WordPress.com now lets AI agents write and publish posts with human approval

WordPress.com Enables AI Agents to Create and Publish Content

WordPress.com announced Friday a significant expansion of its AI capabilities, allowing AI agents to write, edit, and publish content directly on customer websites1

1

. The update, rolled out by Automattic, transforms how website management works by enabling AI agents like Claude and ChatGPT to handle tasks previously requiring direct human input2

2

. Website owners can now control their sites through natural language commands, instructing AI to draft blog posts, build landing pages, manage comments and metadata, and organize content with tags and categories1

1

.

The platform's reach makes this development particularly significant. WordPress powers over 43% of all websites on the internet, with WordPress.com seeing 20 billion pageviews and 409 million unique visitors every month1

1

. While the hosted version represents only a fraction of total WordPress installations, the network's sizable footprint means AI integration in web publishing now operates at considerable scale3

3

.

Model Context Protocol Powers AI Website Management

The new write capabilities build on Model Context Protocol (MCP) support introduced on WordPress.com last fall1

1

. MCP, an open protocol that standardizes how applications provide context to large language models, initially allowed AI assistants to connect and read site content, settings, and analytics from tools like Claude Desktop, Cursor, and VS Code3

3

. For six months, this integration remained fundamentally passive, giving agents a window to view but not modify content3

3

.

The Friday update adds 19 new operations across six content types: posts, pages, comments, categories, tags, and media3

3

. From a single prompt, an agent can draft and publish posts, approve and reply to comments, reorganize category structures, or fix missing alt text across entire media libraries to improve SEO1

1

. The AI can even search a site's theme and design system before creating content, understanding how to use the same colors, fonts, spacing, and block patterns to generate outputs that inherit the site's existing aesthetic1

1

.

Human Oversight and Approval Remain Central

Despite the autonomous capabilities, the feature is designed around explicit human approval at every step3

3

. Before creating, updating, or deleting anything, AI agents must describe exactly what they plan to do and ask for confirmation2

2

. All posts written by AI are saved as drafts by default, giving users a chance to review before anything goes live1

1

. Modifying a published post triggers a warning that changes will be immediately visible3

3

.

Deletions of posts, pages, comments, and media send items to the trash where they remain recoverable for 30 days3

3

. Categories and tags, which WordPress cannot trash, trigger an additional confirmation warning that deletion is permanent. Every action is tracked through the site's Activity Log, and user role permissions are fully enforced—an Editor can create and edit posts but cannot change site settings, while a Contributor can draft but not publish.

Access and Implications for Web Publishing

The write capabilities are available today on all WordPress.com paid plans, with the cheapest Personal plan starting at $9 per month2

2

. Paid subscribers can enable the features by visiting wordpress.com/mcp, toggling on the capabilities they want to use, then connecting their preferred AI client such as Claude, ChatGPT, Cursor, or any other MCP-enabled tool1

1

. Users can configure granular per-operation toggles as the default setup3

3

.

The expanded capabilities could greatly speed up the creation of websites where humans aren't doing much of the content creation, potentially filling the web with machine-generated content1

1

. This reflects a broader shift as generative AI becomes more entwined with popular services, following recent examples like deeper integration between Claude and Microsoft Excel and PowerPoint2

2

. The MCP ecosystem has been expanding rapidly, with Automattic's other products including WooCommerce and Beeper implementing their own MCP integrations, making standardized AI agent access an architectural assumption rather than an experiment3

3

.