In short
- Checked before it answers.OneType measures how well your content matches the question before the AI writes a word.
- Your next step.Show a contact form, just apologize, or hand the chat to a person.
- Your own words.Write the message visitors see when the agent cannot answer.
- Uses no AI credits.A reply that says it does not know never counts as an answer.
The worst thing a support agent can do is make something up. A wrong shipping date or an invented refund rule costs you more than no answer at all. OneType is built to stop before that happens: when your content does not answer a question, the agent says so and helps the visitor move on.
A check before every answer
Before the AI writes anything, OneType searches your content and scores how well the best passages answer the question. If nothing scores high enough, the agent does not try. There is no model call to talk itself into an answer. The visitor gets an honest reply and a way forward.
There is a second check for products and names. If a visitor asks about something your content never mentions, such as a grinder when your pages only cover kettles, the agent treats it as a question it cannot answer, however similar the wording looks.
And even when the check passes, the AI has its own way out. It is told that if the sources do not contain the answer, it must say so rather than guess.
Choose what happens next
In the agent's settings, pick what the agent does when it cannot answer:
- Escalate to human (Premium, with handoff turned on): the visitor is offered a person from your team, and a second unanswered question in a row hands the chat over automatically.
- Collect contact details: the visitor sees a short form to leave their details.
- Apologize: the agent says it could not find the answer and the chat goes on.
In your words, in their language
With Collect contact details or Apologize, you can write the message the agent shows when it cannot answer, up to 500 characters, for example "I'm not sure about that one. Want me to connect you with our team?". If you leave it empty, the agent uses a built-in message in the visitor's language, available in 11 languages.
Along with the message, the agent can show a few topics it does know about, taken from your content, so the visitor sees what they can ask instead.
Small talk is not a failure
A greeting, a thank you or "what can you do?" never gets an "I couldn't find that". The agent answers warmly, describes what it can help with, and suggests a few questions to ask.
No AI credits used
Only real answers use AI credits or, on own-key plans, count as answers. Replies saying the agent does not know, handoffs and answers stopped for abuse are free.
What it cannot promise
The check is careful but not perfect. A question your content does cover, asked in very unusual words, may still get an "I don't know", and no system can promise zero wrong answers. The confidence bar is set by OneType and is not a setting you can change today. When you see a question go unanswered, add a saved answer and the next visitor gets it.
What each plan includes, and how AI credits work, is on the pricing page.