Website Change Detection
Detecting that two versions of a page differ is the easy half. The useful half is deciding which differences are real, and being able to show what a page said three months ago.
Availability
Background, not a product page
This page explains how change detection works, whatever tool you end up using. TagPipes is building page change monitoring and it is not available yet; the monitor pages say so and this one is about the technique underneath it.
Detection is three things, and only one of them is hard
First, a baseline: a record of what the page said at a known moment. Without one there is nothing to compare against, which is why no monitor can tell you anything on its first visit.
Second, a comparison: fetch the page again and work out what is different. This is a solved problem and has been for decades. Comparing two documents is not where anyone struggles.
Third, judgment: deciding which of those differences a person would call a change. This is the entire difficulty. A typical commercial page differs from itself on every single load, because of timestamps, rotating content, session identifiers, advertising and cache-busting parameters. A detector with no judgment reports all of it, is correct every time, and is useless.
What you can compare, and what each layer tells you
The same page can be compared at several levels, and they answer different questions.
| Compare | Catches | Misses | Noise |
|---|---|---|---|
| Raw HTML | Everything, including changes invisible to a reader | Nothing, which is the problem | Very high. A build that reformats the markup registers as a total rewrite |
| Rendered text | Wording, prices, headings, calls to action: the things a person reads | Layout and styling changes that move things without changing the words | Moderate, and manageable once the obvious rotating parts are excluded |
| A specific region | Changes in the part you actually care about | Everything outside it, including a change that moves the part you care about | Low. This is the strongest lever there is |
| A rendered image | Anything visible, including layout and styling | Nothing visible, but it cannot tell you what the words were | High. A different advertisement makes the whole picture different |
Most practical monitoring compares rendered text, narrowed to a region, and keeps an image alongside as evidence rather than as the thing being compared.
Where the false alarms come from
In rough order of how often they are the reason somebody turns a monitor off.
1. Content that rotates by design
Testimonials, featured posts, related articles, carousels and advertising are different on every load and were never meant to be stable. They are the largest single source of noise on commercial pages.
2. Time
Timestamps, copyright years, countdown timers and relative dates such as two hours ago. Always different, never what anyone meant.
3. Identifiers in the markup
Session tokens, cache-busting parameters, build hashes and generated class names. Invisible to a reader and a total rewrite to a naive comparison.
4. Personalization
Pages that differ by location, device, consent state or returning visitor. The page did not change; who was asking did. This one is genuinely hard to tell apart from a real change.
5. The page not loading properly
A slow or partially failed load looks like a page that lost most of its content. Worth catching, but as an error rather than as a content change.
Why the evidence outlives the alert
The alert is read once and forgotten. The record is what somebody asks for months later.
What it said, not that it changed
The wording as it was, kept. Being able to quote it is a different capability from having been told a change occurred.
A dated window
Bounded by the check before and the check after. A change monitor cannot tell you the minute, and the width of that window is set by how often you look.
A picture of the page
More convincing than a description, particularly to somebody who was not watching and is now deciding whether to believe you.
A sequence, not a snapshot
The third change makes sense only in the context of the first two. History is what turns a stream of alerts into an account of what happened.
Observed, not asserted
A record says what was seen at a stated time by a stated method. That is a much stronger and much more defensible claim than a summary of what changed.
Comparable over time
Evidence collected the same way each time can be compared. Evidence collected differently each time is a collection of anecdotes.
What change detection can and cannot tell you
Worth knowing before you build a process on top of it.
It can tell you
- That the text on a page differs from the last time it was checked
- Which part differs, and what it used to say
- The window the change happened in, bounded by two checks
- That a page stopped rendering, or lost most of its content
- That a page you do not control changed without anybody telling you
It cannot tell you
- Why it changed, or who changed it
- Whether the change matters, which is a judgment about your business
- The exact moment, unless your checks are frequent enough to bound it tightly
- Whether you are seeing a change or a personalized variant of the same page
- Whether the change has any legal or regulatory significance, which is a question for your own counsel
Automated testing does not replace a manual accessibility review.
Frequently asked questions
TagPipes monitors, configures and reports. It does not determine legal compliance and does not provide legal advice. Output is informational and reflects observed state at the time of observation. Rawsoft's expertise is technology, not law; please confirm any regulatory interpretation with your own counsel.
Tell me when change monitoring is ready
One email when it is available to use. No newsletter, no drip sequence, and we will not pass your address on.
More on page change monitoring
Watching a set of pages across a site, and watching one page closely.
What TagPipes does today
Change monitoring is not part of the product yet. Tagging automation, measurement, server-side events, consent management and privacy tooling are, and the trial gives you all of them.
See the TagPipes platformFree 14-day trial. No credit card. This page will change when monitoring ships.