Skip to content
Dimitri Robiere PerrietD. R. PerrietCaen, FR — --:--:--EN · FR

AI SEO article generator plugin for WordPress

Under contract · Mediapilote● Live
Client
Mediapilote
Role
Full-stack developer
Duration
17 months
Year
2025
Stack
WordPress · PHP · React · OpenAI · Gemini · Gutenberg · WooCommerce
AI SEO article generator plugin for WordPress — illustration
Fig. 01 — AI SEO article generator plugin for WordPress, illustration
(00) — In short

A WordPress plugin that uses AI to generate SEO articles, pages and WooCommerce product sheets as native Gutenberg blocks, with quotas and costs per client.

(01)

Context

Mediapilote wanted to offer its WordPress clients an AI SEO article generator that could also produce category descriptions and WooCommerce product sheets, without each site holding its own AI API keys.

The project started during my work-study as a single plugin, then split into two parts: the plugin installed on each client site and a central dashboard on the agency side. It continues today as part of my full-time role.

(02)

Challenge

A generated article is only worth something if it is structured, linked to the rest of the site and ready to publish. Raw HTML pasted into the editor was not enough: it needed valid Gutenberg blocks, relevant internal links and images that fit the topic.

Long generations also ran into hosting HTTP timeouts, and each client had to stay within its quotas, with costs tracked model by model.

(03)

Approach

I built a central gateway that manages clients, hashed API keys, quotas and real token-based costs, and is the only component calling OpenAI, Gemini or BFL. The client plugin goes through it: single or bulk generation from a CSV, a quick mode with web search, batch rewriting of WooCommerce products.

Generation follows a multi-agent pipeline: a planner, a writer and a designer produce a block representation, serialised into native Gutenberg blocks, with generated images and automatic repair of invalid blocks. Internal linking relies on embeddings and hybrid search. Long tasks go through a durable job queue and a detached worker, tracked by polling, and the latest versions generate full pages editable in an in-house React builder.

(04) — Results

Key figures
229
commits
2
parts: client plugin and central gateway
3
agents: planner, writer, designer
3.1
plugin version
← Previous project07 · GEO Mediapilote, a brand visibility audit for AI searchNext project09Mediapilote Backlinks, AI agent-driven link building