, and ask for follow-up when it does not.