There Is Almost No Such Thing as a Nine-Year Job

Counts of active postings by stated experience requirement, 1 to 20 years, at the 2026-08-13 16:28 UTC snapshot. Denominator is the 101,397 postings that state a number, not the whole corpus. The collapse at 9, 11, 13, 14, 16, 17 and 19 is the finding: those are the values employers do not write.

Counts restricted to postings stating 1 to 12 years, at the 2026-08-13 16:28 UTC snapshot. Each row's ratios are computed inside that source only. usajobs is highlighted because it is the control: it runs through the same extraction code as every other row, yet its distribution is inverted at 4 versus 5 and nearly flat at 9 versus 10, because United States federal postings specify a pay grade rather than writing 'N+ years' in prose. Same parser, different writing convention, different shape. That is the evidence that the clustering is coming from how employers write, not from what our parser can read.

Source: Seeker job corpus, 247,239 active job postings, snapshot 2026-08-13. seekerscore.com/insights/years-of-experience-rounding-2026

The shape of the data is a short list, not a range

Sort the 101,397 stated requirements and the distribution is spiky rather than smooth. 5 years is the single most common answer at 22,190 postings (21.9% of everything stated). 3 years follows at 13,443, then 2 years at 11,375, then 10 years at 10,546. Now look at the values immediately next to the peaks. 9 years: 700 postings. 11 years: 202. 13 years: 203. 14 years: 222. 19 years: 32. A requirement of 10 years is written 52 times more often than a requirement of 11, and 15 times more often than 9. There is no plausible account of real jobs in which the work needed for a role changes that sharply between the ninth and tenth year. What changes is that a person writing a job description reaches for a round number.

Why this is not our parser inventing a pattern

A clustered distribution is exactly what a lazy text extractor would also produce, so this needed a control before it was worth publishing. Three checks. First, the extractor is demonstrably able to emit the unpopular values: it produced 9 seven hundred times, 11 two hundred and two times, 17 three hundred and ninety-nine times. It is not that those numbers cannot come out. Second, the pattern holds independently inside every large commercial source: the 5-over-4 ratio is 3.68x on himalayas, 3.19x on greenhouse, 2.65x on google, 2.47x on amazon, 2.30x on ashby and 1.96x on workday, and the 10-over-9 ratio runs from 10.7x to 19.6x across all six. A parser artifact would not have to agree across six unrelated feeds, but it would also not vary that much. Third and most useful, usajobs is a genuine negative control: same extraction code, 2,999 stated requirements, and its 5-over-4 ratio is 0.64x, meaning it asks for 4 years more often than 5, with 9 and 10 years almost absent entirely. Federal postings use a pay-grade system instead of prose year counts. When the writing convention changes, the shape changes. That is what tells you the shape belongs to the employers.

What to do if you have four years, or nine

The practical reading is that a stated requirement is a rounded label for a band, not a threshold measured against your resume. If you have 4 years, the number of postings asking for exactly 4 is 7,275, while 22,190 ask for 5. Restricting yourself to postings at or below your exact count removes the largest single bucket in the market for the sake of a rounding convention that the employer applied to themselves, not to you. The same holds harder at the top: with 9 years you are outside 700 postings and inside a 10-year band that contains 10,546. The corollary matters for how you write your own resume: rounding your own experience to the nearest common value, where it is honest to do so, puts you in the bucket recruiters actually filter on. And 60.5% of everything stated is 5 years or fewer, so most of the market is not asking for a decade in the first place.

The 59% of postings that say nothing

The largest group in this data is the one with no number at all. 145,842 of 247,239 active postings, 59.0%, state no year count that we can extract. Whether those roles are more flexible or simply vaguer, we cannot tell from a corpus, and we are not going to guess. What we can say is that the propensity to state a number is itself wildly source-dependent, from 96.6% of google postings down to 3.2% of reed and 6.4% of kaiser. That means the 41.0% figure is a property of our source mix as much as of the market, and it is the reason every percentage on this page is computed against the 101,397 postings that state a number rather than against the full corpus. The clustering finding survives this because it is a statement about the shape of the stated numbers, and we verified that shape separately inside each source rather than pooling them.

How we measured this

One snapshot of the jobs table at 2026-08-13 16:28 UTC, 247,239 active rows, of which 101,397 carry an extracted years-of-experience value. The per-source stated counts and the per-source ratios come from a second query taken 24 seconds later against the same table. Every percentage on this page divides by 101,397, the postings that state a number, and never by the whole corpus; where a count is per-source it divides by that source's own stated count. The requirement is what the posting asks for, not what the employer would accept, and we have no data on the second thing. We did not attempt to distinguish 'minimum 5 years' from '5 or more years' from '5 to 7 years', all of which reduce to the same stored number, so the clustering may be modestly overstated by ranges collapsing onto their lower bound. That mechanism would not produce the 52x gap between 10 and 11.

Frequently asked questions

Related Data

Guides

• Applying Below Your Seniority Level →
• Overqualified for Jobs →
• How to Explain Limited Experience →
• Why You're Not Getting Interviews →