Where the Data Actually Points

Start with WordPress, because it's still the flatly dominant CMS: as of July 2026 it runs 59.2% of the tracked CMS market, and 41.5% of all websites. That share is down recently — from 62.9% as recently as 2024 — but WordPress actually grew for most of the last decade, peaking at 65.2% in 2022. The decline is real, but it's only about four years old, not the whole story some people assume.

For some, that's enough to reheat the "PHP is a dying language" argument. It isn't really worth relitigating here — PHP's aggregate usage is a separate question from any one CMS's market share — so I'll leave that aside and move on.

The more interesting question is where WordPress's recent losses are actually going. Here's what the last decade of W3Techs data actually shows, platform by platform:

• Wix — 3.8% → 6.1% of the CMS market in the last two years alone. Roughly double its historical growth rate — the clearest sign of real, recent acceleration.
• Shopify and Squarespace — both still climbing, but at their normal, steady pace rather than accelerating.
• Joomla — 8.6% → 1.7% over the full decade. A long, slow decline — not a recent event.
• Drupal — 5.1% → 1.0% over the same decade. The same shape as Joomla's line, on the same timeline.
• Sites with no detectable CMS at all — 61.7% → 28.6% (2015 to January 2026). Also shrinking for a decade, so WordPress's lost share isn't going to some invisible category — it's going to Wix, Shopify, and Squarespace.

The clearest read of this data is a decade-long consolidation onto a handful of hosted, subscription-model platforms, squeezing self-hosted open-source CMS in general — WordPress included, not just Drupal. Part of what makes those hosted platforms attractive is that they own the operational overhead a self-hosted CMS pushes onto you: Wordfence attributes roughly 55.9% of known WordPress attack entry points to plugin vulnerabilities, a cost of the plugin ecosystem that's also WordPress's biggest selling point. A hosted platform simply doesn't hand you that trade-off.

That's the market Drupal has actually been losing ground in for over a decade, on the same curve as Joomla. I build custom Drupal at Kellett Communications for enterprise clients, and I'm currently building an AI chatbot on the Drupal AI module for the Mackenzie Valley Land and Water Board site. What follows is where Drupal is actually placing its bets to stay relevant in that market.

Drupal CMS 2.0: Drupal's Answer to the Low-Code Segment

Drupal isn't sitting out the low-code wave. Drupal CMS 2.0, launched January 28, 2026, is a genuine low-cost, low-skill on-ramp aimed at the market WordPress has owned for two decades:

• Drupal Canvas — a visual drag-and-drop page builder with live preview, so building a page no longer means switching between an admin form and a separate preview window. It's built on a "Mercury" component library — cards, heroes, testimonials, accordions.
• AI-powered site building — an admin chatbot handles site-building tasks like creating content types and taxonomy terms; Canvas AI can generate a complete, structured landing page from a text prompt using Mercury components; AI-assisted alt text generation ships with human review built in.
• Site templates — feature-complete starting points that install in under three minutes. The first, "Byte," is a SaaS marketing site with a blog, newsletter signup, pricing, and contact form.
• Recipes — the packaging mechanism underneath all of this: prepackaged module and configuration bundles that make templates and features installable as a unit.
• Drupal Core 11.3 underneath it all, with the biggest performance jump in a decade — 26-33% more requests served on the same infrastructure.

My read: Recipes and Drupal CMS are Drupal's real shot at being a "WordPress killer" — not by beating WordPress on plugin count, but by making Drupal's underlying robustness usable by people who aren't developers.

What I'm Actually Building With: the Drupal AI Module

The piece of this I have hands-on experience with is the AI module itself — 12,676 sites in production as of March 2026. A few submodules matter here:

• AI Automators — field-level automation: populate, transform, or enrich fields via chained prompts, web scraping, OCR, transcript generation. Teams using it report 40-60% reductions in content-processing time.
• AI Assistants API + Chatbot — the backend for building configurable chat-based assistants: what reasoning strategy they use, what data they're allowed to access. Frontend-agnostic.
• AI CKEditor — inline AI (rewrite, translate, summarize, polish) directly in CKEditor 5, so AI assistance is part of everyday editing rather than a separate tool editors have to remember to open.
• Connections to 48+ providers — OpenAI, Anthropic, Gemini, Mistral, self-hosted via Ollama. Nothing here locks you to one vendor.

The part I've actually built against is the agents framework. The MVLWB chatbot uses it to choose between multiple tools per query rather than being locked to a single retrieval strategy: a RAG search tool for general site content (vector embeddings via locally-run Ollama, currently in sqlite-vec for development with Milvus planned for production), and a custom RegistrySearchTool that queries the site's registry entities directly — deterministically, not by similarity match, so it can't hallucinate a result that doesn't exist in the database. Reasoning runs on the Anthropic API, with up to eight tool-call loops per query.

That agent-picks-the-tool shape matches Dries Buytaert's broader 2026 AI roadmap for Drupal, which frames the CMS as a "governed control layer" for agentic workflows rather than just a content store — the CMS defines what an agent is allowed to touch and how, instead of an AI product bolted on top of a static site.

Where Custom-Built Drupal Remains the Right Call

None of this changes who actually needs Drupal. It changes who Drupal also wants to capture.

The 1% overall figure hides a concentration curve, not a flat decline. W3Techs' own ranking breakdown (July 2026) shows Drupal's share climbing the higher up the traffic-rank ladder you go, then falling off again past the largest sites: 6.0% of the top 1,000 sites, 7.1% of the top 10,000, 6.4% of the top 100,000, tapering to 3.7% of the top million. Drupal is the second-most-popular CMS among the top 100,000 sites, behind only WordPress — but that ranking inverts at the low end, where WordPress and no-code builders dominate on sheer volume. The largest active buyer cohort in 2026 is mid-market enterprise: organizations roughly $5M-$500M in revenue commissioning real Drupal work without a Fortune 500 budget, alongside the traditional strongholds of federal government, large universities, and enterprise media.

What custom Drupal work actually involves at that tier is categorically different from what a Recipe or a WordPress plugin installs out of the box: custom module development, complex content architecture, multi-system integration, compliance requirements, granular permissions, multi-language publishing at scale, long-term platform stewardship. Small businesses were never the Drupal market — the low-cost alternatives (WordPress, Squarespace, Wix, Shopify) have simply kept expanding to serve them, and have been for a decade. Drupal CMS is Drupal's way of also competing for that segment on its own terms.

The Bigger Picture

The CMS market has been consolidating onto hosted, subscription-model platforms for over a decade — that's the trend Drupal is actually up against. It's responding by trying to compete on both ends of what's left: Drupal CMS and Recipes at the low-code end, and its existing architecture-level strengths — the kind of thing an agents framework and a custom deterministic query tool can be built on — at the enterprise end it's always actually owned.

That's also roughly where I sit personally: building custom Drupal at Kellett for enterprise clients, while experimenting with the AI module and agent tooling on the MVLWB chatbot on the side. It's a small-scale version of the same pattern the article argues Drupal is well-placed to own — AI at the edges, real architecture underneath.