Only 2.56 percent of US job postings required AI skills in 2025, and premium estimates range from 28 to 62 percent. What works is evidence, not a tools list.
The short answer: far fewer employers require AI skills than the coverage suggests, the pay premium estimates disagree with each other by more than a factor of two, and what moves a hiring decision is a piece of work someone can inspect. A line on a resume saying “proficient in generative AI tools” is worth close to nothing in 2026, because everyone writes it and none of it is checkable.
What the posting data actually says
The 2026 AI Index Report from the Stanford Institute for Human-Centered Artificial Intelligence, in its economy chapter, reports that 2.56 percent of United States job postings required AI skills in 2025, using data from the labour market analytics firm Lightcast. By cluster, postings mentioning artificial intelligence stood at 1.70 percent, machine learning at 0.99 percent and generative AI at 0.41 percent.
Roughly one job posting in forty asked for an AI skill of any kind last year, and roughly one in 240 asked for generative AI specifically. The growth rates are real and steep, and the fastest-growing skill requests inside those postings are infrastructure and data oriented rather than prompt oriented: Python, scalability, workflow management, automation, data analysis and cloud platforms.
A posting that mentions AI is not the same as an employer that values AI skill, and neither is the same as a job where AI skill changes what you are paid. Postings are marketing documents, written by recruiters, copied from templates, and increasingly padded with fashionable terms because the hiring manager was told to sound modern. A mention costs an employer nothing.
The wage premium numbers do not agree with each other
Two credible organisations have measured the same thing and produced very different answers.
Lightcast published “Beyond the Buzz: Developing the AI Skills Employers Actually Need” on 23 July 2025, based on more than 1.3 billion job postings. It reports that postings requesting AI skills advertise salaries 28 percent higher, nearly 18,000 dollars more per year, and that 51 percent of them sit outside information technology, with generative AI mentions in non-technical roles up 800 percent since 2022.
PwC published its 2026 Global AI Jobs Barometer, analysing more than 1 billion job advertisements across 27 countries and territories. It reports a wage premium for AI skills of 62 percent, up from 57 percent the previous year, ranging from 16 percent in government and public sector roles to 118 percent in consumer markets, and AI job postings growing at 69 percent against 9 percent for the overall market.
Twenty-eight percent and sixty-two percent are not close. When two large-scale analyses of overlapping data differ by more than double, the honest conclusion is that the underlying quantity is poorly defined. “AI skills” is not one thing: a posting for a machine learning engineer and one for a marketing coordinator who should be comfortable with a chatbot both register as AI postings, and the first pays what it pays for reasons unrelated to the chatbot.
Lightcast supplies a further caveat itself. Its documentation of the advertised salary metric states that it does not estimate salaries for individual postings, only extracts what employers choose to disclose, that the number of postings containing salary information is much smaller than the total, and that for actual earnings it recommends occupational earnings data instead. The 28 percent figure describes what a self-selected subset of employers advertised, not what anyone was paid.
What employers say versus what employers post
Vendor surveys tell a much more dramatic story, and should be discounted accordingly.
The 2024 Work Trend Index from Microsoft and LinkedIn, based on a survey of 31,000 knowledge workers across 31 markets conducted in February and March 2024 by Edelman Data and Intelligence, reported that 66 percent of leaders said they would not hire someone without AI skills, and 71 percent said they would rather hire a less experienced candidate with AI skills than a more experienced candidate without them.
Set them beside 2.56 percent of postings requiring AI skills. Either two-thirds of leaders are declining to write their stated requirement into their own advertisements, or the survey measured an attitude rather than a behaviour. The second is far more likely: respondents asked whether they value a fashionable capability will say yes, because saying no sounds negligent. Note also who commissioned it. Microsoft sells AI products and has a commercial interest in the belief that AI skill is now a hiring requirement. That does not make the survey dishonest, but it is a different category of evidence from posting counts.
How many employers are even using AI
The most unglamorous and probably most reliable number comes from the United States Census Bureau’s Business Trends and Outlook Survey. Between 14 December 2025 and 3 May 2026, the share of firms reporting that they used AI to produce goods or services hovered between 17 and 20 percent, standing at 19.8 percent in early May 2026.
The size gradient is stark. As of 3 May 2026, 37 percent of firms with 250 or more employees reported using AI, and 32 percent of firms with 100 to 249 employees, while firms with fewer than 20 employees stayed under 20 percent. By sector, information stood at 39.7 percent and finance and insurance at 33.9 percent, against about 14 percent in retail trade.
That reframes the career question. If you are targeting large employers in information or finance, AI capability is increasingly part of the job. If you are targeting a 12-person firm in retail or construction, roughly four in five are not using AI at all, and a candidate who leads with AI enthusiasm is answering a question the employer has not asked.
The only part that matters: evidencing rather than listing
Every resume now says the same thing. The differentiator is whether there is an artefact behind the claim that a sceptical interviewer can poke at.
| What people write | What it proves | What to provide instead |
|---|---|---|
| “Proficient in ChatGPT, Claude and Gemini” | Nothing. Equivalent to listing a web browser. | One named task you rebuilt, the before and after time or error rate, and what you changed when the first attempt failed. |
| “Completed an AI certification” | That you paid for and finished a course. Most hiring managers cannot tell the rigorous ones from the rest. | The thing you built during the course, documented, with the course as a footnote rather than the headline. |
| “Used AI to improve productivity by 40 percent” | Nothing. The number is unverifiable and everyone has one. | A measurement someone else made: a manager’s note, a ticket queue before and after, a cycle time from a system of record. |
| “Prompt engineering” | Diminishing value. Models have become far more tolerant of imprecise instructions. | Evaluation work: how you checked whether output was correct, and what you did where it was not. |
| “Built an AI agent” | Depends entirely on whether anyone other than you used it. | Usage: how many people, for how long, what broke, whether it still runs. |
| “Familiar with AI governance and ethics” | Nothing unless you wrote a policy someone adopted. | A one-page policy you drafted, the objection it answered, and whether it was adopted. |
A worked example
Take a claims processor who uses a large language model to summarise medical reports.
The weak version: “Leveraged generative AI to streamline claims review, increasing efficiency.”
The version that survives a follow-up question: “Our team reviewed roughly 180 medical reports a week, averaging about 22 minutes each. I built a summarisation step with a fixed template of seven fields and ran it alongside manual review on 200 historical claims. It got the injury date and treating provider right consistently, but dropped prior conditions in about one case in eight, so I made that field mandatory manual entry. Review time then went to about 13 minutes and our quality audit scores did not move. It has been in use since March and two other teams copied the template.”
Notice what the second contains: a baseline, a method for checking, a specific failure found and handled, a decision not to automate part of it, and evidence that it outlived the person who built it. It is worth more than any certificate, because it is the kind of statement that is difficult to fabricate: it names a failure mode, and people inventing AI experience do not think to invent failures.
The same shape works in any function. What was the baseline. What did you build. How did you check it. What went wrong. What did you decide not to automate.
A decision rule for the resume
Include an AI skill only if you can answer a follow-up question about how you verified the output was correct. If you cannot describe your checking process, delete the line.
This does most of the work by itself. It removes the tool lists and the unverifiable percentages, keeps the things where you did the work, and prepares you for the question competent interviewers now ask, which is not “do you use AI” but “how do you know it was right.” A claim you cannot defend in a follow-up costs more than the line gained.
What to be careful about
Be sceptical of certification marketing. There is no accreditation regime for AI credentials, no consensus about which ones signal anything, and the incentives run toward selling more courses. A certificate is worth roughly what the artefact you produced during it is.
Be careful about generalising the wage premium to yourself. Both figures aggregate across occupations, and the PwC range, from 16 percent in the public sector to 118 percent in consumer markets, shows how much depends on which job you are applying for.
Be careful about the entry-level picture. The 2026 AI Index reports that employment for software developers aged 22 to 25 has fallen nearly 20 percent from 2024, and that one third of organisations expect AI to reduce their workforce in the coming year. PwC reports that entry-level roles exposed to AI are seven times more likely to require skills traditionally associated with senior positions, such as judgment and leadership. Both point the same way: the AI-exposed entry-level jobs that remain are harder to get into, and AI skill is becoming a floor rather than a differentiator.
Finally, do not overwrite your actual expertise. AI skills are spreading into finance, human resources, marketing and education, which means the valuable person is a finance specialist who can automate part of finance, not a generalist operating a chatbot. The domain knowledge is the scarce half.
Where these figures come from
Stanford Institute for Human-Centered Artificial Intelligence, 2026 AI Index Report, economy chapter, using Lightcast job posting data. Lightcast, “Beyond the Buzz: Developing the AI Skills Employers Actually Need,” July 2025, and Lightcast documentation of its advertised salary metric. PwC, 2026 Global AI Jobs Barometer. United States Census Bureau, Business Trends and Outlook Survey, December 2025 to May 2026. Microsoft and LinkedIn, 2024 and 2026 Work Trend Index reports, fieldwork by Edelman Data and Intelligence.
The honest ending
Most published evidence about AI skills measures what employers wrote in advertisements or what executives told a survey, and almost none measures what employers did when a real candidate sat in front of them. That gap will not close soon, because the people best placed to fund the research are the people selling the tools. What stays true regardless of how the numbers settle is narrower: a person who can show a specific piece of work, explain how they checked it, and name the part they decided not to automate is easier to hire than a person who lists six tools. That was true before any of this, and AI has mostly made the second kind of candidate more numerous.
General information, current as of September 2026.
Frequently asked questions
Do I need to learn coding to be AI fluent?
No. Most postings outside engineering mean tool fluency: prompting well, integrating assistants into research, writing, analysis, and knowing the limits. Coding helps for technical roles only.
Will saying I use AI make me look replaceable?
Framed correctly, the opposite. Position yourself as the person who multiplies output with the tools and judges their work critically, which is exactly who survives automation waves.
