# Workday Sees Prompting Fall While LinkedIn Sees AI Literacy Rise

> Workday's recruiting data show fewer requests for basic prompting, while LinkedIn sees broader AI literacy spreading through job ads. The gap changes how employers should define, test and fund AI skills.

- Published: 2026-11-03
- Author: Gene Dai
- Canonical: [https://digidai.github.io/2026/11/03/workday-linkedin-ai-skills-hiring-divergence/](https://digidai.github.io/2026/11/03/workday-linkedin-ai-skills-hiring-divergence/)
- Topics: Artificial Intelligence, AI Skills, Hiring, Career Development, Workforce, Deep Investigation

---

![Paper job requisitions with a speech bubble face three large use, review and workflow tiles beneath AI SKILLS, TWO SIGNALS.](/images/articles/workday-linkedin-ai-skills-hiring-divergence/cover-v1.jpg)

_AI-generated editorial illustration._

On October 5, Workday reported that mentions of basic AI skills such as simple prompting had fallen **25%** from a
January peak in job requisitions at roughly **550 employers** using its recruiting software. On October 27, LinkedIn
reported that the share of postings asking for **AI literacy** had risen about **65%** over the preceding year. A hiring
manager who copies either number into next year's staffing plan could get the job wrong. The companies count different
skills, employers and periods; neither headline tells that manager what a new hire will have to do on Monday.

A marketing analyst might use an approved model to summarize customer comments. The colleague who must check each quoted
comment against a source record carries a different responsibility. An engineer who changes the reporting workflow has
to keep it running after launch. An employer that writes all three jobs as "AI skills" can attract applicants without
agreeing on the work, its level or the training bill.

That ambiguity reaches inside the company. Workday found fewer internal moves at **57% of employers** in a matched
customer sample, while LinkedIn says members who list AI literacy skills are more likely to be hired and promoted. A
worker seeing "AI fluent" on an internal opening needs more than a course link: they need to know which task qualifies
and who will judge the work. Neither dataset proves that a course secures a promotion.

## Two datasets describe different slices of demand

Workday's
[October workforce release](https://newsroom.workday.com/2026-10-05-Workday-Global-Workforce-Report-AI-Is-Rewriting-Jobs-More-Than-Its-Cutting-Them)
analyzes de-identified skill mentions in requisitions from roughly 550 enterprise employers using Workday Recruiting
between September 2025 and July 2026. Its basic-skill series rose until January 2026, then declined 25%. Its separate
hands-on series, including building AI tools, automating workflows and AI engineering, rose **51%** from September 2025
to July 2026. The time windows are not identical. The public summary does not give the number of postings in either
series, or show that a hiring manager removed prompting from a role after deciding it had become ordinary.

LinkedIn's
[October 27 labor-market release](https://news.linkedin.com/2026/linkedin-data-shows-where-opportunity-is-emerging-as-work-changes)
uses a wider AI-literacy label. It says the share of job postings requiring that label rose roughly 65% in the last 12
months and is more than ten times its 2023 level. The release does not present a Workday-compatible count of
simple-prompting mentions. LinkedIn's [AI reporting methodology](https://pulse.linkedin.com/reporting-ai) treats a
member as "Using AI" when the profile lists at least one AI-literacy skill; it distinguishes that category from
"Building AI," which requires two engineering skills or an AI-building occupation. The exact job-ad taxonomy behind the
65% headline is not fully detailed on the public release page.

An employer might stop naming "prompt engineering" as a discrete requirement while asking project managers and writers
to use AI. Another might write a general literacy clause into every listing without changing the work. Or the two
platforms might simply see different employers and geographies. The public summaries do not provide the matched postings
needed to separate those explanations.

| Published signal                                                | Population and window                                                      | What it measures                                      | What it cannot establish                                     |
| --------------------------------------------------------------- | -------------------------------------------------------------------------- | ----------------------------------------------------- | ------------------------------------------------------------ |
| Workday basic AI skills: down 25% after January 2026            | Skill mentions in requisitions from about 550 Workday Recruiting employers | Change in one basic-skill category after its peak     | A 25% fall in all AI jobs, wages or useful prompting ability |
| Workday hands-on AI skills: up 51%, September 2025 to July 2026 | The same recruiting-customer system, a different skill group               | More mentions of building, automation and engineering | 51% more people hired or promoted into those tasks           |
| LinkedIn AI-literacy posting share: up about 65% in 12 months   | LinkedIn job ads, using its broader literacy category                      | Change in the share of ads asking for literacy        | A matched contradiction of Workday's basic-skill series      |

The table is a reading aid, not a pooled estimate. Workday sells hiring and workforce systems; LinkedIn sells a
professional network and recruiting products. Each sees a large part of labor demand, and each chooses the label through
which readers see it. A buyer deciding on headcount needs its own applications, assessments, hires and retention records
before a vendor's percentage can inform a staffing forecast.

## A skill name cannot carry the whole job

LinkedIn says the demand for AI literacy now extends into marketing, design, project management, administration and
writing. Its figure is about job postings, not a count of workers who have performed those tasks well. The network also
reports that only **1.2%** of its members list an AI skill. Self-reported profile skills are an uneven inventory: some
experienced users may not list one, while a listed skill does not verify depth. The distance between a posting keyword
and demonstrated work is the space where a hiring process must operate.

Indeed asked employers and job seekers to define the term rather than infer it from advertisements. In a
[YouGov survey commissioned by Indeed](https://www.indeed.com/news/releases/what-is-ai-fluency-at-work?co=US), conducted
July 23 to August 13, 2026, the US results cover **500 employers** and **1,007 job seekers**. Its proposed definition of
AI fluency, using tools effectively while judging their output and knowing when not to use them, won agreement from
**78% of employers** and **69% of job seekers**. Among employers, 68% cited generative AI use for writing, research or
summaries; 60% cited effective prompts; 58% cited reviewing output for accuracy, quality or bias; and 56% cited
automating repetitive workflows.

Those are survey responses about capabilities, not measured hiring requirements. The spread still helps explain the
apparent contradiction. An employer can consider prompting useful while writing fewer requisitions that name it as a
special skill. A job ad can ask for AI literacy without specifying whether the worker may publish an unchecked output,
must audit it, or is expected to change the process that produces it. The missing sentence is usually about
responsibility, not software familiarity.

Indeed says its engineers completed hands-on training to integrate AI coding assistants into their workflows. The
company presents this as its own practice, not as a controlled trial showing a performance increase. For a team outside
software engineering, the transferable lesson is narrower: training should use the job's actual materials and an
observable standard of good work. A certificate for attending an introduction cannot tell a manager whether a person can
detect an invented citation in a client brief or keep a sensitive record out of a model prompt.

Workday's recruiting figures show why that distinction is costly. Its release says the median number of applicants per
filled job climbed from **58 to 69** in a year, while time to fill stayed near **60 days**. Workday suggested easier
AI-assisted applications as one possible contributor to the added volume; it did not establish a causal share. More
applications can make a broad "AI fluent" checkbox appealing to a recruiter who needs to sort a queue. If the checkbox
does not match a task in the role, it can screen out a candidate who knows how to verify an answer while admitting one
who knows the vocabulary. A short work sample may demand more reviewer time, but it gives the manager something the
keyword cannot.

The skill requirement also changes the interview. Asking a candidate which model they use reveals familiarity. Asking
them to identify two mistakes in an output, explain which source would settle a disputed fact and decide whether to use
the tool at all reveals more about the work. For an engineering role, the exercise can include recovery when an
automated step fails. Those are proposed assessment designs, not tests that Workday or Indeed has validated across
employers. They should be short, job-related and reviewed for unequal access to tools or preparation time.

## Internal careers have to connect to the new requirement

The same Workday release reports that internal moves fell year over year at **57% of employers** in a matched customer
group, while global promotion rates were essentially flat. That is the share of employers experiencing a decline, not a
57% decline in moves and not a share of workers. Its September survey of about **6,000 full-time workers** found 79%
said they knew what skills they needed to succeed; 66% said their employer helped them develop those skills. Workday's
separate August AI survey found that 62% of heavy AI users expected their current skills to lose value, while 76%
expected new career opportunities. These are distinct samples and questions.

Our [earlier account of Workday's mobility product](/2026/10/09/workday-mobility-agent-internal-career-freeze/) examined
the opening and manager decisions that a match cannot make. The newer cross-platform comparison adds a preceding
problem. A company cannot fairly match a worker to an "AI-fluent" role if the recruiting team, manager and worker have
three different interpretations of that label. A vague requirement can create a skill gap on paper where the worker
already has relevant experience, or hide a real gap in output review and system ownership.

LinkedIn says members listing AI-literacy skills are **78% more likely to get hired** and **11% more likely to be
promoted** than members without them. Those are associations in its network data, not an experiment. People who add a
skill may differ in occupation, education, location, prior experience and willingness to update a profile. The same
release says over 90% of AI roles in 2025 went to workers with at least a bachelor's degree and women accounted for
about a quarter of new AI hires. A badge alone cannot overcome the rest of a selection system.

LinkedIn also reports that US members with AI skills move into occupations paying an average of
**$28,000 more** than occupations entered by members without those skills. It does not say adding a skill raises a person's salary by $28,000.
Occupation choice, prior education and geography may explain some of the difference. A worker deciding whether to pay
for training needs an opening and a credible path to it, not the average salary of people already moving between
different jobs. An employer using that figure to justify a training budget should separate an internal pay change from a
move into a higher-paying occupation.

There is a counterexample to the idea that every move has to start with a newly labeled AI vacancy. In an
[October 29 Indeed account](https://www.indeed.com/lead/the-vanishing-entry-level-role-and-how-to-save-it?co=US),
Chipotle people chief Lois Alexis-Collins argued for giving entry-level workers new tools and work rather than
eliminating the entry role. Indeed reported **23,000 internal promotions** at Chipotle in 2025; the company
independently stated approximately the same figure in its
[2026 proxy filing](https://www.sec.gov/Archives/edgar/data/1058090/000114036126017294/ny20063214x1_def14a.htm). Nearly
90% of restaurant leadership roles were filled internally, according to that filing. This is a restaurant career system,
not evidence that Chipotle's managers passed an AI skills test. It does show a company with a countable progression path
at a time when many labor-market stories stop at the application.

The distinction is useful for a technology employer. It can publish a project assistant job with AI requirements and
count applicants. Or it can let an existing assistant learn to check an AI-generated report, record the review work and
open a funded promotion. Both routes may work; only the second establishes an internal path. A promotion count by itself
still does not tell us whether workers were paid enough, stayed in the role or gained a skill that another employer
would recognize. Chipotle's number answers one part of the question, not all of it.

A requisition can change in days. A worker needs practice, feedback, a supervisor and a real assignment before a new
capability becomes credible. Updating job descriptions alone makes external hiring look faster than internal
development, partly because the internal cost is less visible. A training budget has to cover paid learning time and the
work a manager gives up while a colleague learns.

![A warm desk lamp lights a source sheet under a magnifier between an open work notebook and a three-block workflow.](/images/articles/workday-linkedin-ai-skills-hiring-divergence/interior-v1.jpg)

_AI-generated editorial illustration. The source check sits between using a tool and owning a workflow._

## Training access is uneven before any skills test

PwC's
[September 29 workforce survey](https://www.pwc.com/gx/en/news-room/press-releases/2026/companies-risk-losing-ai-savvy-employees.html)
reached **49,364 workers** across 48 countries and regions. It found that **64%** used AI at work in the preceding 12
months, up ten percentage points from the previous survey. Yet **51%** said they had access to the learning and
development resources they needed, down from 59%. Among the group PwC calls its "engine room," workers with less scarce
skills and lower AI adoption, only about two in five reported such access. The firm also found 29% of its high-AI-skill
"front-runners" said they were very or extremely likely to change employers in the next year.

These are self-reports, not proof that one training program would retain a worker or that AI caused someone to leave.
They challenge a cheap interpretation of the job-ad data. If an employer asks more roles to use AI but reserves
supported practice for already scarce specialists, it may advertise a companywide skill while building a narrow internal
pipeline. The question is where the worker can learn on paid time, with approved tools and a reviewer who knows the
task.

The US entry route has its own constraint. Indeed Hiring Lab
[reported in April](https://hiringlab.indeed.com/2026/04/23/new-grads-looking-for-work-the-struggle-is-real/) that
junior-level postings fell 7% year over year in 2025 while senior-level postings rose 4%. Its later
[July analysis](https://hiringlab.indeed.com/2026/07/23/the-labor-market-is-tilting-toward-seniority/) found entry-level
postings down 7.5% year over year as of May 2026. The two figures use different observation dates and should not be
joined into one trend line. Hiring Lab does not show that AI alone caused either change; broader hiring conditions
matter. A junior worker can be asked to exercise more judgment at exactly the time fewer openings offer supervised
practice.

The aggregate youth picture is less bleak than the job-ad series might suggest. The US Bureau of Labor Statistics
[reported in August](https://www.bls.gov/news.release/archives/youth_08202026.htm) that **53.8%** of people aged 16 to
24 were employed in July 2026, little changed from **53.1%** a year earlier. Their unemployment rate was **9.1%**, down
from **10.8%** in July 2025. These figures cover all industries and include people working outside the jobs Indeed
classifies as junior white-collar posts. They do not cancel the decline in that posting category. They warn against
describing a narrower entry route as a collapse of youth employment or assigning its entire cause to AI.

For the worker, the paths look different. A first job in a restaurant can build supervision and customer judgment
without the employer ever advertising "AI literacy." A junior analyst may need a portfolio of reviewed work before an
AI-labeled role will consider them. Someone must let that analyst handle a source document, find an error and defend a
correction while a senior colleague can still intervene. The employer that wants the finished skill has to create such
assignments. Otherwise it keeps asking applicants for evidence that its entry jobs no longer help them produce.

Indeed's October 29 account gives the older version of that problem a name. Devon Jennings, now a talent acquisition
marketing manager, recalled that his first role gave him routine work and internships that an employer today might
expect a graduate to complete with AI. His account is an interview published by Indeed, not one conducted by this site.
It does not establish how many graduates now lose that first step. It does make the apprenticeship question concrete: if
software handles the early draft, who lets a beginner inspect the messy source, speak with a customer and learn why the
draft was wrong? An employer can preserve those learning moments, but only by assigning someone to review them.

That mismatch can be addressed inside a role. A manager can assign an entry-level worker a small, reversible AI-assisted
task, keep the source material available and review the decision rather than only the polished output. The worker can
then show how they corrected an error and when they decided not to use the tool. This costs review time. An employer
that will not fund that time should be cautious about advertising a higher skill bar as if applicants could acquire it
alone.

## A three-level test for a real requisition

The following matrix is an editorial framework derived from the different task definitions in Workday's release,
Indeed's employer survey and LinkedIn's literacy/building distinction. It is not a vendor certification or a validated
predictor of job performance. Its purpose is to make a hiring manager state the task, the proof and the budget before a
broad "AI skills required" line goes live.

| Level in the actual job    | Candidate or employee evidence                                                                  | Manager review                                                                          | Budget owner                                                     |
| -------------------------- | ----------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- | ---------------------------------------------------------------- |
| Use an approved tool       | Complete a representative task; name the source material and tool limits                        | Check whether the output is useful and whether sensitive material was handled correctly | Team manager for access, orientation and practice time           |
| Judge an output            | Find a planted factual or quality error, verify it against a source and explain a stop decision | Review reasoning, corrections and escalation rather than prompt elegance                | Team manager plus a subject expert for review hours              |
| Build or change a workflow | Map inputs, approvals, failure handling and a measurable result for a bounded process           | Test reliability, rollback and ownership after launch                                   | Product or operations owner with engineering and control support |

The first row does not require every employee to be a prompt specialist. It asks whether the person can use an approved
tool for the work the job already contains. The second makes judgment visible. The third is closer to Workday's growing
hands-on category: a worker or team must alter the way work moves, not merely produce a faster draft. In a small firm
one person may cross all three rows. In a bank or hospital, distinct people may own each approval and failure path. Job
title alone will not reveal the difference.

The assessment should change when the job changes. A customer-support manager reviewing model-drafted replies could be
asked to identify a wrong refund promise and show how they would correct the customer record. A data analyst could
receive an AI-produced chart with a deliberately wrong denominator and trace it to a source. An engineer could describe
what happens when an automated handoff loses permission to read a file. These are illustrative exercises, not accounts
of actual interviews. The chosen defect should be relevant to the job and disclosed to candidates as an exercise. Score
the decision and explanation, not the candidate's access to a premium model at home.

There is also a limit to the matrix. A novice may recognize an error but lack the authority to stop a production system.
A senior engineer may build a reliable integration yet misunderstand the service rule it automates. Pairing an
assessment with supervised work can expose both gaps; a single timed prompt contest can hide them. A manager should
write down who may override the tool, who teaches the business rule and which record will show that the new skill was
used on paid work.

For an employer, the matrix is a drafting constraint. If a role only needs tool use, calling it an AI engineering job
can exclude qualified applicants and raise pay expectations without buying engineering work. If the role needs workflow
ownership, asking only whether a candidate knows ChatGPT can produce a false positive. For a worker, it helps turn
"learn AI" into a request for a project, reviewer and promotion criterion. The matrix does not set salary bands; our
[earlier pay-band analysis](/2026/06/19/ai-skill-premiums-pay-bands/) explains why a company must price changed scope
rather than a keyword.

## A training bill includes the reviewer

Workday's 79% versus 66% survey gap and PwC's 51% access figure are different questions in different samples. Neither
tells a finance team what a particular training program will cost. A course fee is the easiest line to see, but an
employee who practices on company work also draws on a supervisor's time, approved tools, source access and a safe way
to repair mistakes. The manager may need temporary coverage for the worker's ordinary tasks. Those costs remain even
when the online course is free.

An **illustrative budget worksheet**, not a reported company case, can make the choice concrete. Suppose a team wants
one analyst to advance from using an AI assistant to checking its output in a live reporting cycle. For eight weeks it
reserves two paid practice hours for that analyst and one review hour for a senior colleague each week. That is **16
analyst hours** and **eight reviewer hours**, before any tool fees or replacement coverage. Finance can multiply each by
the relevant fully loaded hourly labor cost, add approved software and data expenses, and compare the sum with the cost
of continuing to buy external review. No published source here supplies those rates, so a universal dollar estimate
would be invented.

The worksheet needs a result field. The analyst should finish with a reviewed work sample, a record of errors caught and
an agreed decision about when they may sign off without help. The manager should record how many review hours the task
actually consumed and whether the report improved on a measure the team already uses, such as correction rate or
turnaround time. If the role is supposed to become a workflow-building job, the next assignment should include a
documented handoff and failure test. Counting course completions alone would measure attendance, not capability or
return.

Funding depends on who benefits. A local manager may pay for a tool-use module that makes the team's current work
better. A central product group may pay for the engineering and security review behind a reusable workflow. An internal
mobility program may fund time that prepares a worker for a different team's vacancy. If no one owns the transfer, the
worker can acquire a skill on personal time while the company advertises it as a prerequisite. That is a
labor-allocation decision, even if it appears in the HR system as a learning recommendation.

The employer should also decide whether this is a changed job or an extra requirement added to an old title. If the
worker now approves outputs that affect customers or money, the company needs to revisit authority, supervision and
possibly pay. A broad AI-literacy clause need not change a band. Accountability for an automated process might. The
distinction cannot be settled by Workday's aggregate percentage or LinkedIn's profile association; it belongs to the
actual job description and the results the employee is allowed to own.

## The unanswered conversion question

Neither Workday's skill mentions nor LinkedIn's posting share reports how many people were hired for the newly worded
jobs, how well they performed, or whether incumbent workers moved into them. Indeed's survey reveals what respondents
say they value, not whether a proposed assessment predicts better outcomes. PwC captures worker experience, not a
controlled estimate of training return. These limits matter because recruitment software can make a skill label spread
faster than the work behind it.

A stronger employer readout would follow a cohort from requisition through paid assignment. Record the exact tasks
requested, the assessment offered, who had access to preparation, which applicants were hired, how the work was
reviewed, what training time was paid and whether a worker later changed role or pay. Separate job families and levels.
Then compare like with like across periods. Without that chain, a company can report more AI-literacy ads and course
completions while having no evidence that a worker gained a durable career path.

In the next requisition review, the useful document is the one with a task and an owner beside "AI skills required."
Someone should be able to point to the source the worker will check, the output they may approve and the process they
may change. Workday's 25% and LinkedIn's 65% cannot answer those questions for the manager.

## Continue reading

- [Workday's Mobility Agent Meets a Hiring Pause](https://digidai.github.io/2026/10/09/workday-mobility-agent-internal-career-freeze/): See why an internal skill match still depends on a funded opening and a manager's release decision.
- [AI Builder Ads Grow While Early Careers Move to Contracts](https://digidai.github.io/2026/10/13/ai-builder-ads-early-career-contract-paths/): Separate advertisements for builders from durable first jobs and paid assignments.
- [Workers Fear Skill Erosion as CHROs Ask for More Judgment](https://digidai.github.io/2026/09/23/ai-skill-erosion-critical-thinking-work-redesign/): Examine the judgment workers can lose when AI takes over routine practice.
- [AI Skill Premiums Put Pay Bands on Trial](https://digidai.github.io/2026/06/19/ai-skill-premiums-pay-bands/): Follow the pay-band decision after an employer defines which AI work changes a role.