The question behind the question
"Should we build on WordPress or go custom" is rarely a technology question. It is an ownership question: who edits the site in year two, who patches it, who gets paged when the contact form dies mid-launch. I have shipped more than 12,000 WordPress sites through the agency I co-founded, and I run my own sites on Astro and Next.js, so I hold no flag here. The honest answer depends on the shape of your business, and in 2026 there are three real options, not two.
The three options, defined properly
WordPress, the classic build. One application handles editing, templates, and serving. Themes and plugins cover an absurd range of needs, the block editor has matured into a genuinely capable page builder, and every marketer you will ever hire has already used wp-admin.
Headless. Content lives in a CMS (WordPress with WPGraphQL, or Sanity, Storyblok, Payload), and a separate front-end built in Next.js or Astro renders it. This is the architecture the Jamstack era named: pages pre-built ahead of time, served from a CDN, with APIs handling the dynamic parts. The buzzword faded; the architecture won. Most of the fastest content sites in 2026 are built this way, whether or not anyone says Jamstack out loud any more. There is a full walkthrough in my headless WordPress guide.
Custom. The front-end and the data model are designed for your product: a Next.js or Astro application with a database, auth if you need accounts, and exactly the features you asked for, nothing else. The right shape when the site is the product rather than the brochure.
The comparison, side by side
| WordPress | Headless | Custom build | |
|---|---|---|---|
| Typical build cost | 3,000-30,000 USD | 15,000-60,000 USD | 25,000-150,000+ USD |
| Time to first launch | 2-6 weeks | 6-12 weeks | 10-20+ weeks |
| Editing experience | wp-admin, familiar to every marketer | CMS studio, clean but new to most teams | whatever you build, sometimes nothing |
| Performance ceiling | good, with discipline | excellent by default (static + CDN) | excellent, you control everything |
| Security surface | largest: core, plugins and admin to patch | small: static pages, CMS handles auth | yours to design and defend |
| Ongoing upkeep | core + plugin updates, monthly | front-end dependencies + CMS subscription | a developer relationship, permanently |
| Design freedom | theme-bounded unless custom theme | full | full |
| Hiring pool | enormous | large and growing | framework-specific, senior-skewed |
| Best for | marketing sites, blogs, small commerce | content at scale, multilingual, performance-critical | products, portals, bespoke workflows |
Is WordPress outdated in 2026?
No. It still runs roughly 43% of the web, the block editor has quietly become good, and the ecosystem remains the deepest moat in web software. What is also true: the default path (fifteen plugins, cheap hosting, no performance budget) produces mediocre sites, and it always did. WordPress done with discipline (few plugins, decent hosting, a performance budget) is still the fastest sensible route to a professional marketing site.
Why are people moving away from it, then?
The honest reasons: teams tired of the update treadmill, performance ceilings on plugin-heavy builds, and developers preferring modern JavaScript tooling. The equally honest counter-flow: teams come back for the editing experience, the enormous hiring pool, and the cost. Both migrations are real, and I have run projects in both directions; the fuller picture sits in my piece on WordPress alternatives. A platform switch fixes architecture problems. It does not fix content, positioning, or ownership problems, and those are the usual disease.
Even with AI, custom is not suddenly easy
Here is the 2026 twist. Claude, Cursor, and agentic coding tools have collapsed the cost of producing code. I have built a working site in a day with these tools, and through MCP, the Model Context Protocol, an AI assistant can now talk directly to your CMS, your database, and your deployment platform. The distance between idea and demo has never been shorter.
Production is a different sport. The demo does not include backups, auth edge cases, monitoring, dependency updates, accessibility, SEO plumbing, or the 2am incident. AI compresses the build; it does not absorb the ownership. A vibe-coded site that nobody maintains becomes legacy software within months, just faster and cheaper legacy than the old kind. What AI genuinely changed: custom builds got cheaper and better, because senior developers now ship multiples of what they used to, and WordPress builds got better too, because the same tools write themes and plugins. The gap narrowed from both sides. The decision framework survived intact.
How to actually decide
Pick WordPress when the site is a marketing surface, non-technical people edit it weekly, and the budget is finite. Pick headless when content operations are serious (volume, languages, multiple channels) and performance is a ranking or conversion lever you intend to pull. Go custom when the site is the product: accounts, dashboards, workflows, integrations, the things a plugin was never going to do properly. Real numbers for that route are in my breakdown of custom web development costs.
Then apply the year-two test, which outranks everything above. Who owns this in year two? A perfectly chosen platform with no owner still fails. Choose the stack your actual team, or your actual budget for outside help, can keep healthy, and the rest of the comparison table becomes detail.