WordPress 7.0 Beta 2 Adds an AI Settings Page. Here's What It Can (and Can't) Do.

WordPress 7.0 Beta 2 landed on 26 February 2026, one week after Beta 1, with 70+ updates and fixes. Most of those are bug fixes and polish. But the headline feature is new: a Connectors settings page at Settings > Connectors that lets you configure AI services directly from your WordPress admin panel.

Three connectors ship out of the box. OpenAI for text, image, and code generation with GPT and DALL-E. Claude for writing, research, and analysis. Gemini for content generation, translation, and vision. Each gets an "Install" button, and a single line at the top of the page sums up the pitch: "All of your API keys and credentials are stored here and shared across plugins. Configure once, use everywhere."

That sounds clean. But what does it actually mean in practice? Can you install these connectors and start generating content from your dashboard tomorrow? And if you're a web designer managing client sites, should you care?

The short answer: this is infrastructure, not a feature. And that distinction matters more than any AI settings page.

What the Connectors Page Actually Does

The Connectors page is a credential manager. It stores API keys for AI services in your WordPress database, then makes those credentials available to any plugin that needs them. Before this, every AI plugin brought its own settings page, its own key storage, and its own connection logic. If you ran three AI plugins, you entered the same OpenAI API key three times.

WordPress 7.0 centralises that. One settings page, one key per provider, shared across every plugin that uses the WP AI Client. The system is built on a PHP AI Client library (installable via Composer: composer require wordpress/php-ai-client) wrapped in WordPress-specific features: credential storage from the database, a JavaScript API (wp.aiClient) for the block editor, and role-based access control through a new prompt_ai capability.

Plugins hook into the page through a connections-wp-admin-init action. The route-based architecture uses @wordpress/components and @wordpress/admin-ui, so third-party connectors look and feel native. And there's a link to "Find more connectors in the plugin directory," pointing to a future where AI providers publish their own connectors as plugins.

The technical foundation is solid. Provider-agnostic by design. If someone builds a connector for a self-hosted LLM, Mistral, or Cohere, it plugs into the same infrastructure.

What It Can't Do (Yet)

Here's the part most coverage skips. Installing a connector doesn't give you AI features. It gives you a connection. The Connectors page is a keyring, not an application.

You won't click "Install" on the Claude connector and suddenly have AI writing your blog posts. You won't configure Gemini and get automatic image generation in the media library. The connectors store credentials and provide a standardised API. What happens with those credentials depends entirely on what plugins you install on top.

As of Beta 2, the WordPress core itself ships with minimal AI functionality. The AI Experiments plugin (a separate, optional plugin) adds excerpt generation and an Abilities Explorer screen. But the Connectors page in core is pure plumbing.

For web designers who've worked with the WordPress ecosystem for years, this pattern is familiar. WordPress has always shipped foundations and left the feature layer to plugins. Custom post types, REST API, block editor, Abilities API. The Connectors page follows the same playbook: build the infrastructure in core, let the ecosystem build the features.

That's not a criticism. It's the right approach. But it means the Connectors page in Beta 2 is a promise, not a product.

How Connectors Relate to the MCP Adapter

If you followed our Beta 1 coverage, you'll remember the MCP Adapter was the headline feature. It connects WordPress to external AI agents like Claude Code, Cursor, and VS Code through the Model Context Protocol.

The Connectors page solves a different problem. Here's how they fit together.

Feature	MCP Adapter (Beta 1)	Connectors (Beta 2)
Direction	AI agents reach into WordPress	WordPress reaches out to AI services
Who uses it	Developers, AI tools	Site owners, plugin developers
What it does	Exposes WordPress capabilities to AI agents	Stores AI API credentials for plugins to use
Requires	MCP-compatible AI client	API key from an AI provider
Visible in admin	No (developer-facing)	Yes (Settings > Connectors)

The MCP Adapter answers: "What can AI do with my WordPress site?" Connectors answer: "What can my WordPress site do with AI?" They're complementary, and together they position WordPress 7.0 as the first major CMS with bidirectional AI integration.

The Security Question

Centralising API keys in the WordPress database raises an obvious question: is this safe?

The prompt_ai capability restricts who can use AI features. By default, only administrators get access. Plugins can extend this to other roles, but it's locked down from the start. API credentials are stored as WordPress options, which means they're in the same database as your post content and user data. If someone breaches your database, they get your AI API keys too.

That's no different from how WooCommerce stores Stripe keys, or how SMTP plugins store email credentials. WordPress has always stored sensitive data in the database. But it does mean the usual rules apply: keep WordPress updated, use strong passwords, run a web application firewall, and limit admin access.

For agencies managing multiple client sites, there's a practical concern: each site needs its own API keys. You can't share a single OpenAI API key across 50 client sites without running up a bill the client didn't agree to. The per-site model is correct from a security standpoint, but it adds setup overhead for agencies.

OpenAI, Claude, and Gemini: What Each Connector Offers

The three built-in connectors aren't interchangeable. Each AI service has different strengths, and the connector descriptions make this clear.

OpenAI covers the widest range. "Text, image, and code generation with GPT and DALL-E." If a plugin needs to generate blog post drafts, product descriptions, and product images from a single provider, OpenAI is the only connector that handles all three today. API pricing starts at $0.15 per million input tokens for GPT-4.1 mini.

Claude focuses on quality. "Writing, research, and analysis." Claude Sonnet 4.6 outperforms GPT-5.2 on most writing benchmarks, and the 1-million-token context window means it can process an entire WordPress site's content in a single request. For editorial workflows, content analysis, or SEO audits, Claude is the stronger choice. API pricing starts at $0.80 per million input tokens for Sonnet.

Gemini brings Google's ecosystem. "Content generation, translation, and vision." The translation angle is worth noting; Gemini 3 Pro handles multilingual content better than either competitor, and the vision capabilities mean plugins could analyse images already in your media library. Gemini also offers the most generous free tier through the Google AI Studio API.

The real power will come from plugins that use multiple connectors. Route translation tasks to Gemini, writing to Claude, image generation to OpenAI. The Connectors page makes that possible without each plugin asking for separate credentials. Our model comparison guide breaks down which AI works best for which task.

What This Means for Web Designers and Site Owners

If you build WordPress sites for clients, the Connectors page changes your conversation. Not today, but soon.

For web designers: The standardised AI Client means you can build features that use AI without writing connection code for every provider. Register your plugin with the connections-wp-admin-init hook, call wp.aiClient from JavaScript in the editor, and the credentials are already there. The barrier to adding AI-powered features to client sites just dropped.

But the ecosystem needs time to catch up. Most existing AI plugins (Jeep AI, AI Engine, Jetvision) still use their own API connections. Until they adopt the WP AI Client, you'll have the Connectors page and the old per-plugin settings running in parallel. That transition will take months.

For site owners: Don't rush to get API keys from OpenAI, Anthropic, and Google just because there's a settings page for them. Wait for plugins that use the Connectors infrastructure and solve a real problem you have. Content generation? Translation? SEO analysis? When a plugin you trust adopts the WP AI Client, that's when the Connectors page becomes useful.

For UK businesses: The question isn't whether to connect AI to your WordPress site. It's how visible your site is to AI in the first place. WordPress 7.0 gives your site the ability to call AI services. But AI discovery files and structured data are what make AI services aware of your business when someone asks "Who should I hire for web design in Kettering?" Those are different problems, and both need solving.

So Is It Good, or Irrelevant?

Neither. It's early.

The Connectors page is a well-designed piece of infrastructure that solves a real fragmentation problem. Every AI plugin storing its own keys, building its own connection logic, creating its own settings page? That was messy. Centralising it in core is the right move.

But calling it "AI in WordPress" overstates what's there today. You can't install a connector and start generating content. You can't configure Claude and get automatic SEO suggestions. The Connectors page is plumbing. Good plumbing. Plumbing that the entire WordPress AI ecosystem will build on once the stable release ships in April.

The comparison I keep coming back to: WordPress shipped the REST API in version 4.7 (December 2016). For the first year, most site owners had no reason to know it existed. Then headless WordPress happened, mobile apps connected to it, Gutenberg was built on it. The REST API became the backbone of modern WordPress. It just took time for the ecosystem to catch up.

The Connectors page and WP AI Client might follow the same trajectory. Or they might not. WordPress has shipped developer foundations before that never gained traction (Custom Fields API, anyone?). The difference this time is that AI integration has $14 billion in venture capital behind Anthropic alone, and every major plugin developer is already building AI features. The demand exists. WordPress just gave it a standard interface.

For now, the honest assessment: impressive infrastructure, minimal immediate utility, and strong long-term potential. Check back in six months.

How to Test the Connectors Page

Don't install Beta 2 on a live site. Use one of these approaches.

WordPress Playground (no setup required): Visit playground.wordpress.net with ?php=8.0&wp=beta parameters. Navigate to Settings > Connectors. You can explore the interface without risking anything.

WP-CLI (for developers): wp core update --version=7.0-beta2 on a staging environment.

Beta Tester plugin: Install the WordPress Beta Tester plugin, select "Bleeding edge" channel and "Beta/RC Only" stream.

The WordPress 7.0 testing guide covers what to look for. Report bugs through WordPress Trac or the Alpha/Beta forum.

Release Timeline

Milestone	Date	Status
Beta 1	19 February 2026	Released
Beta 2	26 February 2026	Released (current)
Beta 3-4	5-12 March 2026	Upcoming
Release Candidate 1	19 March 2026	Upcoming
Stable Release	9 April 2026	WordCamp Asia

Seven weeks from Beta 2 to stable. If you're a plugin developer building AI features, that's your window to adopt the WP AI Client and Connectors infrastructure before the ecosystem converges on it. Our sister site has a full feature guide covering the complete WordPress 7.0 roadmap.

Frequently Asked Questions