The Job Posting You Are Reading Is a Lossy Copy

Most advice about where to apply is about applicant tracking systems and recruiter preference. We have no data on either. Here is an argument we can support: the listing you read is a copy of the real posting, copies lose things, and we can prove it because we are the ones who lost them.

How much is missing

Across roughly 250,000 live postings we hold:

• 9.5% carry under 500 characters of description. That is about a tweet and a half to decide a job application on.
• 59.5% state no years-of-experience requirement at all.
• One source runs a median stored description of 123 characters, with 74% under 300.
• Two other sources put every single posting under 500 characters. Another delivers descriptions hard-capped at exactly 500.

Every one of those is a statement about what reached our database. That distinction is the whole point of this post, and we learned it the expensive way.

We blamed the job boards. Twice. We were wrong both times.

First, Apple's salaries. We published that Apple discloses pay on 1 of 4,948 postings, framing it as proof that transparency is a property of the channel. A reader who had actually looked at Apple's careers page said that could not be right. We opened 14 Apple California listings and 10 published an explicit base pay range, one at $144,600 to $263,800. Apple discloses. The pay sits in a section our connector never read.

Then, a bank's job descriptions. We published that one employer appears at a median of 120 characters through one source and over 5,000 through another, a 41.7x gap, and concluded the board decides how much you are told. Then we checked. The 120-character source is the employer's own careers portal. A candidate opening that page sees a full job posting. What we had stored for one of those roles, in its entirety, was a single sentence:

"As a Private Client Banker, you are the point of contact for a select group of Chase's clients as well as customers in the Branch."

That is a clean summary sentence, not a truncation. It is what a short-description field looks like when something reads it instead of the body. Withdrawn, along with the conclusion drawn from it.

The one case where truncation genuinely is the mechanism

Amazon publishes pay at the very end of a long description. We cap stored text at 5,000 characters. One record in our corpus ends mid-sentence on the words "base salary range for this position is listed below" before the cap severs it. The only posting in that sample that came in under the cap, at 3,159 characters, ends with USA, WA, Seattle - 143,700.00 - 194,400.00 USD annually.

Same missing salary. Completely different cause. From the database, they are indistinguishable.

What actually predicts whether a field survives

Not description length, which is the intuitive answer and the wrong one. One source in our corpus carries a median description of just 453 characters and still reports pay on 80% of its postings, because its feed hands pay over as a structured field rather than burying it in prose. Fields delivered as data survive almost anything. Fields living in text survive only if every reader downstream is careful, and at least one of them was not.

What this means for you

Use aggregators to discover, the employer's site to decide. Boards are genuinely good at surfacing roles you would never have found. They are unreliable as the basis for a decision, because you cannot tell from a listing whether a field is absent or merely lost.

Never conclude "they will not say the salary" from a listing. Especially for large employers in states with pay transparency laws, where a range is very likely published somewhere. We made exactly that mistake in public, about Apple.

Treat a 200-character posting as a pointer. It cannot contain a requirement, a level or a salary, so its silence tells you nothing at all.

If the role is gone from the company's own careers page, that is your strongest signal. It beats any posting-age heuristic. Copies routinely outlive the original.

The honest limit

We cannot tell you whether applying through a company site rather than an aggregator changes your odds of a response. Our corpus cannot speak to that and we are not going to guess, because guessing confidently is how both of the errors above happened.

We are also not claiming aggregators are careless. We are one. We lost a large employer's salaries in two different ways and a bank's job descriptions in a third, and each time we only found out because someone checked the page instead of the database.