Trust
AI disclosure
How the AnswerTide assistant answers, when it passes a conversation to a person, and what it tells people on each channel. Last updated 4 October 2026. Examples are shown for a business named Northwind.
How the assistant answers
- It answers only from content the business approves: pages imported from its site, files and documents it adds, and questions and answers it writes. It doesn't answer from general knowledge.
- Every answer cites the page it came from. Every link is checked against the pages imported for that business; an answer with a link that isn't one of them is withdrawn.
- When no imported source supports an answer, the assistant doesn't guess. It replies with a fixed sentence that the AI model never writes:
That isn't covered in Northwind's help content yet. Want me to pass this to a person on the team?
- Answers a business corrects or pins are shown word for word. The AI doesn't rewrite them.
- We don't train AI models on a business's content or its customers' conversations.
Reaching a person
- Every chat has a “Talk to a person” button, on every plan including the trial. The visitor leaves an email address and the business's team gets the whole conversation.
- Asking for a person in the chat, in plain words, does the same.
- If the assistant is paused, for example because the business reached its monthly limit, visitors can still leave a message for the team.
The disclosure sentence
Where a channel shows our disclosure, it is this sentence, with the business's name:
Northwind uses an AI assistant to reply. You can ask for a person at any time.
It reads the same on every channel that shows it. A business can't reword or remove it; only its name changes. When a business turns on no-patient-details mode, every channel adds this notice right after it (on the phone, right after the greeting):
Please don't share medical details here, such as dates of birth, insurance or member numbers, or medications.
What each channel shows
| Channel | What people see or hear | Status |
|---|---|---|
| Chat on the business's website | The business's name with “Answers from this site's help content”, the sources under each answer, the “Talk to a person” button and the “Answered by AnswerTide” badge (paid plans can hide the badge). The disclosure sentence appears under the greeting when no-patient-details mode is on; showing it in every chat is planned. | In use |
| Browser voice (Talk), inside the chat window | The same chat window and header. Spoken answers follow the same rules: cited content or an offer to get a person. A spoken disclosure sentence is planned. Voice is off until the business turns it on. | In use |
| Help-center search box | The answer with its sources, and a “Contact us” button that reaches the team. | In use |
| Phone (Voice receptionist) | Before anything else, every caller hears: “Hi, you've reached Northwind. I'm an AI assistant. I can answer questions from Northwind's information, book an appointment, or take a message for the team.” If the business turns call recording on (it is off by default), the next sentence is “This call is recorded.” | Planned |
| Email replies (Inbox agent) | AI-drafted replies end with their sources, the disclosure sentence and a way to opt out. Drafts wait for a person to approve them unless the owner turns on automatic sending. | Planned |
| Sales follow-up emails and quotes | The disclosure sentence at the end of each email. A person approves every quote before it is sent. | Planned |
| Text messages | The first message carries the disclosure sentence and “Reply STOP to opt out, HELP for help.” | Planned |
“Planned” channels are built but not switched on yet. Related: Trust and security · Security · Subprocessors