DIGITAL INTERACTIVE
WordPress 7.1 Ships in 48 Hours: What wp_knowledge and AI Guidelines Actually Do
The Update Everyone Is Watching for the Wrong Reason
WordPress 7.1 releases in 48 hours — August 19, the closing day of WordCamp US right here in Phoenix. Virtually every roundup you’ll find this week is focused on the enforced iframed editor, the new Tabs and Playlist blocks, and responsive styling without CSS. Those are real and those ship. But the more durable technical change in this release barely makes the headlines: wp_knowledge, a new custom post type that becomes WordPress’s first native storage layer for editorial standards, brand voice, and AI memory.
If you maintain multi-author sites, run an agency with content guidelines stashed in a shared Google Doc, or are building anything that touches the AI Client added in WordPress 7.0, this one is worth understanding before you update.
What wp_knowledge Actually Is
The confusion most teams will hit right away: this is not a user-facing AI feature. No chatbot. No content generator. No model baked in.
wp_knowledge is a storage primitive. Think of it the way wp_template, wp_block, and nav_menu_item work — low-level content types that prevent fragmented parallel implementations across the plugin ecosystem. Right now, every plugin that needs to store AI context or editorial standards ships its own post type, its own permissions model, and its own REST surface. That is the fragmentation wp_knowledge is designed to close.
The merge proposal from core committer Greg Ziółkowski, published on Make/Core in June and confirmed for the 7.1 release, defines a non-public, headless, REST-only custom post type with a type taxonomy (wp_knowledge_type) and three built-in slugs:
- guideline — pure text, the authoritative source of truth for brand standards, tone of voice, and image direction
- memory — durable context explicitly saved or approved for future AI sessions
- note — private freeform working text, not surfaced to AI tools by default
The REST collection endpoint lives at /wp/v2/knowledge. Any plugin, WP-CLI script, or AI integration that knows to ask for it gets the same structured data without any custom implementation. Autosave is intentionally disabled because wp_knowledge is headless storage rather than a live editor session, but revision history is retained. Administrators manage site-wide entries. Contributors and authors only manage their own rows. Subscribers see nothing. The wp_maybe_grant_knowledge_caps filter makes the permission model extensible for teams with more granular access requirements.
One historical parallel worth noting from the Make/Core proposal: this is the same pattern WordPress used with the Abilities API and the AI Client in the 7.0 cycle — prototype in the plugin, harden through community feedback, then merge the storage layer into core. The names (wp_knowledge, wp_knowledge_type, the three type slugs, the capability namespace) freeze at 7.1 Beta 1 and become long-term compatibility commitments. That is the right moment to have argued about naming. That window is now closed.
Guidelines: What It Looks Like in the Editor
Guidelines is the first feature built on top of wp_knowledge. Site admins define editorial rules, brand voice requirements, and content standards in one place. Those guidelines then appear directly in the block editor as an editorial checklist for writers. The same data, over the same REST endpoint, is accessible to any AI tool configured to reference it.
Practical upside for agencies and multi-site operators: guidelines can be exported and imported between sites. If you manage a network of client properties that all need to follow similar editorial standards, that consistency now has a canonical home in WordPress rather than living in a PDF that nobody opens. WordPress.com engineer Artur Piszek built a Telegram agent for WordCamp Europe 2026 that ran entirely on Guidelines — every behavior including personality, schedule lookups, and user memory stored as a published guideline rather than custom application code. Desktop Mode, WordPress Workspace, and several other Automattic products already use the storage mechanism.
One thing confirmed in the RC3 build: the Guidelines feature ships with no connection to any specific AI provider. WordPress 7.1 includes no AI model and no pre-configured provider. AI-assisted drafting that follows your guidelines requires a connected provider through Settings > Connectors, identical to the 7.0 setup.
The AI Client Gets Streaming and Embeddings
Separate from wp_knowledge, WordPress 7.1 also updates the AI Client with two new primitives: generation streaming and embeddings support.
Streaming lets AI responses return text in real time rather than waiting for the full generation to complete. The Make WordPress AI team is explicit in the roadmap post that streaming in 7.1 is introduced at the PHP AI Client layer first. It requires server configurations that support long requests, which are not standard across hosting environments. The 7.1 implementation is the foundation. Hosts and plugin developers build production support on top of it in subsequent releases.
Embeddings are the bigger long-term play. An embedding is a numerical vector representation of content that captures meaning rather than keywords. The Make WordPress AI team’s own example is precise: a user searching “how do I reset my password?” can now surface a page titled “Account recovery steps” even though the two strings share zero words. Recent versions of MySQL and MariaDB added native vector storage support, and the 7.1 AI Client embeddings implementation is built to use it. For content-heavy WordPress sites with large archives, this is the first concrete signal that native semantic search is on the platform roadmap. The vector search experiment continues inside the AI plugin, but client-level embeddings support merging into core is the required infrastructure step.
What This Means for SEO and AI Optimization
The Guidelines system has a search visibility angle that most teams have not processed. Google’s current E-E-A-T evaluation places significant weight on publisher trustworthiness and consistency of content standards. When AI tools assist with drafting, inconsistent brand voice and factual drift are real risks to those trust signals over time. wp_knowledge gives you a single source of truth that both human editors and AI tools draw from during content creation.
The secondary dimension is GEO — generative engine optimization. AI crawlers that index WordPress sites increasingly surface structured, authoritative summaries as direct answers in AI Mode and AI Overviews. A site whose content is governed by explicit, structured editorial guidelines stored in a format that AI tools can retrieve is producing a more coherent, consistent signal profile than one where brand guidance lives in a Slack channel. Whether that translates to measurable citation improvement in AI search surfaces is not yet proven at scale, but the architectural alignment with how AI indexing systems work is sound.
What to Do Before and After Updating
The block compatibility testing guidance from earlier this month still stands. Test on a staging copy before pushing to production, especially if you’re running blocks built on Block API version 2 or earlier. The enforced iframed editor is the most common source of visual regressions in pre-production testing.
Specific to the AI features in 7.1:
After updating, check Settings > Connectors to confirm your AI provider connection survived the upgrade. If you’re on a managed host, confirm whether PHP streaming support is in scope for your plan — the streaming implementation requires server-level long-request support that shared hosting environments often cap. Embeddings require both a provider that supports embedding generation and, depending on your database configuration, vector storage support at the MySQL or MariaDB layer.
The Guidelines admin UI ships via the Gutenberg-side work in the block editor. Navigate to your site editor after updating to verify it appears and that your user roles have the expected access levels. If you are building a plugin that stores any kind of AI context or editorial data, the core team’s guidance is unambiguous: build on wp_knowledge rather than shipping your own post type. The names, capability namespace, and REST surface freeze at the 7.1 branch and become long-term compatibility commitments that will be maintained.
Frequently Asked Questions
Does wp_knowledge require an AI provider to be useful?
No. The storage layer works independently of any AI connection. Human editors see guidelines as an editorial checklist in the block editor without any provider configured. The AI-assisted drafting that references those guidelines does require a connected provider through Settings > Connectors.
Will my existing editorial guidelines in Google Docs migrate automatically?
No automated migration exists. You will need to create guideline entries manually in the new WordPress interface, or build a WP-CLI script that reads your existing documentation and writes to the /wp/v2/knowledge endpoint. The export/import feature lets you move guidelines between WordPress sites once they are in the system.
Is embeddings support ready for production use in 7.1?
Not yet at the host level. The Make WordPress AI team added embeddings support to the PHP AI Client as an initial release to let developers and hosts begin building support. Full production use requires vector storage at the database layer and a provider that generates embeddings. The vector search experiment in the AI plugin is the closest working implementation available today.
Does the wp_knowledge data get indexed publicly by search engines?
No. The post type is non-public and headless. It generates no frontend output and no public queries. Search engines cannot crawl it directly.
