Web change detection: be told when a page or file changes

What gets compared

A change monitor fetches the page like a website check, removes scripts, style blocks, HTML comments and tags, collapses whitespace, and takes a SHA-256 hash of the remaining text. The first run stores that hash as the baseline and reports "first run" in its details. On every later run the new hash is compared with the stored one. If they differ, Tenium opens a "Content changed" incident, shows the monitor as degraded rather than down, includes the first 600 characters of the new text, and promotes the new hash to be the baseline. The next unchanged check closes the incident automatically.

So the behaviour is one alert per change, not a continuous complaint until you acknowledge. There is no side-by-side visual diff; you get an excerpt, the old and new fingerprints, and the page itself one click away. It is a tripwire, not a review tool.

Narrowing the watch with a selector

Whole pages change for many harmless reasons, so the optional selector field limits the comparison to one element. The grammar is intentionally tiny: #id, .class, a tag name, or a combination of tag with id or class, such as table.prices, section#plans or main. Nested selectors with spaces, attribute selectors and positional pseudo-classes are not supported, even though the editor's placeholder shows a longer example. If your element is deep in the page, give a container an id and select that.

If the element is not found, the check fails with 'Selector "#plans" not found' and an ordinary down incident opens. That is often exactly what you want when a redesign silently removes the pricing table, but be ready for it after planned template changes.

Taming noisy pages

A monitor that fires every hour is worthless, so look for volatile content before you commit. The usual noise comes from timestamps and "last updated" lines, CSRF tokens in hidden text, visitor counters, rotating testimonials and banners, cookie-consent text, and A/B tests. Remedies: select the stable block, monitor a plain-text or JSON URL instead of the styled page, or choose a different representation of the same data such as a feed. JSON and text files work well because there are no tags to strip; the hash covers the whole body.

The reverse limitation applies as well: only server-rendered text is visible. A price table filled in by JavaScript after load is invisible, so find the API or JSON file the page reads and watch that.

Good uses

Pages that should change rarely, and where a change matters:

• A supplier's price list, a shipping carrier's rates page, or a vendor's terms and privacy policy, so you hear about changes before an invoice does.
• Your own robots.txt: a staging rule like "Disallow: /" slipping into production can quietly remove a site from search results.
• security.txt, ads.txt and similar files that other systems rely on.
• A brochure site's home page, as a cheap alarm for defacement or unexpected edits by someone with admin access.
• Public notices that you would otherwise check by hand, such as a regulator's page or a client's job board.

What it will not catch

Change detection is not file-integrity monitoring. Because scripts and markup are removed before hashing, a malicious script tag injected into the page, which is how payment-skimming attacks usually work, leaves the text untouched and goes unnoticed. If you need to watch a script for tampering, monitor the script file's own URL; for most minified files the hash covers the whole content, but treat it as a tripwire, not a guarantee. Equally, modified images, CSS and anything outside the visible text will not trip it.

Respect the sites you watch: pick a sensible interval (hourly or daily is plenty for terms and prices), and only monitor pages you are allowed to fetch.

Settings that apply

The monitor uses the same request options as a website check: method (HEAD is treated as GET), expected status range, redirect following, timeout, basic auth, and on Pro custom headers. A WAF challenge is recorded as a blocked incident instead of being hashed, so a bot filter never produces a false "change". Change detection is available from the Starter plan ($7 a month), with 1 min the fastest interval there; the Free plan does not include it. To be told in the right place, attach contacts like for any other monitor, and read the evidence in the incident view described under incident management.