Most WordPress sites contain a useful operational knowledge base before an agency adds a single AI feature. The information is distributed across posts, pages, products, taxonomies, custom fields, order records, media, menus, comments, and plugin tables. The first step in building an AI harness is not generating content. It is identifying which of that information is trustworthy, searchable, current, and safe to expose.
WordPress already contains several kinds of knowledge
Editorial knowledge
Pages and posts commonly contain product explanations, installation instructions, policies, service descriptions, FAQs, troubleshooting steps, and campaign content. Each item also has useful structure: a title, excerpt, author, publication date, modified date, permalink, status, post type, taxonomy terms, and sometimes custom fields.
The modified date is particularly valuable. An AI-powered search experience should be able to distinguish a current returns policy from an older article that happens to contain similar wording. Agencies should preserve that date in the index and use it when ranking or displaying results.
Commerce knowledge
WooCommerce adds a structured product catalogue. Products and variations can include prices, stock status, SKUs, attributes, categories, tags, downloadable files, dimensions, shipping information, related products, and purchase restrictions. This information is more reliable for product discovery than asking an AI model to infer details from marketing copy.
Orders and customer records contain a different class of information. They may reveal purchasing history, refunds, billing details, shipping details, and customer notes. This data can support tightly controlled internal workflows, but it should not automatically be indexed into a public search experience or sent to an external AI service.
Taxonomy knowledge
Categories, tags, product attributes, brands, and custom taxonomies express relationships that are often lost when content is treated as plain text. A product assigned to a “waterproof” attribute or a service assigned to a “small business” taxonomy term can be retrieved using that structured relationship, even when the exact phrase does not appear in the description.
Operational knowledge
Menus, reusable blocks, widgets, media metadata, custom post types, and plugin settings can explain how a site is organised. Some of this information belongs in an internal agency knowledge base rather than a customer-facing index. For example, a custom post type containing delivery zones may be appropriate for a support assistant, while an internal editorial note may be restricted to staff.
Build an inventory before building prompts
Start with a content inventory that records each source, its audience, its owner, and its update process. A useful inventory can include:
- Source: posts, pages, products, variations, FAQs, documents, or custom post types.
- Visibility: public, private, password-protected, draft, or restricted by membership or role.
- Structure: title, body, excerpt, taxonomies, custom fields, SKU, price, stock, and dates.
- Authority: whether the source is the official answer, supporting material, or an obsolete reference.
- Freshness: how often the source changes and which date determines recency.
- Destination: public search, authenticated search, an internal assistant, or no AI access.
In WordPress, the post type is a useful starting point, but it is not a complete security model. A public post can contain a private custom field, and a product page can link to a document that has different access requirements. Review the final rendered content, custom fields, linked files, and permissions together.
Use WPSOLR as the retrieval layer, not the source of truth
WPSOLR can help agencies make WordPress content available to a search index and retrieve relevant documents for search or AI-assisted experiences. The index should be treated as a derived copy. WordPress and WooCommerce remain the systems of record, while the index provides a fast way to find relevant content.
That distinction affects maintenance. When a product price, stock status, policy, or support article changes, the corresponding indexed document must be updated or removed. When a post becomes private, its indexed version must no longer be returned to users who cannot access it. Configure indexing rules deliberately instead of assuming every field should be searchable.
For each indexed document, preserve enough metadata to support reliable filtering and citations:
- The canonical URL and content type.
- The WordPress object ID and modified timestamp.
- Relevant categories, attributes, brands, and custom taxonomies.
- Language, site, store, or customer segment where applicable.
- Access or visibility information used by the application.
- Product identifiers such as SKU, variation ID, and stock state where appropriate.
Separate searchable facts from generated interpretation
Search should retrieve facts that WordPress already stores. Generation can then summarise those facts, compare retrieved products, or turn a support article into a concise answer. It should not invent missing stock levels, shipping promises, compatibility claims, or policy exceptions.
For example, a product assistant can retrieve products matching a taxonomy term, price range, attribute, and stock state before asking a model to explain the differences. A support assistant can retrieve the latest published article and include its URL. A sales assistant can use public product information while keeping customer order data behind authenticated, role-aware application logic.
Audit the data that should not be indexed
Before enabling indexing, exclude or isolate information such as customer names, email addresses, phone numbers, billing addresses, order notes, payment-related data, private editorial comments, API credentials, and internal implementation details. Check media libraries as well: a PDF or spreadsheet may be publicly reachable even if it is not linked from a page.
Also review plugin-generated content. Membership, bookings, subscriptions, help desk, and learning-management plugins may store sensitive records in custom tables or post types. A WPSOLR configuration that indexes all post types or all custom fields can unintentionally widen access to data that was never meant for public retrieval.
A practical agency workflow
- List every post type, taxonomy, custom field group, media source, and WooCommerce data set used by the site.
- Classify each source by audience, sensitivity, authority, and update frequency.
- Choose the fields that improve retrieval, such as SKU, brand, product attributes, policy type, or publication date.
- Exclude private records and fields unless an authenticated use case has been designed and tested.
- Configure WPSOLR indexing and filters around the approved inventory.
- Test searches using real customer and support questions, including obsolete-content and permission cases.
- Monitor reindexing after edits, product changes, status changes, deletions, and permission changes.
The result is a more dependable AI harness: WordPress supplies structured and current knowledge, WPSOLR retrieves the relevant subset, and the application decides what the requesting user is allowed to see. That division of responsibility is more reliable than treating the entire WordPress database as an undifferentiated prompt source.