Most of what we believe about job advert wording does not survive contact with the data
Ask anyone in recruitment which words change how many people apply and you will get a confident answer. Say “no experience required” and you will be swamped. Hide the salary and applications dry up. Write it long and people give up halfway.
We were able to test some of that. Reed publishes a live application count on its adverts, which means it is possible to attach the size of the queue to the wording of the advert. We took 1,846 fresh UK adverts across 21 occupations and tested 16 features that are widely assumed to drive competition.
Fourteen of them did nothing measurable.
The hypothesis we expected to publish
Our starting hypothesis was that adverts welcoming candidates with no experience, or offering full training, would draw fewer applications. The first pass supported it: open-door adverts had a median of 10 applications against 16 for everything else.
Then we controlled for occupation, and it disappeared.
The reason is worth understanding, because it applies to almost every claim made about advert wording. Application counts vary far more between occupations than within them. A receptionist advert and a nursing advert differ by more than any phrasing ever will. So comparing every advert containing a phrase against every advert without it mostly measures which occupations happen to use that phrase. Care and warehouse adverts promise full training constantly. Nursing adverts do not. Our finding was largely a map of which sectors use the words.
Recomputed inside each occupation, on fresh adverts only, with near-duplicates collapsed, the open-door effect was a median difference of -0.25 applications across eight occupations, with a 95% confidence interval running from -1.5 to +4.5. Four occupations showed fewer applications, three showed more. Splitting by pay flipped the sign: below £25,000 the open-door adverts drew 20 applications against 11, while between £25,000 and £35,000 they drew 5 against 8.
That is not a small effect. It is no effect.
What did hold up
Two features separated cleanly.
Adverts requiring a driving licence were consistently less contested, by a median of 5.75 applications, holding in five of the six occupations where the comparison was possible. Some of the within-occupation gaps are large: customer service adverts requiring a licence carried a median of 1 applicant against 11 for those that did not.
This needs its caveat attached rather than trailing at the end. Roles requiring a licence often involve travel, field work or awkward geography, so they may differ from desk roles in the same occupation group for reasons that have nothing to do with the licence. We are not claiming that adding the requirement to an advert would thin the queue.
Customer-facing language showed a smaller effect in the same direction, three fewer applications, holding in 10 of the 14 occupations tested.
One near miss is worth reporting because of why we excluded it. Adverts posted by recruitment agencies looked strong, at 3.5 more applications with 11 of 12 occupations pointing the same way. Agency adverts in our sample were also a median of 14 days old when we collected them, against 11 days for direct employer adverts, and an advert that has been live longer has had longer to collect applications. When we split by advert age the sample fragmented and the direction went mixed among the freshest adverts. It might be real. This data cannot separate it from the age gap, so we are not claiming it.
Why HR teams should care about a null result
Three practical things follow.
The first is about diagnosis. If a vacancy is not attracting applicants, the instinct is usually to rewrite the advert. Our data suggests the wording is rarely the constraint. The occupation, the pay, the location and the hours sit far upstream of any phrase, and the gap between occupations dwarfs anything we could find inside them. A receptionist advert draws a median of 37 applicants and a nursing advert draws 1. No sentence in an advert competes with that.
The second is about the advice market. A large amount of recruitment guidance is built on vivid examples rather than comparisons. Everyone has seen an entry-level advert with 400 applicants. That memory feels like evidence, and it survives right up until someone checks whether similar roles behaved differently.
The third is about honesty in published research. We would have had a better headline if we had stopped after the first pass. Recruitment is full of confident numbers whose method is never shown, and the fastest way to tell the difference is to ask what the author does when the finding fails.
Method
1,846 adverts, drawn from 3,999 unique Reed records: 2,979 carry an application count, 2,216 were posted within 30 days of collection, 1,895 survive near-duplicate collapsing, and 1,846 sit in one of 21 occupations holding at least 30 adverts. A feature is tested inside an occupation only where both arms hold at least 10 adverts. Effects are median differences computed inside occupations, then combined, with confidence intervals from a bootstrap that resamples occupations rather than adverts. Feature definitions are frozen in a versioned dictionary so the analysis can be rerun exactly. It sits inside a wider dataset of 15,291 UK job adverts read in full.
The analysis is cross-sectional and describes adverts as they stood at collection. It is one job board over one month.