A good planned-outage message says four things: that the work is intentional, when it started, when it ends, and what the reader should do in the meantime. Anything else — apologies, jargon, vague "we'll be back soon" phrasing — is filler that makes the page less useful than a blank screen with a timestamp. The examples below cover the places a maintenance notice actually appears: the full-page holding screen, the dismissible banner on a partially working site, the status page entry, the email or SMS notice sent in advance, and the internal message for staff. Each needs a different length and tone, and the full-page version needs a technical detail the others don't — the right HTTP status code.
Table of Contents
- The full-page maintenance screen
- Return the right status code
- Banners for partial outages
- Advance notice by email, SMS, and status page
- Phrasing that causes support tickets
- Who the page needs to reach besides humans
- Frequently Asked Questions
The full-page maintenance screen
This is the page a visitor hits when the whole site is down. Keep it to a heading, two or three sentences, a return time, and one alternative route to help. A short version that works for most sites: > **Scheduled maintenance** > We're performing planned upgrades. The site will be back by 3:00 a.m. UTC on 21 November 2026.
> Need help before then? Email support@example.com or check status.example.com. A longer version for a site with transactional users — a store, a booking system, a client portal: > **We're down for planned maintenance until 06:00 UTC** > We're migrating our database to faster hardware. Orders placed before 02:00 UTC are safe and will ship on schedule. You won't be able to sign in, check out, or download invoices until we're finished. > Live updates: status.example.com The second one is better not because it's longer but because it answers the question the first leaves open: *did my thing go through?* If your outage touches money, bookings, or submitted data, say what happened to in-flight work. Give an absolute time with a timezone, not "for the next two hours." A cached page saying "two hours" is wrong the moment it's served, and a reader in another country can't convert "overnight.".
Return the right status code
A maintenance page served with HTTP 200 tells search engines the page is your real content. Served with 404, it tells them the URL is gone. Both are wrong, and both can affect what's indexed if the outage runs long or recurs. Use **503 Service Unavailable** with a `Retry-After` header. Google's own guidance on site downtime has long recommended 503 for planned maintenance, precisely so crawlers treat it as temporary and come back rather than dropping the URL.
`Retry-After` accepts either a number of seconds or an HTTP date: WordPress handles this for you in some paths and not others. The `.maintenance` file core drops during an update produces a 503 automatically; a maintenance-mode plugin or a hand-made holding page may not. Check with `curl -I https://example.com` before you walk away — the first line of the response tells you what crawlers and uptime monitors are seeing. Drupal's maintenance mode likewise serves a 503 on the standard path, but a reverse proxy or CDN rule you added in front of it can overwrite that. One exception: if the outage is brief and your audience is purely internal, the status code matters far less than reachability. The 503 discipline is for public, indexed sites.
- `Retry-After: 3600` — try again in an hour
- `Retry-After: Sat, 21 Nov 2026 06:00:00 GMT` — try again at a fixed time
Banners for partial outages
Most planned work doesn't take the whole site down. A banner is the right format when the site loads but something is degraded — checkout is read-only, search is stale, uploads are disabled. Write the banner around the restriction, not the maintenance: Compare those with the version most sites ship: *"We are currently performing scheduled maintenance.
Some features may be temporarily unavailable."* It's technically true and operationally useless. A reader can't tell whether to wait, retry, or call. Make the banner dismissible but have it return on a new page load during the window, and keep it out of the way of the primary action. A fixed banner that overlaps a mobile header is its own outage.
- *Checkout is unavailable until 05:00 UTC while we upgrade payments. Browsing and account access work normally.*
- *Search results may be up to 12 hours old until our index finishes rebuilding tonight.*
- *File uploads are paused until 04:00 UTC. Anything you uploaded before 01:00 UTC is unaffected.*
Advance notice by email, SMS, and status page
Advance notice has a different job from the holding page: it exists so the reader can reschedule their own work. That means the window, the impact, and a hard deadline for anything they must do first. Email, sent a few days ahead: > **Subject: Planned downtime Saturday 21 Nov, 02:00–06:00 UTC** > We're upgrading the platform this Saturday. During the window you won't be able to sign in or export reports. > If you need a report for Monday, please export it by Friday 20 Nov, 18:00 UTC.
> No action is needed otherwise. We'll post updates at status.example.com. SMS, where you have roughly 160 characters: > Example: planned maintenance Sat 02:00–06:00 UTC. Sign-in and exports will be offline. Updates: status.example.com Status page entries should be written as a sequence, not a single post — scheduled, in progress, completed — because that's the record people check afterwards. Post the completion note even when nothing went wrong; an entry that stops at "in progress" reads as an unresolved incident for as long as it sits there.
Phrasing that causes support tickets
Several common formulations reliably generate contacts. "We'll be back soon" invites a message asking how soon. "Sorry for any inconvenience this may cause" spends a line on nothing.
"Due to unforeseen circumstances" contradicts the word *planned* and makes readers suspect an incident or a breach. A few replacements: If the window slips, update the stated time rather than leaving the original to expire. A notice promising a return at 06:00 that is still up at 09:00 has stopped being a maintenance message and become evidence that nobody is watching.
- "undergoing maintenance" → "offline until 06:00 UTC"
- "we apologise for any inconvenience" → "here's what still works: …"
- "please try again later" → "try again after 06:00 UTC"
- "technical difficulties" → name the system: "payments", "sign-in", "search"
Who the page needs to reach besides humans
Three non-human audiences read your maintenance page, and each one reacts to something different. Crawlers read the status code, as above. Uptime monitors read the status code too — a 503 will fire your alerting, so either add a maintenance window in the monitoring tool or expect the pages. Some teams keep a separate health-check path that stays 200 behind the maintenance rule so monitoring can distinguish "intentionally closed" from "actually broken." Load balancers and CDNs sit in front of everything and will happily cache the maintenance page.
Send `Cache-Control: no-store` (or a very short max-age) with the holding page, and purge the cache when you come back up. A maintenance page cached at the edge for an hour after the work finished is the most common way a successful deployment still looks like an outage. Finally, anything that calls you over an API needs a machine-readable answer, not HTML. Return 503 with a small JSON body and `Retry-After` so client retry logic backs off correctly instead of parsing a holding page and failing in a way nobody can diagnose.
Frequently Asked Questions
How far in advance should I announce planned downtime?
Match the notice to the disruption. A four-hour outage affecting sign-in and exports warrants several days' notice plus a reminder the day before; a ten-minute restart at 3 a.m. usually needs only a status page entry.
Should the maintenance page still load my site's CSS and fonts?
Keep it self-contained — inline the styles and skip external assets. The holding page often sits in front of the same infrastructure you've taken offline, and a notice that can't render is worse than plain text.
Can I leave a maintenance page up for several days?
Avoid it. A long 503 eventually leads crawlers to treat URLs as unavailable, and visitors stop returning. For work spanning days, serve a reduced but real site and use banners for the parts that are off.




