The AI Wage Premium Is Measured on Postings, Not on People

How the Premium Number Is Actually Built

Two versions of this number circulate, both built the same way: an analyst splits postings within a labor-market dataset into two groups, one that names an AI skill and one that doesn't, then compares the advertised salary between the groups. Lightcast puts the gap at 28%, about $18,000 a year ¹. PwC puts it at 62%, ranging from 118% in consumer markets to 16% in government and public sector ². Different datasets, different years, same method, and neither number describes a raise.

Picture two postings for the same job title at the same company, one during a hiring push that names "experience with AI-assisted analysis" and one from eighteen months earlier that doesn't. Comparing their advertised pay isn't comparing the same role before and after a skill was added, because the role, the year, the labor market, and often the seniority bar changed alongside the wording. That is the comparison both published numbers are actually making, at scale, across a whole dataset rather than one company.

Lightcast does not publish a like-for-like adjustment behind its own figure ¹, and PwC states directly that its estimate 'does not control for other factors that may drive wage differences, such as education, experience, or location' ². The gap between the two headline numbers, 28% ¹ versus 62% ², is itself evidence of how much a raw comparison like this can move depending on dataset and year, before anything about AI changes at all. Neither figure is wrong, exactly. Each is answering a narrower question than the one most readers bring to it.

What the Comparison Can't Tell You

The comparison can't tell you what adding an AI skill would do to your own pay, because it never watched one person do that. It watched two different sets of job postings, and postings that name an AI skill already skew toward more senior levels, more urban markets, and higher-paying industries before the AI language enters the picture at all.

PwC's own page names this limit for its number; Lightcast's does not name it as explicitly, but the same selection problem applies to a posting-level comparison built the same way.

The number reflects something real about how employers price a posting today, though it still answers a narrower question than the headline suggests: employers who write AI skills into a posting today pay more for that posting than employers who don't, on average, across whatever mix of roles and seniority happened to fall into each group. That's a fact about the current market's pricing, not a promise about your next raise.

Treat the two published figures as separate, dated snapshots rather than a trend. The 28% and the 62% come from different datasets in different years, measured differently enough that averaging them or reading a rise from one to the other would be inventing a trend line that isn't in either source.

The One Question Worth Asking About Your Own Market

What the same reports show more usefully is where the requirement is concentrating, which you can check yourself. By December 2025, AI-related language had reached about 15% of marketing postings and 9% of human resources postings ⁴, both outside any technical field. Nationally, 2.5% of all US postings now name an AI skill directly, growing 55% in a single year ⁵. Small, and moving fast.

A separate Indeed measure counts a broader category, any AI-related language rather than a named skill, at 6.3% of US postings as of August 2026 ³. The two figures track different things and shouldn't be added together or read as one trend line; treat each as its own dated snapshot. By Indeed's earlier count, postings mentioning AI ended 2025 about 134% above their February 2020 level ⁴, a large growth rate produced by a small starting base. What both agree on is direction: check your own function's postings from a year ago against today's, and a rising share of AI language in the postings you'd actually apply to is the leading indicator worth more than either national percentage.

Reading for this signal takes less than an hour. Pull the last twenty postings in your target role and note which ones name a specific AI-assisted task rather than a generic "AI familiarity" line, then do the same search from a year ago if the postings are archived anywhere. A requirement that appeared in three of twenty last year and eight of twenty now is a local, checkable version of the trend the national percentages can only gesture at.

Should You Spend Money on This Number?

Not on the strength of the premium alone. Spend money on an AI course or certificate when postings in your own function are already naming the requirement, which you can check in an afternoon by reading fifteen recent postings for your target role. Spend it on the strength of a national percentage, and you're pricing a decision on a number that was never measuring you.

A rising premium doesn't mean aiming for a dedicated AI job title is the right move either; it's evidence that a skill is being priced into roles generally, which is a different claim than a specific title being worth chasing. Employers writing that requirement into a posting are told to name the task the assistant is used for rather than the tool itself, which is exactly the language worth searching for in your own field's postings before you decide the premium applies to you.

The honest use of a headline wage number is as a reason to look, not as a reason to spend. Look at your own field's postings first. Spend only once you've found the same signal there, in language specific enough to put on your own resume afterward.

See a sample report >