What happens when you negotiate a salary

Most people who ask for a higher number get one. That is the most defensible thing anyone can say about salary negotiation, and it is also the most misquoted.

It is also an American finding. The two surveys behind it are both US. In the research this post is built on, Canadian negotiation base rates were not found: no share of Canadians who negotiate, no share of those who succeed, no provincial breakdown. So read every figure below as imported, and read the mechanism as the part that might travel.

Here is what the better of the two surveys found. In April 2023, Pew Research Center surveyed 5,775 US adults, 5,188 of them employed and not self-employed, about the last time they were hired. 32% of men and 28% of women said they had asked for more than the offer. Of the people who asked, 28% got exactly what they asked for, 38% got something in between, and 35% got only what they were first offered. Two-thirds gained something. One in three gained nothing.

The numbers, and what kind of numbers they are

These two get quoted as though they were one number. They are not one number, and the research they come from says so directly: the survey methodologies differ, so the 66% and the 80% should not be merged into a single figure.

The difference is not cosmetic. Pew used its American Trends Panel, recruited through national random sampling of residential addresses and weighted to represent the US adult population. Employ commissioned Zogby Analytics to run a nationwide online survey of more than 1,500 US adults who were employed or looking for work, fielded 2025-03-25 to 2025-03-28, for a company that sells recruiting software. Pew separates the outcomes into three categories. Employ reports a headline success rate. One is a measurement of a population; the other is a marketing asset that happens to contain measurements.

Two further limits on reading either. Pew's denominator is people who were hired: their question looks back at the last hiring, so it cannot tell you the probability of getting an offer after you ask. And both are self-report, which means the number is what people remember happening, not what the employer's file says.

One useful detail from Employ: the rise from 29% to 37% is 8 percentage points across eight years, which the report describes as taking nearly a decade to gain. Negotiation is becoming more common at the pace of about a point a year. It is not a cultural tide.

How the employer's framing moves the number before you say a word

The actionable finding in this evidence base is not about your technique. It is about what the employer told you before you opened your mouth.

A field experiment by John List and Andreas Leibbrandt, reported by the University of Chicago, placed 18 online job ads for administrative assistants across nine US metropolitan areas between November 2011 and February 2012. 2,422 people expressed interest and gender was determinable for 2,382 of them. Each applicant was then randomly told one of two things: that the job paid $17.60 an hour, or that the salary was negotiable.

Faced with a fixed stated rate, 8% of women and 11% of men started a salary negotiation. Told the salary was negotiable, 24% of women and 22% of men did. The application rate moved too: on the fixed offer, 47% of men and 32% of women applied; on the negotiable one, 42% of men and 33% of women, which cut the gender gap in applications by about 45%. Adding four words to the employer's message changed who bothered to try, before any conversation about money happened.

Three limits, stated plainly. The manipulation measures whether people start a negotiation, not what they end up with. It covers one occupation, in 2011 and 2012, and the researchers hired 10 people. And the finding is causal only because the message was randomised; every salary survey in the table above is observational.

What survives those limits is the mechanism: how the employer frames the number constrains what you ask, and a stated rate does more of that constraining than an invitation. Nothing in your control is worth as much as knowing which frame you have been given.

In Ontario the frame is no longer entirely the employer's to choose. Since 2026-01-01, an employer with 25 or more employees on the day of posting must include expected compensation or a range in a public posting, and the range may not be wider than $50,000 a year, with no disclosure required where compensation or the top of the range exceeds $200,000. The employer chooses between a single figure and a range, and is not required to give both. What the posting does not do is bind the offer: the same guide states plainly that the requirement does not establish an obligation to offer employment at the level, or within the range, set out in the posting. A posted range is a frame, not a floor. The rest of Ontario's hiring rules are covered in interview questions Ontario employers cannot ask.

Width still matters even when disclosure is legal. Across three preregistered experiments, Kuhn (Journal of Applied Psychology, 2024) found that very broad advertised pay ranges lowered perceived organisational trustworthiness and skewed what applicants predicted they would be offered. That study is about the width of a posted range, not about the offer conversation, so it does not transfer directly. It does establish that a range can mislead about the offer, which is the opposite of what transparency is for.

What none of this measures

Here is the part that the guidance industry skips.

Every study above records what happened to people who asked. None of them records what happened to the people who did not ask, or to the people who asked and then lost the offer. Pew asks employed workers to recall their last hiring; it has no comparison group. Employ's survey has no comparison group. The field experiment randomised the employer's message, not the candidate's decision to ask.

So the downside of asking is not a smaller number in this evidence base. It is an absent one. There is no measured base rate for the probability that negotiating costs you the offer, because no study in this research pass was built to produce one. Anyone telling you that asking is riskless is selling something, and anyone telling you it will cost you the offer is making the same move in the other direction. Both are inventing a figure.

The honest version of the summary is: among people who asked, two thirds to four fifths got something, and the cost side is unmeasured. That is a real finding and it is more useful than a false one, because it tells you where to spend your attention. What you can control is the ask, the number behind it, and the evidence you bring. What you cannot control is how a particular manager reacts on a particular day.

What the evidence supports, and no more

• Know the number before you are asked. Employ's own report notes that candidates turn to peers and transparency platforms to gauge the market rate for a role before they are in the room, and since 2026-01-01 an Ontario posting from an employer with 25 or more employees has to carry the compensation or a range. There is no longer a good reason to reach the offer stage without the figure already written down.
• Ask for a specific figure, not a feeling. The reported outcomes cluster: two thirds of Employ's successful negotiations landed between 5% and 10%, and Pew's askers split into three outcomes rather than a smear. What these studies describe is a request for a number. "Whatever you can do" is not a category either survey measured.
• Put it in writing. This one carries the least evidence and the most convention. None of the sources above tracked the medium of the negotiation. The single field experiment ran its entire job-related conversation by email, which is a description of how the study was built, not a finding that email negotiates better. Written requests are the default in hiring administration; that is a convention, and it is a reasonable one.

The gap is the finding

A category that tells you negotiation is riskless is selling certainty that nobody has measured. The published record is narrower and more useful than that: asking usually pays, sometimes by a few percent, sometimes not at all, and the cost of asking has not been quantified in any of the research behind this post.

That last sentence is the honest core. It is also the thing that makes the rest of the numbers worth reading, because every figure in this article is American, the two surveys measure different populations in different ways, and the one thing you would most want to know is not in either of them.