Monitor source changes without losing the original link

The first job of an information flow is not to write an elegant summary. It is to retain what it found in a form you can inspect again. Without a source URL, tracing a sentence a week later becomes difficult. This guide describes records for a personal experiment monitoring three technical sources. It does not assume an automatic publishing service or guarantee that every source is correct.

Create a source card first

For each source, record its address, short description, expected content type and monitoring purpose. “Technology news” may be too broad; tracking a particular product’s release notes is more testable. Identify whether the source is official and which version it describes. Keep community commentary visibly distinct from vendor documentation so that later summaries retain that context.

At item level, separate the title, original link, publisher identifier where available, publication time and your retrieval time. Finding a page today does not mean it was published today. Leave missing source dates unknown instead of inventing them. Record time zones as well. Comparing the displayed date strings alone can produce the wrong ordering across sources.

Separate reading, comparison and notification

The n8n RSS Read node can read data from a published RSS feed. Begin with one feed and inspect the returned fields. An identifier you expect may not be supplied by every publisher. Save a sample before sending it to a summarization model. Seeing missing fields helps prevent fluent but unsupported explanations from filling those gaps later.

Comparison should distinguish an unchanged item, a new item, a changed item and an unavailable source. Failure to read a source does not mean there is nothing new. Finding a new record does not prove that the underlying event just happened. Preserve these distinctions in a status field instead of returning only an empty or populated list. Even a personal spreadsheet can make this behavior clear.

Choose duplicate keys for the context

If the publisher offers a stable item identifier, investigate how it behaves. Comparing only titles can miss an updated document with the same heading. Comparing only URLs may count tracking parameters as new pages. If you normalize addresses, explain which parameters are removed and verify that distinct content is not merged. Observe the result on a few concrete examples.

Where a source supplies ETag, consider conditional requests. Under appropriate conditions, a server can return 304 to indicate that the selected representation has not changed. That mechanism does not establish a claim’s accuracy or identify the editorial importance of a sentence. Test source support and keep failed requests separate from unchanged content. Network efficiency and editorial judgment complement each other but remain different responsibilities.

Prepare an inspectable change note

Before sending a long AI summary, show what changed, which source changed and when the previous record was collected. Label any summary as a separate interpretation layer. Make it easy to return to the source when a statement lacks support. If a comparison could not be completed, “change not verified” is more useful than an invented certainty.

Test by reading the same source twice, changing controlled sample content and treating one source as unavailable. Write the expected record for each case in advance. The experiment here sends no network requests; it compares four illustrative events. Choose an actual checking interval, retention period and recipient separately when building your own system. Otherwise a small flow can gradually become an opaque notification generator.

Rehearse with a sample table

Create three imaginary items: a newly published document, an older document updated at the same link and a source whose title stays unchanged while a note is added. Record the comparison field for each. This exposes why title-only monitoring can miss a change. Then add a fourth case where the source does not respond. Explicitly prevent that outcome from being labeled “no new content.”

Return to the table a week later. Can you still distinguish information supplied by the source from your own interpretation? If everything is buried in one summary column, reorganize it. Keep source URL, identity, retrieval time and commentary separate. Reconsider the interval against actual change frequency. Reduce noise by combining repeats into an inspectable record rather than deleting useful evidence. The result is a small observation log that can be questioned later.

Sources

• n8n RSS Read
• n8n Remove Duplicates
• MDN: ETag
• MDN: 304