Which produces better outcomes?

A manual application takes somewhere between twenty minutes and two hours once you have read the posting, tailored the CV and written the answers. An automated one takes a minute of review, or none.

What you spend the time on is control: knowing exactly what went out, catching the answer that was wrong, choosing not to apply after reading properly. Automation buys back the hours and gives up some of that.

The third of those is the one people forget. A meaningful share of the value in reading a posting slowly is discovering you do not want the job — and an automated pipeline never delivers that outcome, because deciding not to apply is not a state it can reach.

Roles you genuinely want, applications with substantial free-text questions, small companies where a person reads every application, and anything where you have a connection or specific knowledge worth conveying.

In all of these the application is read carefully by someone who will notice the difference, and the difference is what decides the outcome.

Free-text questions deserve particular attention because employers increasingly use them deliberately as an automation filter. A question like "what would you change about our product" cannot be answered well without having used it, which is precisely why it is asked.

• Roles you would accept immediately
• Applications with real free-text questions
• Small employers where every application is read
• Anywhere you have a connection or specific insight
• Anything where the posting names the team or the problem

High-volume portals with structured forms and no free text, roles you are exploring rather than targeting, and situations where speed matters because the posting is new and the queue fills quickly.

It also wins on everything mechanical regardless of the role: tracking, follow-up timing, checking whether a posting is still open. That part has no downside at all.

The tedious re-entry case is worth calling out separately. A portal that demands you retype ten years of employment history it already parsed from your CV is not testing anything, and automating it costs you nothing in signal.

Reliable public data comparing the two is scarce, and anyone quoting a precise multiplier is likely selling something. What is consistent is the direction: applications that demonstrate specific engagement do better than generic ones, from every side that reports on it.

Since automation makes generic easy and specific hard, the effect of automating everything is predictable even without a number attached to it.

You can also generate your own evidence cheaply, which is better than any published figure. Tag each application as automated or written, leave it four weeks, and look at the response rate by tag. Fifty applications is enough to see a large difference, and a large difference is all you are looking for.

Something wrong going out in your name is the main one. A generated answer that overstates a qualification, names the wrong company, or claims a certification you do not hold is worse than no application, and it is worse still if you never see it.

Applying twice to the same employer through different sources is the common second one. It signals carelessness at best, and some tracking systems flag it automatically.

Both are avoidable rather than inherent. Deduplicate against employer rather than posting URL, keep a human review step on anything with free text, and read a random sample of what actually went out each week — because the alternative is discovering the problem in an interview.

• Overstated claims you never saw before they were sent
• Duplicate applications to the same employer via different boards
• Applications to companies you would refuse, wasting everyone’s time
• Answers that name the wrong company — the classic template failure

Automate discovery, screening, tracking and the mechanical parts of form filling for everything. Write the applications that matter yourself, with assistance rather than delegation.

That is not a compromise between the two — it is using each where it is strong. The failure mode is picking one approach and applying it to every role regardless of what the role is worth.

A workable rule: decide the tier at the moment the role is surfaced, not at the moment you apply. Deciding later means deciding when you are tired, and tired always chooses automated.