Make your business easy for AI assistants

More and more people ask an AI assistant instead of searching. With agent-ready switched on, your website assistant's public page tells search engines and AI assistants what your business does and what people ask it, and an AI assistant can ask your avatar a question on someone's behalf and get a written answer, with the same rules your avatar follows in a typed conversation on your page.

You need: a published portal with a website assistant, open to visitors. Asking needs no key.

1. Switch it on#

An owner or admin switches on AI assistants may read your FAQ and ask your website avatar in Settings, under Features (it comes with the Starter plan and up), then opts each public page in: on the portal's Developers tab, or, for an avatar's website widget, on the avatar's own page. Nothing is published until both are on. Switching either off stops the answers within a minute and takes the documents down within five.

It is never available for a portal for children or for a candidate's interview, whatever is switched on.

2. What AI assistants can read#

Structured data on your public page (schema.org, as JSON-LD): your organisation, the service your assistant gives, and the questions and answers you published (a FAQPage).
Your organisation's llms.txt, at https://app.zentiqai.com/v/<org>/llms.txt: each opted-in page, what it helps with, how to ask it, and its questions and answers.
An agent card for each page, at https://app.zentiqai.com/v/<org>/<page>/agent-card.json, and at /.well-known/agent-card.json on the page's own verified domain: its name, what it does (its skills) and the address to ask it.

Only what your page already shows visitors goes out: your organisation's and the page's names, what it helps with, and the questions and answers you published. Email addresses, phone numbers and payment or identity numbers are taken out of all of it.

3. Ask a question#

Shell

curl -X POST https://app.zentiqai.com/v1/agents/acme/front-desk/ask \
  -H "content-type: application/json" \
  -d '{"question": "Are you open on Saturdays?"}'

{ "data": { "answer": "Yes, we're open on Saturdays from 9am to 1pm.", "url": "https://app.zentiqai.com/v/acme/front-desk", "aiGenerated": true }, "error": null, "requestId": "req_…" }

One question, up to 800 characters. Each answer stands on its own: there is no conversation, nothing is stored, and no details are taken.
url is your assistant's own page, where a person talks to it, books or reaches your team. The answer points there whenever a person is needed.
aiGenerated is always true: say the answer was written by an AI when you pass it on.
Email addresses, phone numbers and payment or identity numbers are removed from the question before your assistant reads it, and from the answer. Leave people's names out of questions.

4. Limits and errors#

Ten questions a minute from one caller to one page, 60 an hour from one caller, 600 an hour for one page and 3,000 a day for one organisation: beyond them, 429 rate_limited with Retry-After.
404 not_found for any page that isn't opted in, or doesn't exist: the answer is the same either way.
503 capacity when no answer could be written just then: try again shortly.

Good to know#

The answers follow your assistant's own rules for written replies: what it helps with, your documents and knowledge, your opening hours and the safety rules. It never takes a booking or a message here: it points to your page instead.
It is always truthful about being an AI. The answer is marked as AI-generated, and your page says so to every visitor.
Nothing about a question is kept: not its words, not the caller's address. Only short-lived counters for the limits.
Your questions and answers come from your website assistant's own FAQ, as published: change them there.