Website maintenance is the recurring work that keeps a site secure, available, fast and correct after launch: applying updates and security fixes, test-restoring backups, watching for outages and expiring certificates, checking speed and broken links, reviewing who has access, and renewing the domain. Each task runs on its own clock, from continuous alerts to a yearly review.

Below: the tasks by frequency, the dates that set the schedule, a monthly checklist, agreement clauses and cost drivers. For recovery targets, see RTO vs RPO; for how sites are built, our web development guide.

Which website maintenance tasks run how often?

The cadence below is a suggested baseline, not a standard: a shop that takes payments tightens it, and a brochure site can relax it. Staging means a private copy of the site where updates are tried before visitors see them. Security controls get more depth in our website security checklist.

WhenTaskDone means
ContinuousUptime and key-flow checksA named person is alerted when the home page, a form or the checkout fails
ContinuousCertificate, domain and DNS expiryAlerts from outside the host reach two people well before each date
ContinuousBackup and vulnerability alertsA failed backup or a new flaw in something you run opens a ticket that day
WeeklySecurity fixes and alert reviewCritical fixes tested on staging, then applied; logs and alerts read
MonthlyUpdate roundCore, plugins, themes and libraries updated on staging, then live, and logged
MonthlyBackup checkOff-site copies exist at the expected size and one database copy opens
MonthlySpeed, links and admin accountsWeb Vitals read, 404s fixed, admin list and unused plugins pruned
QuarterlyTimed restore drillA backup restored elsewhere, the site checked, the time compared with the target
QuarterlyScan and access reviewExternal scan triaged; users, API keys and third-party scripts reviewed
QuarterlyAccessibility and content reviewKey flows tested by keyboard and screen reader; prices, contacts and legal text checked
YearlyDomain, DNS and certificatesRegistrar contacts current, renewal paid, stale DNS removed, renewal automation proven
YearlyVersion plan and agreementEnd-of-life dates checked, upgrades booked, agreement and hosting plan reviewed

A monthly round alone is too slow for security. In Patchstack's data on 2025, about half of high-impact WordPress vulnerabilities were exploited within 24 hours and, weighting by how intense the exploitation was, the median time to first exploit was 5 hours. Patchstack sells WordPress vulnerability protection, so treat the figures as vendor data. They are why critical fixes go out of cycle.

The continuous rows follow one rule: check what earns money, not only whether the server answers. A ping of the home page can stay green while the checkout is broken. A synthetic check runs from outside your host, walks the flow (product page, basket, form) and alerts a named person at the step that fails.

A round probe sends an orange dashed path through a chain of four browser windows, home page, list, basket and form. The link before the last window is a gap and its frame is cracked.
Fig. 1 A home-page ping proves one link; a flow check walks the chain and shows where it breaks.

Which version and expiry dates set the schedule in 2026 and 2027?

Other people set several dates, all as published in October 2026.

PHP gives each branch two years of active support, then two more of critical security fixes only, after which it is end of life (EOL).

PHP branchActive support untilSecurity fixes until
8.231 Dec 202431 Dec 2026
8.331 Dec 202531 Dec 2027
8.431 Dec 202631 Dec 2028
8.531 Dec 202731 Dec 2029

PHP 8.1 reached end of life on 31 December 2025, 8.0 on 26 November 2023 and 7.4 on 28 November 2022. WordPress recommends PHP 8.3 or greater, MariaDB 10.11 or greater or MySQL 8.0 or greater, and HTTPS, so a site on PHP 8.2 gets security fixes until 31 December 2026 but is already below that recommendation.

A headless front end or a custom app has the same question about Node.js. Version 20 reached end of life on 30 April 2026, 22 follows on 30 April 2027 and 24 on 30 April 2028, and the project says production applications should only use Active LTS or Maintenance LTS releases.

Certificate lifetimes are the third clock. Under the CA/Browser Forum Baseline Requirements, a publicly trusted TLS certificate issued on or after 15 March 2026 may be valid for at most 200 days, on or after 15 March 2027 for 100 days, and on or after 15 March 2029 for 47 days. A yearly manual renewal stops being workable, so automate it and watch the expiry date from outside; our guide to renewing an SSL certificate covers both. Let's Encrypt announced it would end its expiration notification emails on 4 June 2025, so do not wait for a warning.

A cloud hands a certificate card to a server that holds a padlock and a small renewal gear. Outside the server, orange binoculars on a tripod look at the padlock along a dashed sight line.
Fig. 2 Renewal runs inside the server and can fail quietly, so the check that counts looks from outside.

Domain registrations run on a fourth clock, and the failure is harder to undo. For generic top-level domains, ICANN's Expired Registration Recovery Policy requires registrars to interrupt the name's DNS resolution before deletion, where the registry permits, while the owner can still renew. Per ICANN's renewal FAQ, a deleted name then spends 30 days in redemption, where it can still be restored, and 5 days in PendingDelete before it is released for registration on a first-come-first-served basis. For a country-code domain, check your registry's own rules.

Tip

Give the registrar a second contact address, and keep the card behind auto-renew current. The policy says registrars should advise holders to provide "a secondary email point of contact that is not associated with the domain name itself", so that reminders still arrive if the domain expires.

What does WordPress update by itself, and what is left to you?

The defaults, from the WordPress upgrading guide and the auto-updates documentation, differ by item and by the age of the install.

ItemDefaultChange it with
Minor core releasesOnWP_AUTO_UPDATE_CORE set to 'minor', true or false; AUTOMATIC_UPDATER_DISABLED switches off all
Major core releasesOn for installs created on 5.6 or later, unless a version-control checkout is detected; older installs keep minor only until it is turned onWP_AUTO_UPDATE_CORE, or the allow_major_auto_core_updates filter
Plugins and themesOff, except in special cases the WordPress.org API sets for critical vulnerabilities; since 5.5 an administrator can opt in plugin by plugin and theme by themeThe Plugins and Appearance screens, or the auto_update_plugin and auto_update_theme filters

By default, opt-in plugin and theme auto-updates run twice a day, and WordPress sends email notifications when they succeed or fail. They rely on WP-Cron, which checks for due tasks only on page load, so a quiet site can run late. Check Site Health monthly: among its possible critical issues are background updates not working, a loopback request that cannot complete and an outdated PHP version. Point the site's administration email at a mailbox someone reads.

Staying on an old major version is not a safe default: the hardening guide says older versions of WordPress are not maintained with security updates.

Since WordPress 6.6 in July 2024, a failed plugin auto-update can be rolled back automatically. According to the core merge proposal, WordPress requests the home page after the update. If that loopback request returns an error, it assumes a PHP fatal error, restores the previous version, emails the site's administration address and tries the same plugin again at the next auto-update check.

A cutaway server holds a row of puzzle-piece modules. A looping orange arrow runs from the newest module to a cracked browser window on the open door, while a grey arrow lifts the older module back from a shelf.
Fig. 3 The rollback test is a request to the home page: it catches a crash, not a quietly broken feature.

Warning

The rollback test described in the proposal looks for a PHP fatal error, not wrong behaviour. A plugin that loads without crashing but breaks the checkout, the contact form or the layout would not trip that test. Auto-updates do not replace a staging test or a backup you have restored.

How does the work change for static, headless and custom sites?

The cadence stays the same; what you patch changes. A headless site adds a CMS and its connection (see what a headless CMS is), and a custom application adds your own code.

QuestionWordPressStatic or headlessCustom application
What you patchCore, plugins, theme, PHP, databaseBuild tools and packages; the CMS and server runtime if you host themLibraries, framework, runtime, containers, operating system
Where known flaws appearMostly plugins: 91% of 2025's new WordPress vulnerabilities in Patchstack's dataPackages in the lockfilePackages in the lockfile, plus your own code
How updates arriveDashboard or WP-CLI, staging firstDependabot pull requests, a rebuild and a preview deployThe same, through the CI/CD pipeline, with tests as the gate
What a restore needsDatabase, uploads, configurationSource repository, CMS content export, build settingsDatabase, uploaded files, configuration and secrets

For a hosted shop or site builder, the Canadian Centre for Cyber Security notes that who patches depends on the cloud service model, and that with software as a service the provider does. Our website security checklist has the table for each hosting model.

On a custom application the clocks are mostly automated. For Minute Notary, an online notary platform with its head office in Ottawa that we build and run, base-image updates arrive automatically, synthetic monitoring watches the live site, every outside dependency has its own health alert, and a breach response runbook is written down before it is needed.

What does a monthly maintenance checklist look like?

Run the round on a staging copy first, then on production in the agreed update window, and keep the output as the change log. For WordPress, start with wp db export, which runs mysqldump; uploads and files still need your normal backup job, covered in our website backups guide. Then list what is waiting with wp plugin list, which accepts a field filter such as --update=available.

wp db export ~/backups/pre-update-$(date +%F).sql
wp plugin list --update=available --fields=name,version,update_version,auto_update
wp plugin update --all --dry-run
wp plugin update --all
wp theme update --all
wp core update
wp core update-db

wp plugin update previews with --dry-run, and its documented example output shows maintenance mode switched on while it works, so keep to the update window. Before updating, load the key flows on staging and note what they do: the form sends mail, login works, the basket takes a product. Repeat after the update and compare.

Three packages ride a conveyor into a tunnel where one browser window is shown twice, before and after the update, and an orange magnifying glass compares them before the belt reaches a server.
Fig. 4 Run the same flows on staging before and after the update, and the difference is the test.

Then check the result and the things updates do not fix:

wp core verify-checksums
wp plugin verify-checksums --all
wp user list --role=administrator --fields=ID,user_login,user_email,user_registered
wp plugin list --fields=name,status,version,wporg_status,wporg_last_updated
wp plugin list --status=inactive --field=name    # review, then delete what you do not use

wp core verify-checksums and wp plugin verify-checksums compare installed files with WordPress.org's checksums, and wp user list takes --role. The wporg_status and wporg_last_updated fields show whether a plugin is still active on WordPress.org and when it was last updated. The hardening guide says to delete a plugin you are not using.

Outside WordPress, the same round runs on the lockfile. npm audit exits with a non-zero code when it finds a vulnerability, and --audit-level sets the lowest severity that fails the run. composer audit exits with 1 when it finds packages that match its dependency policies, such as vulnerable ones. Dependabot can open the update pull requests, on an interval from daily to yearly, and by default waits 3 days after a release before proposing a version update, except for security updates.

npm audit --audit-level=high
composer audit --locked --no-dev

Two more checks run from outside your host. openssl x509 -checkend exits non-zero if the certificate expires within the given number of seconds, openssl s_client opens the TLS connection, and curl prints a status code, a total time or where a redirect points.

# Expiry date; exits non-zero if the certificate expires within 14 days
openssl s_client -connect example.com:443 -servername example.com </dev/null 2>/dev/null \
  | openssl x509 -noout -enddate -checkend 1209600

# Status code and total time of a key page
curl -s -o /dev/null -w '%{response_code} %{time_total}s\n' https://example.com/contact/

# An old URL: expect 301 and the new address
curl -s -o /dev/null -w '%{response_code} %{redirect_url}\n' https://example.com/old-page/

For performance, the targets are LCP within 2.5 seconds, INP of 200 milliseconds or less and CLS of 0.1 or less, measured at the 75th percentile of page loads. Search Console's Core Web Vitals report shows real-world field data for groups of similar pages, with group figures for the last 28 days. That window confirms a regression late, so test the key templates in a lab tool after every change. Fixes are in our Core Web Vitals guide.

For links, Google's documentation on HTTP status codes says it removes URLs that return a 4xx code (other than 429) from its index, treats a 301 as a strong signal that the redirect target should be processed and a 302 as a weak one, and slows its crawling when a site returns 5xx errors. After any URL change, run the curl redirect check above; our WordPress SEO guide covers migrations and redirects.

For accessibility, automated checks are a start. The W3C Web Accessibility Initiative says tools "cannot check all accessibility aspects automatically" and that human judgement is required. Each quarter, scan the key templates, then do a keyboard-only pass and a screen-reader pass on the main flows. The WCAG 2.2 AA process is in our WCAG guide.

Who does what, and what belongs in a maintenance agreement?

Split the work by who can actually do it. The owner keeps the registrar, DNS and hosting accounts, the budget, and the accuracy of prices, contacts and legal text. The maintainer does the technical upkeep: updates, monitoring, backups and restore tests, incident response and the monthly report. Hiring one does not hand over the responsibility: the Canadian Centre for Cyber Security's guidance says your organization "is legally responsible for protecting all customer data that is processed on your website, even if your organization does not host the website".

Keep those accounts in your own name and give the maintainer named, multi-factor access you can revoke. If a previous developer holds the only login, nobody else can renew, restore or move the site.

A laptop holds a large ring of three keys linked to a globe, a server and a cloud inside a dashed outline. A second laptop with a gear holds one small key that reaches only the server through a padlock.
Fig. 5 The owner holds the master keys; the maintainer holds named access that can be revoked.

A maintenance agreement should make the work checkable. These are the clauses worth writing down:

ClauseWhat to write down
ScopeSites, environments and components covered (core, plugins, theme, runtime, hosting), and what is not: new features, redesigns, content writing
Response and fix timesBy severity, in business or clock hours; whether "response" means an acknowledgement or a working fix; who covers evenings, weekends, holidays
Update windowThe day and time the maintainer may deploy, blackout dates such as campaigns, and which critical fixes go out of cycle
Staging and rollbackUpdates tested on staging first, key flows checked before and after, how a bad update is rolled back
Backups and restore targetWhat is copied, how often, where, for how long, who can delete it; recovery time and point per site (RTO vs RPO), proven by a timed restore drill
Monitoring and reportingWhat is watched and who is alerted; a monthly note on updates, failures, incidents, restore tests and upcoming expiry dates
Access and exitAccounts in your name, named engineers, credentials rotated at handover, and a final backup, runbook and change log at the end

What does website maintenance cost?

Prices differ too much by country, platform and provider to quote one honestly, so here is what moves the price. Ask every bidder to price the same tasks and agreement clauses.

DriverWhy it changes the work
Platform and extensionsEach plugin, theme or package is another stream of updates and advisories (plugins held 91% of 2025's new WordPress vulnerabilities in Patchstack's data), and nobody upstream patches custom code
Age of the siteWordPress says to consider upgrading incrementally across more than two major releases, so a neglected site may need a one-off catch-up first
What the site earnsA checkout, booking or login flow needs key-flow checks, a tested rollback and faster response than a brochure page
Response and coverEvening, weekend and holiday cover means someone is on call; ask what is included and what is billed per call-out
Environments and toolsStaging, off-site backup storage, monitoring and premium plugin licences; ask for these itemized
Change requestsNew features, redesigns and content work sit outside maintenance; agree how they are quoted

Who does the work is a trade-off. An in-house engineer knows the business but may be one absence away from a gap, and a freelancer can carry the same single-person risk. An agency or managed host can bring cover, a documented process and reporting, at the price of a contract to manage. Pay for more only where an hour of downtime or a missed update costs more than the work.

What happens when maintenance stops?

Nothing, at first, which is the trap. Each skipped task fails on its own timetable, and several failures give no visible warning.

Skipped taskWhat followsWhy you may not notice
UpdatesKnown flaws stay open; in Patchstack's 2025 data, about half of high-impact flaws were exploited within 24 hoursThe site looks normal
Certificate renewalThe validity period no longer includes the current time, which fails path validation under RFC 5280Renewal ran inside the server, and a reminder email may never arrive
Domain renewalFor generic domains, registrars interrupt DNS resolution where the registry permits, and after redemption and PendingDelete anyone can register the nameReminders may go to a mailbox on the expiring domain, or to someone who left
Backup testsThe first restore happens during the outage; CISA says to test backups regularlyA backup job's success message does not prove that a restore works
Version upgradesPHP 8.1 and older and Node.js 20 are past end of life; PHP 8.2 security support ends on 31 December 2026The site works, so nobody books the upgrade
MalwareCloaking can show search bots spam and owners a clean pageNothing looks wrong until rankings drop or customers complain

Patchstack's report, with data from Monarx, which runs server-level malware detection, says the dominant attack families all use cloaking: search engine bots see keyword-stuffed spam, human visitors are redirected to phishing sites or fraudulent stores, and security scanners and site owners "often see clean content". Check the site the way a crawler sees it. The hardening guide says an external, web-based integrity monitor can detect a defacement or added malware.

A server sends a plain page to a laptop on the left and a crowded page packed with link marks, in orange, to a magnifying-glass crawler on the right, past a switch set toward the crawler.
Fig. 6 Cloaking can show the owner a clean page and the crawler a spam page, so your own browser is not enough.

How do you start a website maintenance plan this month?

  1. List what you run: registrar, DNS host, hosting plan, CMS, plugins or packages, runtime versions, and who holds each login.
  2. Move the accounts into your name. Turn on multi-factor authentication for every administrator, and add a registrar contact that is not on the domain.
  3. Add outside monitors for the home page, one key flow, certificate expiry and domain expiry, alerting two people.
  4. Restore a backup somewhere else and time it. That time is your first honest restore target.
  5. Set up staging and run the first update round, recording what changed.
  6. Check your versions against the dates above. On PHP 8.2 or older, book the upgrade before 31 December 2026.
  7. Calendar the cadence, with a named owner for each row and an agreed monthly report.

If you would rather hand the routine over, our hosting and maintenance service starts with a read-only hosting health check: who owns the domain, DNS, certificates and accounts, the end-of-life and patch status of every component, a test restore of the current backup, and the risks ranked.