Find what the claim is about — occupation, area and period — then go to the published table for that combination and compare. If the source will not tell you which table it used, that is usually the answer. The check matters most for statutory figures, which change on fixed dates, and for any number that is round enough to look repeated rather than read.
Four questions that settle most claims
Who collected it? An employer survey, a self-reported platform, or job postings. Each samples a different population.
What period does it describe? Not the publication date — the reference period. A May 2025 survey published in 2026 is 2025 data.
Median or mean? Across large occupations the mean sits about 6.4 per cent above the median, and far more in skewed ones.
What counts as pay? Base only, or with commission, bonus and equity. For commission-heavy work the two can differ by a factor of two.
Most published salary figures cannot answer all four, and that inability is itself the verdict.
Follow the citation until it stops
Articles cite articles. Click through until you reach something that collected data rather than repeating it, and be prepared for the chain to end at a page that cites nothing.
Chains also mutate. A median becomes “average” at the second hop, loses its reference period at the third, and acquires a confident adjective at the fourth. By the time a figure is widely repeated it frequently describes something the original never claimed.
Read the methodology, not the summary
Official releases publish their scope, exclusions and definitions openly. That is where you learn that self-employed workers are excluded, that annual figures assume 2,080 hours, and that some cells are suppressed rather than zero.
None of that appears in a summary article, and all of it changes how a figure should be read. Ten minutes with a methodology page is the difference between quoting a number and understanding it.
Sanity checks that catch most errors
Does it beat the occupation’s own ceiling? If a claimed typical salary sits above the published ninetieth percentile, something is wrong — usually the figure is a mean, a total compensation number, or a different occupation.
Is the geography stated? Registered nurses run from $76,540 in Wichita to $216,740 in San Jose against a $97,550 national median. A figure without a place attached could be almost anything.
Does the spread appear at all? A source giving one number and no distribution has thrown away the most useful part of its own data.
The claims that deserve the most scepticism
Anything with a marketing purpose behind it. Training providers publish graduate salaries; recruiters publish market reports; platforms publish figures that make their own users look successful. None of that is dishonest and all of it is selected.
Ask what the publisher gains if you believe the number. If the answer is a course fee, weight accordingly.
What to do when the chain dead-ends
Discard the figure and look it up yourself. The official data is free, published at a fixed schedule, and searchable by occupation and metro, so there is rarely a reason to rely on somebody else’s summary of it.
That takes a few minutes and produces something you can cite with a date attached, which is worth considerably more in any conversation about your pay.
Citing a figure properly
Name the source, the reference period, the geography and whether it is median or mean. “The May 2025 wage survey puts the median for this occupation in this metro at X” is difficult to argue with.
Compare that with “I read that people in this role earn X”, which invites exactly the conversation you do not want — about your source rather than your salary.
Common questions
What counts as a primary source for a wage figure?
The published survey table itself, for the specific occupation, area and reference period. For a rule, the statute or the enforcing agency's own notice rather than a summary of it.
Three sites agree. Is that enough?
Not on its own. Sources copy each other, so agreement between three publishers frequently means one original and two repeats. Only the primary instrument settles it.
What if an official site contradicts itself?
Prefer the duller page. A technical bulletin, FAQ or taxpayer guide is consistently more accurate than the marketing page on the same domain.
How do I judge a source that stands alone?
Test it on a claim you can verify independently. If it is wrong there, discount it here — but do not assume the opposite of what it says, because that creates a second error from the first.
What if I genuinely cannot settle a figure?
Say it is unsettled. That is a legitimate finding, and it is more useful than a confident choice between two numbers you cannot tell apart.
How do I check a salary claim?
Ask who collected it, what period it describes, whether it is median or mean, and what counts as pay. Most published figures cannot answer all four.
What is the quickest sanity check?
Whether the claimed typical figure sits above the occupation's published ninetieth percentile. If it does, it is probably a mean, a total-comp number, or a different occupation.
Why do figures change as they get repeated?
A median becomes 'average' at the second hop, loses its reference period at the third, and gains a confident adjective at the fourth.