Set Up Alerts So Roles Come to You

Of everything in a job search, discovery is the part that most deserves to be automated. It’s repetitive, it has no judgement in it worth preserving, and doing it manually means the search only progresses on days you remember to go looking.

This is the opposite of automating the submission itself, which removes the judgement that makes an application work — the distinction argued in when to use auto-apply tools, and when not to. Automating discovery removes typing and gives you more things to make decisions about. Automating submission removes the decision.

So: build a feed, then triage it. Here’s how to make the feed good enough to be worth having.

Why manual sourcing fails

Not because it’s hard, but because of when it happens. Sourcing is the block that gets skipped — it feels like the least productive part of the week, since you finish it with nothing sent. So it slides, a fortnight passes in which you saw no new roles at all, and the applications you didn’t send show up as a quiet month six weeks later.

An alert flips the default. Instead of remembering to look, you have to remember to ignore things, which is a far more reliable human behaviour.

There’s a second benefit. Good roles are frequently filled from early applicants, so a role seen on day one is a different opportunity from the same role on day eighteen. Cutting your latency from a fortnight to a day changes what you’re competing for, and costs nothing after setup.

Build the queries badly first, then fix them

The mistake is trying to write one perfect query. You can’t, because you don’t yet know which words the roles you want are actually advertised under — and it’s routinely not the words you’d use for the job yourself.

Run three or four overlapping alerts instead, deliberately different in shape:

A title alert. The two or three titles that most often mean your job. Note that the same work is advertised under several titles, and the variation is worse across industries and countries than most people expect.

A skill or domain alert. The specific technology, tool, regulation, or subject matter that defines your work. This is usually the highest-precision query available, because it appears in the posting body rather than the title and it filters out roles that share your title but not your job.

A company alert, for the employers on your target list — see a target list of companies, not a feed of postings. Lower volume, much higher relevance.

A deliberately wide one. Broader than you’d normally search: adjacent titles, a nearby seniority level, the sector rather than the role. This exists to show you what you’re missing, not to be applied to directly — read it monthly and treat it as research into what your work is called elsewhere.

Then refine from evidence. After two weeks, look at what each alert delivered and adjust the wording. Most search interfaces support some combination of quoted phrases, excluded terms, and location bounds; the specifics differ everywhere and change, so read whatever the interface offers rather than trusting a recipe. What matters is the iteration loop, not the syntax.

Alert hygiene: measure precision, kill the bad ones

An alert delivering thirty results a day of which one is relevant gets unsubscribed from mentally long before you unsubscribe formally. That’s the real risk: a noisy alert doesn’t only waste time, it teaches you to skim past the whole channel, including the good ones.

So treat each alert as a channel with a yield, and track it crudely:

ALERTS

Name              Frequency   Per week   Worth a look   Applied   Verdict
Title: "analyst"  daily       ~40         3             0         too noisy — narrow or kill
Domain: "SQL +
  reporting"      daily       ~9          5             2         good, keep
Target companies  daily       ~2          2             1         best precision
Wide: sector      weekly      ~60         n/a           n/a       research only, monthly read

Review it once a month. Any alert where “worth a look” is consistently near zero gets narrowed or deleted. Two or three well-tuned alerts beat nine untuned ones by a wide margin, because you’ll actually read two or three.

One destination, not seven

The setup collapses if the results arrive in six different places.

Route everything to one inbox or folder — a filter that files all alert mail into a single label, out of your main view, is enough. Nothing about the search should be arriving in the inbox you use for everything else; that’s the subject of keep the search out of your main inbox.

Then the sourcing block becomes: open the folder, work through what accumulated, empty it. Bounded and finishable, which is what makes it survive a bad week.

Triage rules, so the block stays short

You want to spend seconds per posting, not minutes. Three outcomes only:

Bin it. Wrong level, wrong location, wrong content, or something you’d refuse if offered. Most of the feed. Don’t deliberate; a posting you’re unsure about after twenty seconds is a bin.

Add it as Found. A row in the tracker with the link and the posting text saved. Saved, not just linked — postings come down and you’ll want the text later.

Fast-track it. Occasionally something is a strong match at a company you want and it appeared today. Note it explicitly, and let it jump the queue in your next applying block.

And one rule that matters more than the other three: do not apply during the sourcing block. Sourcing is scanning and applying is focused work; mixing them produces half-tailored applications sent because the posting happened to be on screen. The separation is one of the load-bearing parts of a weekly cadence for a job search.

Alerts are not complete, and they fail silently

Two limitations to plan around.

Coverage is partial. No feed sees everything: employers post in different places, aggregation is uneven, and some roles are never advertised publicly. Alerts reduce manual checking; they don’t replace the weekly pass over your target list’s careers pages.

Alerts break quietly. Saved searches get reset, filters catch things they shouldn’t, mail lands in spam. The failure is invisible — no error, just nothing arriving, and you read an empty folder as a quiet market. Add one line to your monthly review: did each alert deliver something this month?

What not to automate on the sourcing side

Auto-saving everything. A tracker with two hundred Found rows you never triaged is a landfill, not a pipeline, and it makes the tracker feel like homework. Triage on arrival or bin it.

Anything that applies on your behalf, including the “apply to everything matching this alert” features that sometimes sit next to the alert settings. That converts a good automation into the exact thing this site argues against.

Anything needing your account credentials. Reading public postings doesn’t require deep access to your accounts; treat a request for it as a reason to check what the service does with your data.

The setup, in one sitting

Ninety minutes, once:

Write three or four queries in the shapes above.
Create them wherever you search, with alerts enabled, at whatever the shortest useful frequency is.
Add per-company alerts or watches for your A-tier target list.
Set up a filter routing all of it to one folder, out of your main inbox.
Put a fifteen-minute triage block in your week, and a monthly ten-minute alert review.

After that, discovery costs fifteen minutes a week and stops depending on your motivation on any given Wednesday — which is the point, because that motivation is the least reliable component in the whole system.