AI website change alerts after OnWebChange: summarize or judge

An AI website change alert can mean two quite different things, and the gap between them explains why some people are leaving OnWebChange even though it has AI features of its own. In one design, AI rewrites an alert that a rule has already decided to send. In the other, AI decides whether to send the alert at all. This piece looks at both, with OnWebChange and AyeWatch as the examples.

The old pipeline: select, compare, notify

Page monitors have worked the same way for years. OnWebChange, which has run since 2009, is a clean example:

1. Fetch the page on a schedule, including pages that build content with JavaScript.
2. Cut out the part you selected with a CSS selector.
3. Compare it with the last copy.
4. Apply rules: a keyword must appear, a number must pass a condition.
5. Notify by email, Pushover or a URL callback.

Every step is deterministic. That is a strength. You know exactly why an alert fired, and a rule on one converted number is precise and cheap to run. OnWebChange's paid plans start at €0.89 a month.

The weakness is that none of these steps knows what the text means. A keyword rule sees the word "discontinued" in a customer review the same way it sees it in a product status line. A selector does not know that a redesign moved the price into a new block.

AI as a summarizer

OnWebChange lists AI summarization and AI translation. In the pipeline above, that slots in after step 4: once a tracker fires, a language model can condense a long text change into a few lines, or render it in another language.

That is useful. Reading a summary is faster than reading a diff. But the decision about whether to alert you is still made by the selector and the rule. If those fire on noise, you get a well-written summary of noise.

AI as the judge

AyeWatch moves the model earlier. You give a webpage monitor a URL, or several, and a rule written as a sentence:

"Alert me when this regulator's guidance page adds or removes a requirement for AI model providers. Ignore formatting changes and new links in the sidebar."

There is no selector and no keyword list. Each time the page changes, AyeWatch judges the change against the sentence and alerts only when it matches, with a short summary of what happened. Its structured data describes this as semantic alerts rather than keyword matching.

Two practical effects follow. Layout changes stop breaking monitors, because nothing points at a DOM path. And the rule can express things a keyword cannot, like "a new requirement" or "a price increase but not a currency change".

The same idea extends past single pages. AyeWatch also runs topic monitors across news, blogs, forums, government pages and social sources such as X, Reddit and YouTube, where there is no fixed page to select from at all.

The costs of putting a model in charge

Judgment by a model is less transparent than a rule. You cannot always predict at the edges what it will treat as a match, so the first week with a new monitor is partly spent tightening the wording. It also costs more to run than a string comparison. AyeWatch prices this in monitoring credits: most checks use one, the app shows an estimate per day before you save, and plans run from a Free Preview of three lifetime runs to Pro at $9 a month and Pro+ at $19 with the fastest ASAP checks.

There are capability gaps too. OnWebChange tracks PDFs and files directly and reaches login pages with custom headers. AyeWatch covers the public web and does not list PDF tracking.

Which design fits which job

What this says about monitoring tools

Adding a summarizer to an existing tracker is the easy step; OnWebChange and Visualping both list AI summaries. Handing the alert decision to a model is a bigger change in how the product works, and it is where the noise problem actually gets solved. AyeWatch publishes a feature comparison with OnWebChange for readers who want the detail.