Why a page change alert fires when nothing changed
Published 19 October 2026
Short answer: most false alerts come from parts of a page that change on their own: clocks, visitor counters, "posted 3 minutes ago" labels, adverts, cookie banners and rotating "recommended" blocks. A good monitor ignores the first group and tells you about real edits. None of them is perfect, so choose the narrowest watch that answers your question.
You set up a watch to hear about a real change, and you get an email because a counter went from 1,234 to 1,301. After two or three of those you stop reading the emails, which is worse than having no watch at all. This guide is about why it happens, which noise Osmanth Watch ignores, and which it doesn't, with checks we ran on 19 October 2026.
Where the noise comes from
- Things that count or tell the time. Clocks, "last refreshed" lines, visitor counters and relative times such as "3 minutes ago".
- Things that are not the page's content. Cookie banners, adverts, and the navigation, header and footer, the page regions that the HTML Standard defines for links, introductory material and footers.
- Things that rotate. Sliders and carousels, where the W3C notes that a carousel can advance automatically, and "recommended for you" blocks that show different items each time.
- Things that are real but not what you want. A news front page that changes every few minutes is working as intended, if you wanted to know about a headline and not if you wanted to know about one article.
What Osmanth Watch ignores
On a whole-page watch the product drops cookie and consent banners, elements marked as adverts, and the navigation, header, footer and sidebars. It also treats dates, times, long numbers and counters as the same from one look to the next. These are rules of thumb, not a guarantee.
We tested them on 19 October 2026. For each row we made two versions of a small page that differ in one thing, and asked the product's own reader whether it sees a change.
| What differs between the two copies | Seen as a change? |
|---|---|
| A clock line: "Last refreshed 12:03" and "12:09" | No |
| A date: "Updated 5 October 2026" and "6 October 2026" | No |
| A counter: "1,234 people viewing" and "1,301 people viewing" | No |
| A relative time: "Posted 3 minutes ago" and "9 minutes ago" | No |
| The text of an advert (a block marked as an advert) | No |
| A cookie banner that appears | No |
| A new link in the navigation | No |
| A changed line in the footer | No |
| A reference code with letters and digits: "Ref 8F3A29B1C7D2E4" and another | Yes |
| The text of a "Recommended" block | Yes |
| A real edit: opening hours from "Monday to Friday" to "Monday to Thursday" | Yes |
| A real new notice added to the page | Yes |
The pattern is that the things that are plainly times, dates and counts are folded together, and anything that looks like ordinary text is not. A reference code, or a block of suggestions that changes, looks like text.
A two-reading test on nine real pages
On 19 October 2026 we fetched ten pages twice, eleven minutes apart (at 07:20 and 07:31 UTC), and ran the product's own reader and comparison on the two copies. One page returned an error page both times, so it is left out. These are the other nine.
- Identical both times (7): Python's downloads page (1,190 lines), a GOV.UK guidance page, the GOV.UK Japan travel advice page, an Allbirds product page (213 lines), Mozilla's careers listings (137 lines), the Wikipedia main page (554 lines) and the SAT dates page (74 lines).
- Different (2): the We Work Remotely programming jobs page and the BBC News front page. Both are feeds that publish all day. The jobs page went from 730 to 737 lines, with new jobs added and old ones removed. The news page swapped two headlines for two others.
On the jobs page, one of the changed lines was a relative time, "Latest post less than a minute ago" in place of "Latest post about 24 hours ago", which was not folded away. In this case it arrived together with real new listings, but it shows that a relative time can still leak through when it is worded in some ways.
This is nine pages, eleven minutes apart. It is not a false-alarm rate. What it does show is that stable pages were stable in the product's reading, and that the two that changed were feeds that really had new items.
What to do about it
- Watch a smaller thing. A price, a stock status, or a few words are far less noisy than a whole page. See how to monitor a webpage for changes.
- Choose one part of the page. If only a list matters, pick that block.
- Use words for feeds. For a busy page, ask for new content that mentions your words, rather than any change. How to get notified about new job or rental listings shows that on real pages.
- Rely on the second look. A change to a page is read again about ten minutes later and is only emailed if both readings agree, so a block that flickers once does not email you.
Limits to plan for
- A relative time can still look like a change, as in the jobs page above.
- Reference codes and suggestion blocks are text. If a page prints either, a whole-page watch will see them.
- Once a day. On the free plan a page is read once a day.
Starting one
Paste the page on the home page and read what the starting reading says it will watch. If it is a very busy page, choose a narrower watch before you give an email address. The limits are written out in the terms.