AI-Exposed Jobs Are Changing. Here's What Employers Are Adding.
The federal government now rates every occupation for AI exposure. We read the job postings to see what that means in practice. In the jobs rated most exposed, employers aren't removing duties. They're adding new ones, and the new duties tell students what to build and what to show.
Dave McNasby · data pulled Oct 23, 2026
Cite this analysis
McNasby, D. (2026). AI-Exposed Jobs Are Changing. Here's What Employers Are Adding. semanticinsight.tech/ai-exposure-change.html (data as of Oct 23, 2026).
7,148
job-posting analyses, read duty by duty
316
occupations compared with BLS AI exposure
760
new duties confirmed across employers and metros
74%
of Very high exposure jobs show new duties, vs 41% of Low
The question
"Highly exposed to AI" is a warning. It isn't a plan.
In August 2026 the Bureau of Labor Statistics began publishing an AI exposure category, Low to Very high, for every occupation it projects. For college students the label is nearly universal: in our sample, 94% of bachelor's-level occupations are rated High or Very high. Marketing, accounting, PR, HR, software, management. Almost every degree path a student might choose carries the same warning.
The label answers "could AI touch this work?" It doesn't answer the questions a student actually has. Is the job changing yet? What's being added? What should I learn, and what should I put in front of an employer?
Job postings can answer those questions, because employers describe the work they're hiring for. But a posting is written in an employer's own words, mixed in with benefits, boilerplate, and house style. Finding what's new means comparing every duty in every posting against the federal description of the job, then confirming it isn't one employer's quirk. That's what our pipeline does.
How we know
Reading postings against the federal standard, sentence by sentence
O*NET, the Department of Labor's occupational database, lists the tasks that define each job. It's the federal standard that career tools, college programs, and the BLS exposure ratings all build on. Our pipeline measures how far real postings drift from it.
119K+
Collect postingsFrom 10,000+ employers across 138 U.S. metros.
Pull out every dutyEach posting is broken into individual duty statements, separated from benefits and boilerplate.
Match to O*NETEach duty is compared by meaning, not keywords, against every federal task for that occupation.
Group what's leftDuties with no federal match are clustered with similar duties from other postings.
760
ConfirmA cluster counts only if independent employers in at least two metros ask for it, and its sentences describe one clear duty.
Most unmatched sentences don't survive. A duty one company asks for is that company's job, not the occupation's. Requiring independent employers in more than one metro, plus a check that every grouped sentence describes the same duty, is what makes a short list trustworthy. Across 7,148 analyses of 316 occupations, 760 new duties passed.
Finding 1
The jobs rated most exposed are the ones employers are rewriting
The BLS rating holds up. The higher an occupation's exposure, the more likely its postings include duties the federal description doesn't cover.
Share of occupations with at least one confirmed new duty, by BLS AI exposure
All 316 occupations. A confirmed new duty appears at independent employers in two or more metros and has no match in O*NET.
Very high110 occupations
74%
High88 occupations
57%
Moderate69 occupations
45%
Low49 occupations
41%
The amount of change rises too. The typical Very high occupation has about one new duty for every ten federal tasks; the typical Low or Moderate occupation has none (Spearman ρ = 0.31, p < 0.001).
The change is uneven, though. Very high includes software developers, with 22 new duties, and customer service representatives, with 2. Anthropic's Economic Index, which measures how much of each occupation's work people actually bring to its Claude AI model, helps explain the gap. Where real AI use is higher, employers have added more.
New duties per 10 federal tasks, by observed AI use
Median. Observed exposure from the Anthropic Economic Index (0 to 1). 17 occupations without a match are not shown.
High (0.35+)30 occupations
1.9
Mid (0.15–0.35)60 occupations
0.9
Low (under 0.15)93 occupations
0.5
None116 occupations
0.0
Among the 61 bachelor's-level occupations BLS rates Very high, those with high observed AI use show about three times the new duties of those with little use (median 2.4 vs 0.8 per 10 tasks). The pattern holds outside computer and math jobs.
Finding 2
Jobs are adding work, not shedding it
If AI were simply replacing tasks, the tasks people use AI for should be vanishing from postings first. In most jobs they aren't. We matched Anthropic's task-level data to 99.1% of O*NET Core tasks: tasks people do with AI disappear from postings at about the same rate as other tasks. Employers still list the drafting and analysis. What changes is what sits on top of it.
Federal tasks where AI is in use
Research and compile information
Draft reports, releases, and content
Analyze data and forecast
Screen, sort, and summarize
→
What employers are adding
Judgment: evaluate results and decide what changes
Systems: own the data, tools, and definitions
Proof: measure outcomes and report them
Coordination: work across sales, legal, and IT
Only about 4% of the new duties mention AI by name. Employers aren't mainly asking for "AI skills." They're asking for the work AI makes more important: deciding, checking, owning, and showing results.
What it looks like
Four careers, up close
Four bachelor's-level careers that BLS rates Very high, where people already use AI heavily, and where our postings confirm real change. For each: the federal tasks people already do with AI, the duties employers are adding, and what a student can build to show them.
Marketing specialists and managers
BLS Very highAI use 0.65 / 0.3223 new duties66 analyses
Already done with AI
Research market conditions and competitors
Gather and analyze customer data
Forecast sales and trends
Measure campaign effectiveness
Prepare reports
Employers are adding
Own datasets and set data definitions and reporting standards22 employers · 8 metros
Target priority accounts with the sales team19 employers · 5 metros
Create content for B2B buying committees17 employers · 7 metros
Run campaign reporting and post-campaign evaluation16 employers · 6 metros
Use AI and automation to scale programs12 employers · 5 metros
What to show
A campaign dashboard with a one-page recommendation
A metric dictionary that defines each KPI
A target-account brief written for a sales team
AI-drafted content with your edits marked, and why
The parts of the job AI doesn't touch in Anthropic's data are the people parts: hiring, negotiating, managing budgets. Market research analysts have the highest observed AI use of any business occupation in our sample.
Public relations specialists
BLS Very highAI use 0.4513 new duties62 analyses
Already done with AI
Respond to media requests
Research public opinion
Plan communication strategy
Write press releases
Employers are adding
Evaluate media coverage in PR software20 employers · 7 metros
Run email campaigns and webinars17 employers · 5 metros
Produce video and multimedia13 employers · 6 metros
Track and report engagement metrics7 employers · 3 metros
What to show
A coverage report from a media-monitoring tool, with what you'd change next
A short video or multimedia piece you produced
An email or webinar campaign with its results
Writing the release is the part AI already helps with. Employers are asking PR staff to prove what the coverage did.
Accountants and auditors
BLS Very highAI use 0.358 new duties88 analyses
Already done with AI
Examine financial records
Analyze for fraud and control weaknesses
Prepare financial reports
Employers are adding
Build and run a risk-based audit plan64 employers · 24 metros
Research technical accounting questions11 employers · 6 metros
Write technical accounting memos6 employers · 2 metros
Follow up on remediation of findings
What to show
A risk-based audit plan for a sample company, with your reasoning for what to test
A technical accounting memo on one real standard
A remediation tracker that follows findings to close
The risk-based audit plan is one of the most widely confirmed new duties in our data: 64 employers in 24 metros. Reviewing records is becoming the input. Deciding where the risk is, and defending it in writing, is the job.
Human resources specialists
BLS Very highAI use 0.409 new duties27 analyses
Already done with AI
Compile employment data and reports
Search for candidates
Interview applicants
Screen applications
Contact applicants about status
Employers are adding
Process payroll and wage garnishments13 employers · 3 metros
Advise on automated HR systems11 employers · 4 metros
Run desk audits and job classification11 employers · 4 metros
Manage workers' compensation claims10 employers · 4 metros
Oversee the HR information system
What to show
A review of an automated screening process: what to check, and where a person decides
A job classification write-up from a desk audit
A walk-through of a payroll or workers' comp case
HR is the exception to Finding 2. Here the recruiting tasks people do with AI (interviewing, screening, status updates) are also going missing from postings. The added duties move toward systems, compliance, and casework.
For career preparation
How to prepare, and how to present yourself
Across all four careers the pattern is the same. The tasks AI handles are becoming the baseline, and employers are writing new duties around judgment, systems, and measured results. That points to three rules for students, and for the career staff who coach them.
1
Treat AI-assisted work as the floor, not the pitch
Research, drafting, and first-pass analysis are what AI already does in these jobs. Do them well and fast, with AI, but don't build your story on them. Every other applicant can say the same thing.
2
Build proof of the duties employers are adding
Pick two or three new duties from your target occupation and make something that shows each one: a plan, a dashboard, a memo, a process review. Choose duties confirmed by many employers in many metros.
3
Lead with the decision and the result, then name the tool
Employers rarely ask for "AI skills" directly. Show what you decided, what you checked, and what changed. Mention the tool as part of how you got there.
Instead ofProficient in ChatGPT and AI tools for marketing content.
TryBuilt a campaign dashboard and recommended shifting budget to the best-performing channel; used AI to draft ad variants and edited them for brand voice.
The same logic applies to occupations that aren't changing yet. If postings for your target job still match the federal description, the traditional preparation still fits. Check where your occupation lands in the table below.
Look up an occupation
All 316 occupations
Each occupation is placed in one of four groups using two questions. Does BLS rate it High or Very high? And are employers changing it, with at least three confirmed new duties and at least two for every ten federal tasks?
53
Exposed and changingPrepare for the new duties. 32 of the 53 also show real AI use.
145
Exposed, not changing yetPostings still match the federal description. Watch for change.
8
Changing for other reasonsLow exposure, real new duties from equipment, rules, or safety practice.
110
StableLow exposure and little change, mostly hands-on and service jobs.
Search any occupation, or click a column heading to sort.
Occupation
BLS exposure
AI use
New duties
Per 10 tasks
Group
AI use is Anthropic's observed exposure (0 to 1); a dash means no match in that data. New duties are confirmed across independent employers in at least two metros. Small counts are approximate, so use them to compare groups rather than to rank single occupations.
Who this helps
From a label to a plan
Students and job seekers can prepare for the specific duties employers are adding, instead of reacting to a category label.
Career staff and faculty can see which highly rated occupations are changing now, and build projects and resume coaching around the new duties.
Workforce and Workforce Pell programs need evidence that an occupation is durable. A Very high rating with real AI use and new duties is a different signal than a Very high rating with neither. See the Durability Screen for the state-level view.
Putting it to work. semanticinsight.tech is operated by OptimalResume.com, Inc., the company behind SkillsFirst, a career readiness platform used by colleges and workforce programs. We are using this data to build occupation-specific tools in SkillsFirst: portfolio projects built from the duties employers are adding, reviewed by each school's career staff, and resume guidance that helps students show those duties when they have done the work.
Method.
New duties. Duty statements from job postings that match no O*NET 30.2 task for the occupation, grouped into clusters. A cluster counts only if it is active, appears at independent employers in at least two metros, and its grouped sentences passed a coherence check. This is stricter than the count on the Occupational Drift Landscape, which also shows emerging duties still being confirmed. Employer and metro counts on the career cards are for each duty cluster.
Scale. New duties are divided by the occupation's O*NET Core task count and shown per 10 tasks. Our sampling saturates once an occupation's duties are covered, so dividing by the number of analyses would understate heavily studied occupations.
Sample. The 200 largest U.S. occupations by employment, which together cover over 80% of the workforce, swept across the 25 largest metros, plus occupations linked to college programs through the NCES CIP-to-SOC crosswalk. We kept the 316 with at least five completed analyses and an O*NET Core task list. The relationship between observed AI use and new duties is about the same in both halves of the sample. Postings were collected before August 2026.
BLS exposure. AI exposure categories published with the 2025–35 employment projections, joined by SOC code using BLS's own aggregate crosswalk.
Observed use. Anthropic Economic Index, labor market impacts release. Occupation-level observed exposure, and task-level penetration matched to O*NET Core tasks by exact statement text (99.1% matched). "Already done with AI" lists Core tasks with measured use in that data.
Missing tasks. A Core task counts as missing when it is absent from at least half of the analyses for that occupation. Across bachelor's-level occupations, 21% of AI-used tasks and 18% of other tasks were missing, a difference within noise (p = 0.23).
Four groups. Exposed means BLS rates the occupation High or Very high. Changing means at least three confirmed new duties and at least two for every ten Core tasks.
Limits
These are associations across occupations. They do not show that AI use causes new duties.
Observed exposure reflects Claude usage only, one AI product. A zero means no measurable use in that data, not zero use anywhere.
Each occupation has a handful of confirmed new duties, so single-occupation counts are approximate. Duty wording on the career cards is summarized from the cluster statements.
Core task counts vary because O*NET detail codes are combined to the six-digit SOC level, which can raise or lower an occupation's rate.
Postings measure what employers write down. Duties can change before job descriptions catch up.
Sources
U.S. Bureau of Labor Statistics, Employment Projections 2025–35 and AI exposure categories. bls.gov/emp ↗