Is WordPress Dying? The True State of WordPress in 2026

The WordPress obituaries have a real number this time: 43.6% of surveyed websites in January 2025, 40.7% by August 2026, most of the fall inside 2026 and accelerating. If you run sites or a store on WordPress, the question is simple: is the platform still a safe bet?

That number is weaker than it looks. But a share can only answer so much, and the real question splits into three markets with three different years.

Three groups, three different verdicts

Split the market by the shape of the job: does real complexity exist, and is it shared by thousands of businesses or unique to one?

• Group A, smaller websites: a restaurant page, a portfolio. Little complexity; any tool works.
• Group B, online stores: a hard job shared by millions of merchants. On WordPress that means WooCommerce.
• Group C, complex builds: custom workflows, requirements that exist nowhere else. A hard job with no standardized shape.

Group A: the small website became a prompt

A 3-page site is a prompt now. AI builders run $2 to $32 a month, and Lovable alone reports 1 million new projects a week, aimed squarely at the work WordPress once got by default. This group is gone for new work: nobody starts here anymore, and it was the platform’s lowest-margin territory, the floor. That flood of new AI-built sites has a second job: it crowds the very statistic that says WordPress is dying.

Group B: WooCommerce lost the center, kept the fringe

WooCommerce’s storefront count peaked at 4.78 million in Q2 2024 and has fallen since, to 4.34 million by May 2026 (Store Leads via Mobiloud), a 4 to 5% yearly leak, with Shopify winning the center: standardized retail.

The twist: on the one detector that counts active storefronts on both platforms, WooCommerce still runs 1.5× Shopify’s stores, 4.34 million against 2.84 million. A counted Shopify store is almost always a paying merchant. A detected WooCommerce storefront may not be. That asymmetry runs against WordPress: the 1.5× lead is softer than it looks.

What WordPress keeps is the fringe where retail grows bespoke edges: B2B pricing and approval logic, content-led commerce, checkout and data-model control, platform-fee avoidance at volume. The fringe is eroding too (Shopify Plus now ships native B2B). Verdict: default status lost, existence kept. The verdict flips if native B2B absorbs those named fringe cases while the leak continues.

Group C: complex builds are holding

Group C has held, as of 2026. Pew Research Center, Al Jazeera, Ubisoft and CNN Brasil run WordPress at the enterprise tier. Each is publicly verifiable in its own page source or the WordPress.org showcase. The stickiness is real machinery: installed base, the maintenance liability of custom code, teams who run WordPress for years. A custom build has no drop-in replacement; leaving means rebuilding, and organizations rarely rebuild working systems.

The threat I take seriously here is agentic codegen making bespoke builds cheaper from below. Headless CMS vendors pitch at this group from above, and from where I sit that migration is rare, loud case studies aside. If organizations like the ones above start moving off WordPress, that flips this verdict.

From default pick to reasoned pick

Group A: new work gone, and it was the floor. Group B: center lost, fringe kept for reasons you can list. Group C: held. WordPress went from what you pick by default to what you pick because you have a reason. That is a weaker market position, and the same sentence described Drupal in 2015. The distinguisher is mass: on the very leaderboard I take apart below, WordPress holds 40% of all surveyed sites, the largest share any CMS has ever posted there, against Drupal’s peak of 2.3% on the same series. The leaderboard’s flaws blur trends; a seventeenfold gap survives them. Add a commerce base still bigger than its challenger’s on the one count that covers both platforms.

So if the map says weakened but standing, where does “dying” come from? One survey, and it cannot carry the verdict.

The dying number counts a top-10-million leaderboard instead of the whole web

W3Techs defines its survey as the 10 million most-visited websites: a leaderboard where a seat costs a few hundred visits a month on every ranking source that could plausibly feed it. Parked domains and empty pages are filtered out. That is the whole method, and the definition dates to 2013. The ranking source behind it, Alexa, shut down in 2022, and W3Techs never published what replaced it, nor whether the sample stayed the same size. The industry’s standard statistic runs on an undocumented ranking basis, a fun property for a number that decides platform funerals.

The same data also reads two ways: 40.7% of all surveyed sites, or 59.0% of sites with a known CMS. A “43 to 41” post and a “60 to 59” post describe the same August; the first counts all surveyed sites, the second only sites with a known CMS.

On W3Techs’ own definition, web growth can look exactly like WordPress decline

On its stated definition the sample is always 10 million sites. So a falling share means one thing mechanically: the leaderboard now holds roughly 290,000 fewer WordPress sites (the 2.9-point fall from 43.6 to 40.7, applied to 10 million, back-of-the-envelope), every vacated seat filled by something newer. New sites outranking the WordPress sites at the bottom of the leaderboard is enough on its own to produce the whole fall. Not one of those sites has to be dead. They only have to be outranked.

And the most obvious source of the push is the group A flood. Registered domains grew from 350 million in late 2022 to 401.6 million by mid-2026. Recent growth ran at 6 to 8% a year, up from about 1% in 2023 (Verisign). 74.2% of newly created English pages carry AI-generated content, and W3Techs’ filters catch empty domains while plausible-looking AI content sails straight through (NewsGuard tracks 3,749 AI content farms, growing by 300 to 500 a month). “WordPress lost sites” and “the web grew around WordPress” produce the identical curve, and the published number cannot tell them apart.

The strongest counter-exhibit cuts against the “web grew around WordPress” story. Among sites with a known CMS, WordPress fell from 65.2% in January 2022 to 59.0% now, a slide that started three years before the AI flood; part of that is plain competition, and rival explanations (the Automattic governance rupture among them) fit the same window. A smaller bias runs the other way: detection is surface fingerprinting, so the minority of WordPress sites running behind a JavaScript front end get counted as something else; magnitude unmeasured.

And the famous “518 million WordPress sites”? That is W3Techs’ share multiplied by Netcraft’s hostname total: quoting it as corroboration is quoting the same number twice. Nobody publishes an absolute WordPress count on a consistent population. The demonstrator below makes the two worlds concrete: set WordPress shrinking or set the web growing, and both presets land on the identical published curve.

The bet: two numbers end this argument

The bet defines dying as platform death. If dying means “no longer the default pick,” that already happened. WordPress is genuinely dying the day two published datasets exist: a falling absolute WordPress count on a consistent population, and a new-build stack mix (what new sites actually choose) showing groups B and C leaving. The counters that would settle it: HTTP Archive on a fixed cohort, a Store Leads full-platform series, or W3Techs disclosing absolutes. They stand even where they undercount headless builds, no discounting after the fact. Half the bet has already fired: for e-commerce, the absolute count is falling while the same detector counts the challenger growing. Conceded in full. Publish both numbers for the platform as a whole and I drop the argument the same afternoon. Until then, the verdict is unproven dying, never proven health.

Key takeaways

• New work in Group A is gone to AI builders, and it was the floor; the platform bet lives in groups B and C.
• WooCommerce leaks 4 to 5% a year yet still counts 1.5× Shopify’s stores on the one detector measuring both.
• Complex builds have held: installed base, the maintenance liability of custom code, and no drop-in replacement to leave for.
• The 40.7% is a share of W3Techs’ top-10-million leaderboard: the fall equals ~290k fewer WordPress sites above the popularity bar, and it cannot distinguish “WordPress lost sites” from “the web grew around it.”
• Default pick became reasoned pick. The two numbers that would prove dying are named above; neither exists yet for the platform as a whole, and for e-commerce half the bet already fired.

That is the true state as of August 2026. If you’re weighing a platform decision that this map applies to, I’m happy to point you in the right direction.

Where the numbers come from

• WordPress share, all-sites series (43.2 · 43.1 · 43.1 · 43.6 · 43.0 · 40.7) and known-CMS series (65.2 → 60.2 → 59.0): W3Techs historical trends and known-CMS history, accessed 2026-08-19. Sample basis: 2013 “top 10 million” post, 2020 Tranco extension, FAQ (which describes the target as “the relevant web”). The post-Alexa ranking basis, and whether the sample size stayed fixed, are unpublished; the displacement reading in the body follows W3Techs’ stated definition, not a verified census.
• The “few hundred visits a month” entry bar: archived Alexa support pages (rank reliability degrades sharply at low traffic; ranks beyond 100,000 “not statistically meaningful”) and CrUX methodology (threshold undisclosed, estimates near 100 visits/month). Ranking windows are short (Tranco 30 days, CrUX 28, Cloudflare Radar 7), so whatever moved the curve moved recently. Alexa’s 2022 retirement: Cloudflare, “Goodbye, Alexa”.
• The ~290k displacement figure is this article’s own arithmetic: 2.9 percentage points of a nominal 10-million-site sample. No source publishes it. An alternative reading treats the surveyed sample as growing with the web; the fall then needs ~4.3%/yr growth against 6 to 8% observed, if ranked-population growth tracks domain growth (unmeasured). W3Techs’ stated definition is fixed-size, so the displacement reading leads in the body.
• Registered domains: Verisign DNIB Q4 2025 and Q2 2026 release, accessed 2026-08-17; the ~1% comparison point is 2023 growth. Only about a third of registered domains resolve to a real site (one registrar’s portfolio, may not generalize), and a new domain displaces nothing until it clears the ranking bar; DNS-based rankings count resolver queries, bots and AI crawlers included.
• AI content share: Ahrefs, May 2025, 900k pages, one per domain, Ahrefs’ own detector, labeled imperfect. Accessed 2026-08-17.
• AI content farms: NewsGuard AI Tracking Center, updated 23 June 2026; news-adjacent sites only, a floor. Accessed 2026-08-19.
• Store counts: Store Leads active-storefront series via Mobiloud, figures as captured 2026-08-17 (the live roundup has since updated; the captured series is the one used throughout). The 4 to 5%/yr leak rate is this article’s calculation from the Store Leads endpoints; quarterly movement is lumpy. Per-platform detection error unknown in both directions. BuiltWith counts 6.77M sites with the WooCommerce plugin present, a different question (plugin presence versus working checkout). Shopify’s own merchant figures come from other methods and are never mixed with Store Leads counts here.
• Shopify absorbing the majority of WordPress’s share losses: Shopifreaks, secondary reporting on W3Techs monthlies. Accessed 2026-08-17.
• Site generation: TechCrunch on Lovable, June 2026, vendor claim; projects are not live domains. AI builder pricing ($2 to $32/month): builder pricing pages surveyed at research, 2026-08-17.
• Headless detection mechanism: Wappalyzer’s public WordPress rule file (community-maintained fork; rules include markup paths, block classes, and response headers), accessed 2026-08-17. No published measurement of how much WordPress runs headless exists; the “rare migration” read is the author’s practitioner judgment.
• Enterprise WordPress: verified 2026-08-19 via first-party WordPress fingerprints (/wp-content/ asset paths in the organizations’ own homepages: Pew Research Center, CNN Brasil, Al Jazeera) and the WordPress.org showcase (Ubisoft). Only names with a neutral public verification route are printed.
• The 518M circle: wordpress.com blog, April 2025, W3Techs share applied to Netcraft’s hostname total.
• Drupal peak: 2.3% of all surveyed sites (W3Techs all-sites yearly history, link above, accessed 2026-08-19); its known-CMS peak was near 5%.
• Practitioner fringe list (B2B logic, content-led commerce, checkout/data-model control, platform-fee avoidance): consistent across 5+ 2026 practitioner sources, no primary measurement; Shopify Plus native B2B (company accounts, price lists, net terms) has shipped since 2022.