---
id: aikit-answers-setup
title: Set up Answers and live product information
sidebar_label: Answers setup
slug: /ai-kit/answers-setup
sidebar_position: 35
description: Start from an assistant preset, connect approved live catalog data, and guide visitors to your own WordPress product pages.
tags: [ai-kit, chatbot, wordpress]
---

Open **SmartCloud → Answers → Run setup**. Setup stages your changes until
**Finish setup**. Existing detailed settings remain available and are preserved.

## Start with your business

Choose **Site guide**, **Services and enquiries**, **Product discovery**, or
**Customer support**. Enter your business name and describe what visitors should
achieve in one or two sentences. Audience and main topics are optional.

For a sports shop, topics might be “Snowboards, winter sports equipment, product
selection and care.” Describe your actual offering; do not enter sample prices
or claim stock levels here. Live prices and availability belong to the catalog.

Select **Preview assistant profile**, review the suggested role, tone and tasks,
then apply the reviewed profile to your draft. A preset supplies behavior; it
does not connect a service or provide verified product facts.

## Choose the information sources

- **WordPress content:** select public content types for the knowledge base.
  Synchronization and backend indexing must finish before that content can
  support answers. See [Knowledge Sync](../knowledge-sync/).
- **Live catalog:** choose an optional supported catalog connection.
- **Approved web pages:** add specific public HTTPS sections that the assistant
  may read. This is page retrieval, not unrestricted web search.

These sources complement each other. WordPress can provide usage guides and
detailed descriptions while the external catalog supplies current product data.
When setup proposes an evidence-policy change, review its effect before approving.

## Connect a Shopify catalog

This guided connection uses the Shopify UCP catalog. For the separate typed
Storefront GraphQL adapter, inventory-aware searches and visitor-confirmed
shipping estimates, follow the [Shopify Storefront setup guide](./shopify-storefront.md).
That guide also covers Shopify apps, permissions, publication and filter setup.

1. Select **Shopify product catalog**.
2. In Shopify administration, open **Settings → Domains** and copy the original
   `your-store.myshopify.com` domain. Enter the domain only, without `https://`,
   an admin path, or a product path.
3. Optionally enter a two-letter market code and a fallback language tag.
   Visitor-language support sends the selected reply language; it does not
   change market or currency. Catalog translations must exist in Shopify.
4. Keep the **Public catalog agent profile** filled from your configured AI Kit
   backend. An override is for an independently hosted compatible profile.
5. Select **Check connection and available reads**. Review the server identity,
   endpoint and operations, then explicitly approve read-only access.
6. Add the reviewed connection to the draft. It is saved with **Finish setup**.

This connection check lists supported schemas. It does not execute a product
search, save a profile, or change the store. A connection already configured in
detailed settings is recognized, so setup will not add the same endpoint twice.

No catalog JSON export is required. Advanced MCP JSON describes reviewed tool
names, schemas and policies, rather than exporting products. Its remote schemas
come from the service, and the site owner supplies the purpose and approval.

## Send visitors to your WordPress product pages

A compatible backend can resolve a catalog product to your own public page.
The original product ID, price, variants and availability remain catalog data.

Choose a link policy for the connection:

| Policy | Destination |
| --- | --- |
| Original store links | The approved catalog product URL |
| Prefer WordPress pages | Your matching page, otherwise the original product URL |
| WordPress pages only | Your matching page; no external product link when unmatched |

Search the live catalog and select a product. Search your published WordPress
pages and select the page dedicated to that product. Review and add the pair to
your draft; you do not need to look up the external product ID yourself. Finish
setup or Save profile saves the rows.

Only published, public pages without a password are offered. Selecting a page
does not create or publish content. A collection page that lists products does
not automatically become a dedicated page for each one.

An optional language tag identifies a localized page. Resolution uses an exact
visitor-language match, then its base language, then a row without a language.
Explicit pairs take precedence over reviewed integration suggestions. You can
store up to 50 rows per connection.

Some installed integrations can supply verified product identities. Review their
page suggestions before adding them. Other integrations use the manual picker;
recognition is not assumed from a product name, SKU, generic block ID or URL slug.
The presence of WooCommerce page identities does not itself enable a live
WooCommerce catalog connector.

Review saved pairs after moving or unpublishing a page. Discovery does not
automatically rewrite the profile. If several sites share one backend profile,
choose the public site intended for those clients.

## Test before enabling visitor answers

Try a known product, its price and variants, a missing product, and a comparison.
Repeat in the visitor languages you expect. Inspect the source references and
local page links, including an unmatched product under each selected policy.

“Orderable” does not necessarily mean immediately in stock. The assistant should
not invent quantities, unavailable attributes, or variants missing from a bounded
result. Missing catalog information should remain explicit.

These settings enable read-only discovery and product-page links. Cart,
checkout and order changes are not enabled.

## Configure approved tools in detailed settings

Open **Conversation profile → Tools** for approved REST reads, MCP connections
and tool workflows. MCP is a connection/tool protocol; UCP describes commerce
operations a compatible service exposes. Neither name alone grants permission
or proves a particular inventory, shipping or payment capability.

For an MCP connection, review its server, HTTPS endpoint, advertised tool names,
input schemas and intended purpose. Enable only the required reviewed tools.
Discovery is not a product search, and saving a connection does not install a
Shopify app. Authentication must use owner-configured runtime references, not
credentials pasted into JSON. For the distinct fixed Shopify adapter and its
merchant-side setup, follow [Shopify Storefront](./shopify-storefront.md).

Tool workflows connect approved reads within one assistant turn. Inputs describe
values the assistant may supply. Steps select approved tools and explicitly
declare prerequisites and argument bindings; list order alone is not a dependency.
Independent branches may run in parallel, while failed or missing prerequisites
prevent dependent reads. These workflows do not enable arbitrary scripts or
purchase/booking actions, and do not require the Flow plugin.

## Configure source-backed display templates

In **Display templates**, choose **Add template**. The creation dialog asks for a
**Starting preset** and **Approved source**; these are creation choices, not list
filters. The preset supplies starting mappings/layout and is not a persistent
business rule. Existing custom templates need not retain a preset identity.

Approved sources can include structured WordPress content and approved external
REST or MCP reads enabled in the same profile. A Shopify source appears only
after its applicable read tool is configured and enabled. If structured WordPress
content is the only approved source, that is the only choice. Select compatible
approved WordPress content types to share one template across those types;
this does not authorize arbitrary reads or combine incompatible external schemas.
Review overlapping source/type bindings instead of creating competing templates.

Configure the heading, layout, maximum items and verified-field mappings. The
field picker offers known pointers for the selected source; a custom JSON Pointer
is advanced configuration, not an expression or code. For a structured WordPress
record, `/entityRef` identifies the record, `/title` supplies its public title,
and `/fields/summary` refers to an approved mirrored field only if that mapping
exists. Do not guess field names from their visible WordPress captions.

Use **Ordered attributes** and **Ordered navigation links** to add and arrange
compact entries. Open an entry to edit it; several entries may remain open.
Navigation can use a verified URL or an approved fixed destination and explicit
conditions. Displaying an appointment or product link does not book or buy.
Missing source facts stay absent; a label such as “Availability” cannot establish
live appointment slots or stock on its own.

**Approved read arguments for verified sample (JSON)** contains the inputs to
one real approved read, not pasted response data. A mirrored-record preview needs
an actual `entityRef` and its `expectedSourceVersion`; an external read uses its
approved schema. Run **Read verified sample and preview** to check mappings.
The preview does not save the template or grant further tool permissions.

For declared HTML/Markdown fields and restricted rich-text opt-in, see
[Knowledge Sync](./knowledge-sync.md#review-structured-fields-and-text-formats).
Card and navigation labels use the shared translation catalog in the chatbot UI
locale; authored record values and the answer text are not automatically rewritten.

## Understand nested drafts and confirmations

Compact lists cover template mappings, tool-workflow inputs/steps, suggested
visitor actions and external tools. Opening or closing an unchanged item should
not create an unapplied change. Closing a changed editor and **Cancel** use the
same discard check; confirmation is needed only for actual unapplied edits.
Removal requires confirmation and affects the enclosing draft, not source data.

Apply an edited item to its enclosing draft, then apply the complete template or
workflow to the profile draft. Finally choose **Save conversation profile** to
activate the configuration. Applying a nested draft is not a backend save.
Cancel discards only the current unapplied editor changes; it does not undo edits
already applied to the enclosing draft.

Where a list header exposes **Enabled**, toggling it updates the enclosing draft
independently; canceling a clean item editor does not undo that toggle. A REST
tool's enabled switch is inside its editor instead: it belongs to that item's
draft and must be applied or discarded with the other tool edits. Neither case
changes the live profile before the final profile save.
