The Relevance Digest

Have AI read a noisy source on a schedule and email you only the handful of items actually worth your attention.

What it is/Overview

You've got a source you're expected to monitor. New public tenders. Competitor announcements. A busy trade newsletter. Job postings in your niche. Most of it doesn't apply to you, but you skim all of it anyway, because every so often something does and you can't afford to miss it.

This automation puts an AI in front of that stream. You tell it, once, what "relevant to me" actually means. From there on, it scores every new item against your criteria and hands back only what meets your criteria, each with a one-line reason attached.

The manual version is a prompt you paste items into. The automated version runs on a schedule and drops the digest straight into your inbox or Slack.

Why someone should implement it

The expensive part of monitoring is skimming the 90% that doesn't matter, to find the few that do. That skim is tedious and easy to put off, so what usually happens is one of two bad outcomes: someone drowns trying to keep up with the source, or they quietly stop checking it and start missing things.

Handing the first filter pass to AI removes that trade-off: you read a shortlist instead of the whole stream. The judgment about what matters is still yours, written down once as criteria, and the machine applies it consistently on the days you're too busy to look.

Expected results/outcome(s)

• A ranked shortlist instead of a raw stream, every item scored with a one-line reason it made the cut.
• A few minutes of reading where you used to lose the better part of an hour.
• A filter that keeps applying your standards on the weeks you're too slammed to check.

Step 1: Pick one source and define "relevant."

Start with a single stream. Then write down, in plain language, what makes an item worth your attention and what makes it noise. Be specific and a little ruthless.

This definition is the entire automation. Everything downstream is just applying it, so spend your effort here.

Step 2: Build your scoring prompt.

Wrap that definition in a prompt the AI will reuse every time.

That last sentence matters more than it looks; it's what stops the AI from padding your digest with things that sound plausible but were never there.

Step 3: Run it by hand first.

Before you automate anything, do it manually for a few days. Copy a batch of raw items out of your source, paste them under the prompt, and read what comes back. That's the whole test.

You're checking two things: whether it's catching the items you'd have flagged yourself, and whether it's correctly throwing out the noise.

If the shortlist misses your judgment, the fix is almost always the wording of your criteria. Tune it and run again.

This manual phase does double duty. It gets the criteria right, and it tells you whether the automation is even worth building before you spend an afternoon wiring it up.

Step 4 (optional): Automate the whole loop.

Once the manual version is clearly earning its keep, connect it so it runs without you.

What you'll need: a trigger that fires on new items (an RSS feed, a Google Alert pointed at a dedicated inbox, or a scheduled scrape), Zapier or Make to move the data around, the Claude or ChatGPT API to do the scoring with your prompt, and a place for the result to land.

Basic flow: new items show up, Zapier or Make collects them on your chosen schedule, sends them to the AI with your scoring prompt, and posts the filtered digest to your email or a Slack channel.

Set the cadence to fit the source. A fast-moving news feed might run daily. Public tenders might only need a weekly pass. Once it's set up, it runs on its own and you review a clean shortlist instead of a raw stream.

You're confirming two things: nothing relevant is being filtered out, and nothing fabricated is sneaking in. Once it's earned your trust, let it run and stop checking.

Summary

Start with one source, get the criteria right, and then automate. The manual version works today with nothing but an AI chat tool and a clear definition of what "relevant" means to you.

If you want to figure out which of your recurring "someone has to keep an eye on this" tasks are actually worth automating, and get them built to hold up, book a quick call.