Screening questions are the cheapest filter in hiring and the most commonly wasted. Most application forms either ask nothing beyond a CV upload, or ask a long list of questions that deter strong candidates while filtering nobody.
Done properly, two to three questions remove a large share of applications that were never going to progress, before anyone spends time reading.
What a good screening question does
It has a right answer that you can act on without judgement. That is the entire test.
"Why do you want to work here?" has no right answer and cannot be screened at volume — it is an interview question on an application form. "Do you hold a valid Category C driving licence?" has a right answer, and it either qualifies the candidate or does not.
Good screening questions fall into four categories:
1. Hard requirements
Legal or physical prerequisites without which the person cannot do the job. Right to work in the location, required licence or certification, ability to work the shift pattern. These are binary and uncontroversial.
2. Verifiable experience thresholds
"How many years have you worked with [specific technology]?" as a numeric field rather than free text, so it can be filtered rather than read. Use these sparingly — years of experience is a weak proxy for capability, and over-filtering on it removes good candidates.
3. Logistical fit
Location, notice period, shift availability, willingness to travel. These do not measure quality at all and they resolve a large proportion of eventual drop-outs before either side invests time.
4. One short work-relevant question
A single free-text question tied to the actual work, capped at a few sentences. Not "tell us about yourself" but something like: "Describe a time you had to fix a process that was failing. What was broken and what did you change?"
This one is not automatable, but it is the highest-signal item on the form. It separates people who read the posting from people who applied in bulk, and at scale that is most of the filtering value.
Questions that do not work
- Anything answerable from the CV. Asking candidates to retype their employment history raises abandonment and adds nothing.
- Salary expectations as a hard filter. Restricted in a growing number of jurisdictions, and it filters on negotiating confidence rather than capability. Publish your range instead.
- Long free-text essays. Every additional paragraph raises abandonment sharply, and strong candidates with options abandon first.
- Self-rated skill scores. "Rate your Excel skills from 1 to 10" measures self-assessment calibration, which correlates with confidence rather than competence.
- Anything touching protected characteristics. Age, marital status, health, national origin. Where you collect demographic data for monitoring, it must be voluntary, separate from the evaluation record, and invisible to evaluators — see outcome testing.
How many, and where
Three to five questions total. Each additional question costs completions, and the cost is not evenly distributed — candidates with more options abandon soonest, so a long form filters hardest against exactly the people you want.
Put the hard requirements first. A candidate who cannot meet a legal prerequisite should learn that on question one, not after fifteen minutes.
Auto-rejection: use it narrowly
Automatic rejection is defensible on unambiguous hard requirements: no right to work, no required licence, unavailable for the only shift pattern. It is not defensible on experience thresholds or self-assessed skills, and using it there will remove good candidates silently while creating exposure you would rather not have.
A safer default: auto-reject only on binary legal or logistical requirements, and use everything else to sort rather than to exclude. Where automated evaluation contributes to the outcome, disclose it — see the compliance checklist.
Measuring whether they work
Two numbers, before and after any change:
- Qualified application rate — applications passing first review, divided by total. This should rise.
- Application completion rate — started versus submitted. This should not fall much. If it drops sharply, the form is too long and you are filtering on patience.
If the qualified rate does not move, the questions were not filtering anything and the problem is upstream in the job posting rather than in the form.
For configuring these in the application flow, see the pipeline.