A Correction: The Salary Apple Was Not Showing You, We Were Not Reading

Share of active postings carrying a structured pay figure. The top of this chart is not a transparency ranking. It tracks whether a feed hands us pay as its own field, which survives anything, or buries it in description text, which our 5,000-character cap can sever.

Reed is the row that breaks the tempting story. It carries the shortest text of any large source at 453 characters and still reports pay on 80.0% of postings, because its feed hands us pay as a structured field. Description length does not determine whether we see pay. Whether the feed gives us pay as data, rather than as prose we may truncate, does.

Source: Seeker job corpus, 257,534 active job postings, snapshot 2026-08-16. seekerscore.com/insights/who-shows-you-the-salary-2026

Correction, 2026-08-16

What this page claimed: 'Apple discloses on 1 of 4,948 postings through the feed we read from Apple's own careers site and 98 of 103 through TheMuse,' and 'Nationwide, 0 of 389 via oracle_cloud and 45 of 45 via himalayas. Lyra Health, 0 of 463 via lever and 38 of 58 via himalayas.' The page presented these as an employer-held-fixed experiment proving that the channel, not the employer, decides whether you see pay. What is true: Apple publishes a base pay range on its California postings, as California law requires. We sampled 14 Apple California listings live from jobs.apple.com and 10 of them carry an explicit range, including a Cupertino role advertising $144,600 to $263,800. Two of eight sampled non-California US roles also carry one, both in Seattle, which also has a disclosure law. The Apple zero in our corpus is our own data loss. Every per-employer claim on this page is withdrawn.

The mechanism, and it is not one bug

Two separate failures produced the same misleading zero. First, Apple. An Apple job page carries the pay figure in a Pay and Benefits block that sits outside the description field our connector reads, so we store the opening marketing paragraph and nothing else. The median Apple description in our corpus is 871 characters and 62% are under 1,000. Second, Amazon, where the description is complete but too long. We cap stored description text at 5,000 characters, and Amazon puts its pay range at the very end of a long posting. One Amazon row in our sample ends, mid-sentence, on the words 'base salary range for this position is listed below' before our cap severs it. The only Amazon 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.' The pay was there in both. We kept it in one.

Why our own control did not catch this

The original page ran what looked like the right check. It asked whether the pay was merely sitting in the description text unparsed, counted dollar amounts in the descriptions of postings with no structured pay, and found apple 0 of 4,947. That result was reported honestly and it was worthless, because it searched the text we had stored rather than the posting that existed. For Apple, the stored text is a 871-character teaser that could not contain a pay figure under any circumstances. A control that runs against your own transformed copy can only ever confirm your own transformation. The check that would have caught this is the one a reader performed from memory: open the employer's actual page and look.

What stands

Three things survive. First, the corpus-wide figure: 84,862 of 257,534 active postings carry a structured pay figure, 33.0%. That is now stated as a FLOOR on how many postings disclose pay, because we have just demonstrated at least two ways we lose a figure that was published. Second, sources genuinely do differ in what they hand us, and reed is the clean demonstration: it carries the shortest description text of any large source, 453 characters, and still reports pay on 80.0% of postings, because its feed delivers pay as structured data. Third, the TheMuse observation is real as a statement about TheMuse: it carries full posting text and consequently we see pay on Apple roles through it. What we cannot do is turn that contrast into a claim about Apple.

What we are no longer claiming

We are not claiming that salary transparency is a property of the channel rather than the employer. That was the thesis of this page and we could not support it, because the evidence we offered turned out to measure our own pipeline. We are also not publishing a per-source transparency ranking, and the chart above is deliberately relabelled to say what it actually measures: where a pay figure survives ingestion. Any source whose stored description is short or truncated has an unknown true disclosure rate, not a low one. That currently includes apple, amazon, oracle_cloud and lever. Finally, we are not claiming employers in general withhold pay at the rate our corpus suggests. Our 33.0% is what we captured, and the gap between that and reality is at least as large as the failures documented here.

How we measured it (and what we did not claim)

Corpus figures come from all 257,534 active postings on 2026-08-16. The Apple verification is a live sample, not a corpus query, which is the entire point: 14 Apple postings whose location is a California city and 8 whose location is elsewhere were selected at random from our own corpus, and each URL was fetched from jobs.apple.com and searched for a published range. 10 of 14 California and 2 of 8 other, both Seattle, carry one. That sample is small and we are not extrapolating a precise Apple disclosure rate from it. It is sufficient for the only claim it needs to support, which is that our published figure of 1 in 4,948 is false. Two limits on the correction itself: we have not re-derived what the corrected corpus-wide disclosure rate would be if we captured everything, because we cannot until the ingestion is fixed, and we have not audited every employer named on the original page individually. Nationwide and Lyra Health are withdrawn on the same reasoning as Apple rather than on separate checks, which is the conservative direction. The original figures are quoted unchanged above so the error stays auditable. Thanks to the reader who did not believe the number.

Frequently asked questions

Related Data

Guides

• Salary Negotiation →
• How to Read a Job Description →