How to conduct a salary survey: the process, the sources, and the legal limits

To conduct a salary survey: define the jobs you need priced, write a short content summary for each, choose your peer group and geography, pick sources you are allowed to use, collect the numbers, age them to today, then publish percentiles and build ranges. The hard part is job matching and documentation. The statistics are arithmetic.

There is one thing to settle before any of that, and most guides skip it. In the United States, how you collect the data decides whether the exercise is routine or a legal problem. That question has a different answer in 2026 than it had in 2022, so it goes first.

Is it legal to ask other companies what they pay?

Asking a direct competitor what it pays is the exact fact pattern US antitrust enforcers care most about. For nearly thirty years there was a safe harbor that made structured exchanges workable. The 1996 Statement of Antitrust Enforcement Policy said the agencies would not challenge a wage information exchange when four conditions held:

• a neutral third party ran the survey, not one of the participating employers;
• the data collected was more than three months old;
• at least five employers reported into each statistic, and no single one made up more than 25 percent of it;
• the output was aggregated so no individual source could be identified.

The Department of Justice withdrew that guidance on February 3, 2023, moving to case-by-case enforcement. On January 16, 2025, the DOJ and FTC issued new antitrust guidelines for business activities affecting workers, replacing the 2016 guidance for HR professionals. Those guidelines dropped the safe harbor language rather than restoring it, and they specifically address sharing competitively sensitive information through an algorithm or a third party.

Read that carefully, because it is routinely overstated in both directions. Salary surveys are not illegal. Reputable providers still build to those four conditions, and they remain sound practice. What changed is that meeting them no longer buys you a promise that the agencies will not look. The practical line for a small employer is clear enough: running your own survey against published sources is ordinary market research, and collecting current pay figures directly from competitors is where the exposure lives, with or without software in the middle. If someone invites you into a peer pay-sharing group, that is a conversation for your counsel before it is a conversation for HR.

This is also why the source of your data matters beyond accuracy. Public federal wage statistics involve no exchange at all: nothing of yours goes into a pool, and the provenance is a government dataset you can cite to a regulator or a board.

Step 1: decide which jobs you are actually pricing

Not all of them. A company of 120 people typically has 25 to 40 genuinely distinct benchmarkable jobs, and the rest are variations on those. Start from your org chart, group roles that share scope, and write down the list. Anything you cannot describe in two sentences of what the person does all day is not ready to be benchmarked yet.

Prioritize within that list. The jobs worth the most effort are the ones you hire for repeatedly, the ones where you have lost a candidate on money, and the ones in states that require a posted range. Everything else can be handled by grade once you have a salary structure in place.

Step 2: write job content summaries, not titles

Job matching is where surveys go wrong, and the error is invisible afterwards because the output still looks like a number. One company's Analyst is another's Manager, and titles inflate fastest at exactly the companies you most want to compare against.

For each job write three or four lines: what the person is accountable for, who they manage if anyone, what decisions they make without approval, and the level of experience the work needs. That summary is what you match on. The rule of thumb used by compensation analysts is a 70 to 80 percent content match, and if you cannot get there, price the job by slotting it between two jobs you can match instead. The full method is in our guide to market pricing a job.

Step 3: define the peer group before you look at any data

Decide in advance who you are comparing against, and write it down. Three dimensions do most of the work: industry, size, and geography. A 90 person software company in Denver competing for engineers against remote-first employers has a very different peer group from a 90 person manufacturer hiring locally.

Choose the cut first because choosing it afterwards is how you end up picking whichever number supports the decision you already made. Narrow the cut until the number of contributing employers gets thin, then stop narrowing. Geography deserves its own decision: see geographic pay differentials for how to handle multi-location and remote staff without inventing a policy per hire.

Step 4: pick sources you are allowed to use

There are four practical sources, and most companies end up blending two.

The third row is the one most people underuse. A dozen states now force employers to publish a good-faith range on every job ad, which has quietly turned the job boards into a live, lawful wage dataset. If you are pricing a role you would hire remotely, scanning what comparable remote roles are posting right now gives you a current read that no annual survey can match. Treat those numbers as hiring ranges rather than paid salaries, record the date you captured them, and keep a screenshot.

Step 5: age every number to today

Every source has a reference date, and it is never today. Survey data collected in March and published in October is seven months old on the day you open it. Public wage data runs about a year behind. If you compare a March number to an October number without adjusting, you have measured the calendar, not the market.

The fix is one line of arithmetic and one line of documentation. Pick an annual movement figure, divide it across the months between the reference date and today, and apply it. US salary budget surveys have been landing in the low three percent range for 2026, so a number with a reference date ten months ago carries roughly 2.7 percent of aging at a 3.2 percent annual rate. Write down the rate you used and where it came from. That note is what makes the range defensible when someone asks in a year.

Step 6: produce percentiles, then turn them into ranges

Report P25, median and P75 for each job rather than an average. Averages get dragged by one highly paid outlier, and with a handful of data points that outlier is often a single company overpaying to fix a hiring problem. If you have fewer than five contributing employers for a job, say so on the output rather than presenting a median as though it were solid.

The percentiles are not the deliverable. The deliverable is a midpoint and a range for each job: set the midpoint from the aged market rate at the percentile your compensation philosophy targets, then set a spread around it. Entry roles commonly run a 25 to 30 percent spread and management 40 to 50 percent, with the reasoning laid out in how wide a salary range should be.

Step 7: write down the method before you publish the numbers

One page: which jobs, which sources, which cut, which reference dates, what aging rate, who approved it, and when it gets refreshed. This is the least interesting step and the one that pays off. A posted salary range in California or Washington has to be a good-faith range you intend to pay within, and good faith is demonstrated with a documented method, not with a confident tone. The state-by-state requirements are in the pay transparency hub, and the mechanics of turning a band into an ad line are on job posting salary ranges.

When to stop and buy instead

Running your own survey makes sense when your jobs are common enough to appear in public occupation data, your peer group is defined by geography more than by name, and you need a defensible number rather than a competitive edge. That describes most companies under 200 people.

Buy instead when you need pay for a named peer set, when equity is a material part of the package, or when a board will not accept a number without a recognized survey behind it. The salary survey providers comparison covers who sells what, which ones publish a price, and how the four categories of provider differ. What a purchase does not remove is steps 1, 2, 5, 6 and 7, which are yours either way.

Four mistakes that waste the whole exercise

Matching on title. Covered above, and still the biggest one. Match on what the person does.

Choosing the cut after seeing the data. If you widen the geography because the local number came in high, you are not benchmarking, you are justifying. Decide first.

Averaging instead of using percentiles. With ten data points, one outlier moves the mean and leaves the median alone. Use the median.

Doing it once. A survey with no refresh date drifts below market in about 18 months, and the first symptom is offers being declined for reasons nobody writes down. Set an annual refresh and put it in the calendar the day you finish.

Frequently asked questions

How long does a salary survey take? Running your own across 25 to 40 jobs realistically takes three to six weeks of part-time work, and most of that is job matching. Buying a survey removes the collection time but not the matching, aging and range building, which is usually another one to two weeks.

How many companies do you need for a salary survey? The withdrawn 1996 safe harbor used at least five reporting employers per statistic, with no single one contributing more than 25 percent of it. That is no longer a legal shield, but it remains the sensible statistical floor. Below five, one outlier moves the median enough to make the number useless.

What questions go in a salary survey questionnaire? A job content summary rather than a title, base salary, bonus target and actual, equity where relevant, location, how many employees sit in the role, and an effective date. Titles alone produce noise.

What is the purpose of a salary survey? To replace opinion with a market rate you can defend. It answers what a defined job paid across a defined set of employers at a defined date, which is the input you need to set a midpoint, build a range, and post a compliant salary range in a job ad.