How Many Websites Use PHP? 36.8M Sites and the Upgrade Queue Behind Them
As of November 16, 2026, webatla counts 36,829,757 websites running PHP, 11.30% of the 325.8 million websites it tracks. The live figure sits on webatla's PHP technology page, which recounts it every day. That is the he
As of November 16, 2026, webatla counts 36,829,757 websites running PHP, 11.30% of the 325.8 million websites it tracks. The live figure sits on webatla's PHP technology page, which recounts it every day.
That is the headline count. Measured as PHP market share among the 24 programming languages webatla tracks, the figure is 91.7% of all language detections. If you maintain PHP code for a living, the follow-up questions matter more. Which branch is a given site on, does that branch still receive security releases, and how much work stands between it and one that does? This post reads PHP usage statistics as an upgrade queue. It leans on the version table of the PHP page (the 25 most common version strings), on the support schedule the PHP project publishes, and on a few checks you can run against any host yourself. Rival languages, the history since 2015 and the country split are covered in webatla's PHP statistics article.
What webatla records for each PHP site
For a maintainer, webatla's scan answers two separate questions about each domain. Does anything the domain serves point to PHP, and does the server name the PHP release? A yes to the first puts the domain in the 36.8 million, counted once whatever HTTP status it returned. The second answer is optional, and most servers leave it out.
PHP sites in the November 16 scan and how many of them show a release number. Roughly two in five do.Everything below works from the 25 most frequent of those 714 strings, which together cover 85.7% of the sites that report a version.
That optional answer skews the sample. A hardened server discloses nothing, as the PHP page points out, so the hosts that still print a release lean toward stock settings. Every share below describes that self-selected group. Group totals are also floors, because about 2.06 million versioned sites run a string outside the top 25.
Check one host yourself
You can look at the same signal for a single site from a terminal. A HEAD request filtered to two header names is enough.
curl -sI https://example.com | grep -iE '^(x-powered-by|set-cookie):'
Here is how to read what comes back.
-
x-powered-by: PHP/8.3.xmeans PHP handled the request and printed its release. PHP adds this header on its own unlessexpose_phpis switched off in php.ini. -
set-cookie: PHPSESSID=...is PHP's default session cookie name. It appears when the application starts a session under that default name, and it carries no version. - No match proves nothing. The header may be disabled, a proxy or CDN may strip it, the page may come from a static cache, or the site may not run PHP at all.
In browser DevTools the same headers sit in the Network tab, on the document request. Add -L to the curl call when the first response is a redirect and you want the final page too. Treat the string as a claim about the interpreter. A distribution package may have patched that interpreter without touching the number, which matters a lot further down.
Where the versioned sites sit on the support calendar
php.net lists an end date for every PHP branch. At the time of writing, 8.3, 8.4 and 8.5 get fixes through 2027 at least. 8.2 gets security fixes until December 31, 2026. Every older branch, 8.1 included, gets nothing from upstream.
Sorting the 25 most common strings by that schedule gives the picture below.
PHP sites that send a version, grouped by the support status of their branch. Built from the 25 most common strings, so every group is a minimum.Read it top down. 5,076,777 sites, or 35.3% of those with a known version, are on a branch with a future. 2,443,102 (17.0%) are on 8.2. Under that line, 1,431,353 sites run 8.0 or 8.1, 2,608,189 run some 7.x release and 752,114 still run 5.x. Those three groups hold at least 4,791,656 sites, a third of the versioned base, on interpreters that upstream no longer patches.
Move the calendar to January 1, 2027 and 8.2 joins them. The unsupported group then reaches at least 7,234,758 sites, about half of every PHP site that reports a version. Nothing changes on those servers overnight. The schedule just runs out under them.
The version string is upstream's number. Linux distributions routinely backport security fixes into their own PHP packages while keeping the original version, so an old string can sit on a patched binary. The chart counts strings, not vulnerabilities.
PHP 7.4.33 and other final releases still in use
A version string can tell you more than its branch. Several of the most common ones are the very last release the PHP project shipped on a branch before it closed.
Sites reporting the final upstream release of a closed PHP branch. The seven strings add up to 3,181,535 sites.7.4.33 is the clearest case. It closed out the 7.4 branch, and 1,867,498 sites report it, 13.0% of all versioned PHP sites. Add 5.6.40, 8.0.30, 7.2.34, 7.3.33, 5.3.29 and 7.1.33 and the total comes to 3,181,535 sites. That is 22.1% of the versioned base and about two thirds of everything on an unsupported branch in the top 25.
For many of these servers the story is probably simple. They took patch releases until the branch ended, and the next minor or major upgrade never happened. An install that follows upstream patch releases on a closed branch, such as the official php:7.4 Docker image, ends on its final release because nothing newer exists there. Most likely the code on top needed changes before it would run on the next branch, and nobody scheduled that work. Updating a package and porting an application are different jobs, and these sites got only the first.
Distribution packages behave differently. A distribution usually settles on one PHP version for a release and backports fixes into it, so its builds keep a number of their own instead of ending on upstream's last one. Three strings in the table look like such builds. 7.2.24 (149,614 sites) is the PHP 7.2 build of Red Hat Enterprise Linux 8 and Ubuntu 18.04. 5.4.16 (95,617) and 5.3.3 (91,992) are the PHP versions of RHEL 7 and 6, and CentOS along with them. RHEL 7 and 6 also shipped Apache 2.4.6 and 2.2.15, and webatla's Apache HTTP Server page lists those two strings on 296,783 and 122,193 sites. The numbers come from separate tables, so they do not show how many are the same servers. A string like 5.4.16 may carry years of vendor backports. Whether those still arrive is a question for the operating system vendor.
Patch lag on the branches that still get fixes
The branches with a future look healthier, at least at the patch level.
Sites on the highest patch each 8.x branch reaches in the top 25, against sites on a lower patch or a bare branch number. Percentages give the top patch's share of its branch.On 8.2 to 8.5 together there are 7,519,879 sites in the top 25. Of those, 5,120,891, about 68%, report the highest patch number their branch shows in the table, such as 8.3.36 or 8.2.35. The remaining 2,398,988 sit on a lower patch or report only "8.4" with no patch number. 8.5 is the tidiest branch, with about 80% on its top patch, and 8.2 the least tidy at about 64%.
The table cannot tell whether 8.3.36 is still the newest 8.3 release on the day you read this, only that it is the newest 8.3 string common enough to make the list. The contrast with the previous section still holds. Patch updates inside a branch mostly happen, most likely because a package manager or a hosting panel applies them. Moving to the next branch is the step that stalls.
What breaks when you upgrade from PHP 5, 7 or early 8.x
If you inherit one of these sites, its branch gives a rough idea of the upgrade effort. The changes below come from the official PHP migration guides. They cover the common breaks and leave out many smaller ones.
| Starting point | Sites in the top 25 | What tends to break or warn on the way up |
|---|---|---|
| 5.x | 752,114 | 7.0 removed the mysql_* functions and ereg. Database code has to move to mysqli or PDO. |
| 7.x | 2,608,189 | 8.0 removed each() and create_function(), changed how strings compare with numbers, and turned many warnings into thrown errors. |
| 8.0 or 8.1 | 1,431,353 | Fewer hard breaks. 8.1 makes mysqli throw exceptions by default and deprecates passing null to non-nullable parameters of built-in functions. 8.2 deprecates dynamic properties. |
| 8.2 | 2,443,102 | Smaller steps. 8.4 deprecates implicitly nullable parameter types such as Foo $x = null. |
The 7.x row carries the weight. 1,986,123 of those 2,608,189 sites are on 7.4. Many removals in 8.0 had already been deprecated somewhere between 7.2 and 7.4, so a codebase that runs on 7.4 with every deprecation notice visible and none firing tends to be closer to 8.x than it looks.
A cheap first pass is to lint the code with the target interpreter. Syntax that 8.x no longer accepts, such as curly-brace string offsets like $str{0}, fails right there. The official Docker image saves you from installing anything.
docker run --rm -v "$PWD":/app -w /app php:8.3-cli \
sh -c "find . -name '*.php' -not -path './vendor/*' -print0 | xargs -0 -n1 php -l | grep -v '^No syntax errors'"
Linting only catches what fails to parse. Removed functions and new comparison rules surface at runtime, so the test suite has to run on the target version as well, with deprecations shown. Dependencies are often the first wall. In a Composer project, composer why-not php 8.3 lists the installed packages whose requirements rule that version out, and each of them needs an update or a replacement before the application can follow. Rector, which ships rule sets for PHP version upgrades, can automate a good share of the mechanical edits. Read the migration guide for every branch you skip. A move from 7.4 to 8.3 crosses four of them.
What the turnover says about PHP work
webatla's PHP page puts the change over the last 12 months at -1.5%. In that time 6.68 million PHP sites shut down, while 4.06 million stayed online and stopped showing PHP. Net of the sites that switched in from other stacks, switching cost PHP only about 460,000 sites, about 1.2% of its base, and part of that may be a measurement effect, since a site that starts or stops sending PHP headers looks like a switch from outside.
So the installed base shrinks mainly through shutdowns, and most PHP sites that stay online also stay on PHP. An application running on 7.4 today keeps its upgrade debt until someone pays it down or the site goes offline. For a PHP developer, a good share of the steady work is likely to be carrying existing applications from one branch to the next.
A short checklist for your own hosts
If you run PHP in production, a few habits follow from how these numbers come about.
Check the public hostname. Run the curl line from earlier against the public hostname rather than the origin. A CDN in front can strip the header or pass it through, and the public answer is what any outside scanner sees. To compare with the origin, keep the hostname and point curl at the origin's address with --resolve example.com:443:203.0.113.10, using your own origin IP.
Decide on expose_php on purpose. Setting expose_php = Off in php.ini removes the X-Powered-By header that PHP adds, once you reload PHP-FPM or the web server. It is a php.ini setting, so ini_set() in application code cannot change it. The setting changes only the response header and leaves the interpreter as patched as it was before.
Check the version on the box itself. php -v on the command line and phpversion() inside the application can disagree when the CLI and FPM come from different packages. The one serving requests is the one that counts.
Put the end date of your branch in the calendar. php.net lists it. For 8.2 that date is December 31, 2026.
FAQ
How many websites use PHP right now?
36,829,757 as of November 16, 2026. That is 11.30% of the 325.8 million websites in webatla's index, about one site in nine, and the PHP page refreshes the figure daily.
Which PHP release is most common?
8.3.36, on 2,372,491 sites, or 16.5% of PHP sites that report a version. Second is 7.4.33 with 1,867,498 sites, the final release of a branch the PHP project no longer supports.
Does an old version string mean a site is vulnerable?
Not by itself. Distributions backport fixes into their packages and keep the upstream number, so only the package version and the vendor's advisories on the host settle it. A string from an unsupported branch is a good reason to look both up.
Why do most PHP sites show no version?
The release shows only when the server sends it. expose_php = Off removes the whole X-Powered-By header, and a proxy or CDN in front can strip it as well. PHP can still be visible through other traces, such as a PHPSESSID cookie or the CMS running on it, and those carry no release number. Together that most likely explains why about 22.5 million PHP sites in webatla's scan show no version.
Sources
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.


