Most hiring teams are pointed at the wrong target when they think about CV screening. The conventional advice (“make the first six seconds count”) accepts the broken default and tries to optimise within it. The fix is upstream of the CV itself.
Across the jobs Join’s customers multipost into Indeed, LinkedIn, XING, Stepstone, and the rest, the highest-impact hiring decision is not how much attention you give each incoming application. It is what you decide is worth attention in the first place. Filter the inbox on hard, binary, legally defensible facts. Then spend real time, five to ten minutes per CV, on the candidates who survived. The teams that hire well do both halves. The teams that struggle do neither.
The 7.4-second problem
The most cited number in this space comes from Ladders’ 2018 eye-tracking study, which tracked the eye movement of professional recruiters reviewing CVs and clocked the average initial screening at 7.4 seconds per resume (up from 6 seconds in 2012). The study is real, the number is real, and the response in the recruiting blogosphere has been almost entirely wrong.
The headline reading is “make those 7.4 seconds count” and the prescriptions that follow are about CV layout: bold the job titles, simplify the timeline, drop the headshot. None of that fixes the underlying problem, which is that 7.4 seconds is not enough time to read a CV at all. It is enough time to confirm a snap pattern-match the reviewer was already going to make. Optimising for 7.4 seconds is optimising for the lazy version of the work.
The right reading is: 7.4 seconds is the symptom of a missing layer. Teams that read 100 CVs in 12 minutes are not reading CVs; they are running an unstated knockout filter in their head, badly and inconsistently. The fix is to write the filter down, run it once at application time, and free up the attention budget for the candidates who deserve it.
Knockout questions: the layer most SMB teams skip
A knockout question is a structured application question with a binary answer designed to disqualify candidates who do not meet a hard, non-negotiable requirement. It takes the candidate thirty seconds to answer and zero seconds of recruiter time to process. The ones who answer “no” never reach the CV review queue.
Two practical points before the examples:
- Keep it short. One to three knockout questions per role, three maximum for high-volume positions. Every additional required field measurably reduces application completion, and most of the requirements people are tempted to encode as knockout questions are better as scoring criteria during the deep CV read.
- Stay on the legal side. The bar is set by Article 4 of the EU Employment Equality Directive 2000/78/EC: a difference of treatment based on a protected characteristic is permitted only where that characteristic constitutes a genuine and determining occupational requirement, the objective is legitimate, and the requirement is proportionate. Anything that doesn’t clear that test is not a defensible knockout question.
What good knockout questions look like
Five categories carry almost all of the legally defensible knockout filters at SMB scale.
- Work permit / right to work in the country of the role. Asked plainly: “Do you have the right to work in [country]?” The DE term is Aufenthaltstitel, FR titre de séjour, ES permiso de residencia. The question is binary, the answer is verifiable at hire, and the requirement is proportionate.
- Language proficiency required to do the role. “This role requires fluent German (C1 or above) to handle customer calls in Munich. Do you meet this requirement?” Language proficiency that is genuinely required for the work qualifies as an occupational requirement under Article 4. The phrasing matters: it has to be the role’s requirement, not a preference for native-language candidates.
- Hard-required certifications. Forklift licence, CPA, registered nurse, electrician’s licence, security clearance. Specific, verifiable, often legally mandatory for the work itself.
- Location and arrangement. “This is an on-site role in our Lyon office four days a week. Are you able to work on this basis?” Honest about the arrangement, binary about whether the candidate can.
- Minimum hard experience floor where the role genuinely requires it. “This role requires at least three years of hands-on experience administering Kubernetes in production. Do you have that?” Use sparingly: this category is the one most prone to becoming a preference dressed as a requirement.
These are all questions the candidate can answer truthfully in thirty seconds. None of them require interpretation. None of them touch protected characteristics.
What good knockout questions don’t look like
The pattern below is the most common way SMB teams accidentally cross the line. Avoid it.
- Anything that proxies a protected characteristic. “When did you graduate university?” (proxy for age). “Do you have family obligations that would prevent travel?” (proxy for parental status). “Are you willing to work on Sundays?” (potential proxy for religious observance). Even if the intent is logistical, the effect is exclusionary in a legally fraught way.
- Soft skills phrased as filters. “Are you a self-starter?” / “Do you thrive in ambiguity?”. Every applicant will answer yes, the question filters nothing, and you have given the candidate the impression you do not know what you are screening for.
- Culture-fit framings. “Do you embrace our values?”. Same problem, plus a candidate-experience hit.
- Salary history. After 7 June 2026, asking candidates about their salary history is no longer compliant under the EU Pay Transparency Directive (2023/970). If your current screening flow includes a “current/previous salary” field, that field is on a deadline.
The test: would you be comfortable defending this question in writing to a regulator? If yes, it’s a knockout question. If no, it’s something else.
What the deep read should look like
Once the inbox is filtered, the remaining stack is the one that deserves real attention. Five to ten minutes per CV, not seven seconds. The signals that predict, in rough order of weight:
- Relevant work samples or outputs. Code repositories, portfolio pieces, published writing, shipped products, named campaigns. Specific things you can click through.
- Specifics on past roles. What was shipped, what was measured, what changed. Compare “Led marketing for a B2B SaaS” (vague) to “Took outbound lead volume from 60 to 280 per quarter over two quarters by rebuilding the SDR cadence and switching ICP focus to mid-market” (specific). The second version is harder to fake and more predictive.
- Tenure patterns. A career with consecutive 18-month stints in similar roles is signalling something the CV is not stating. The pattern is not always negative, but it warrants a question in the interview.
- Adjacency to the role. Does the experience map to the job, or does it map to a different job that happens to have a similar title? SMB roles especially blur titles; the substance of past work matters more than the label.
What predicts much less than the SEO listicles claim: degree quality and prestige (compare Schmidt and Hunter’s 1998 meta-analysis putting years of education at .10 validity for job performance), gaps without context, and almost anything about CV formatting. Work-sample evidence is the strongest CV-readable predictor; Schmidt and Hunter put work-sample tests themselves at .54 validity, higher than any interview format.
What we see across multiposting boards
Spending time inside Join’s multiposting product is mostly seeing CVs sorted by board, day after day, across hundreds of customer accounts. A few patterns are stable enough to call out:
| Board (representative) | Pool characteristic | Implication for the screen |
|---|---|---|
| Indeed | Highest volume, broadest range. Many borderline-qualified applications. | Knockout questions earn their keep most here. Without them, the deep-read budget gets eaten by candidates who shouldn’t have made it past application. |
| Lower volume, more curated. Stronger profile data, weaker CV consistency. | The CV is less the document and more the profile next to it. Read both. | |
| XING (DACH) | Mid volume in DE/AT/CH, often more senior. Tenure patterns are more typical (4-7 year stints). | Less knockout-heavy, more deep-read at the experience tier. |
| Stepstone (DACH) | Mid-to-high volume, broader generalist pool. Strong on regional roles. | Similar profile to Indeed for screening purposes; knockout questions are useful. |
| Talent.com (FR/ES) | High volume in FR and ES generalist roles. | Knockout questions on language and location are the highest-impact filter. |
The cross-board observation: the same role posted to four boards produces four meaningfully different applicant pools, with different baseline qualification rates and different signal-to-noise ratios on the CV itself. A screening flow that treats every inbound the same is leaving real attention on the table. A screening flow with one or two well-chosen knockout questions per role does most of the work that “make six seconds count” cannot.
Where AI helps, and where it stops
Drafting the screening-question copy is a 30-second task for a model and a reasonable productivity assist. Summarising a qualified CV into a structured pre-interview brief is also fine. Both fall inside the limited-risk side of the EU AI Act’s Annex III classification for recruitment AI.
Auto-filtering applications based on an AI-generated match score is on the other side of that line. The Act’s Annex III explicitly covers AI used “to analyse and filter job applications”, with full high-risk obligations (risk assessment, bias testing, human oversight, transparency) applying from 2 August 2026. A structured knockout question with a binary answer is not AI screening; it’s a candidate-provided fact. An AI ranker on top of CV content is. The difference matters legally as well as ethically.
The principle we ship at Join is the same one we apply across the product: AI is the assistant on the drafting side, not the decision-maker on the filtering side. Knockout questions are how you get most of the filtering benefit without crossing the regulatory line.
The shape of the screen that works
Put the layers together and the operational picture is straightforward. One to three knockout questions on the application, written to clear Article 4 of Directive 2000/78/EC. Five to ten minutes per CV in the qualified pool, looking for relevant work samples and specifics. AI for drafting the screening questions and structuring the deep-read summary, not for ranking candidates. No salary-history question after June 2026.
Filter first, read second, decide third. Most SMB teams flip the first two steps and wonder why their hiring is slow.
