Which Jobs Are Less Exposed to AI? A Task-by-Task Career Check

October 30, 2026 · By Lumen

Exposure is not a prediction of job loss

In its October 22, 2026 analysis, Pew Research Center measured how much AI can perform or assist with occupational tasks. It found that employment in highly exposed occupations had grown overall, while outcomes differed by gender. The study does not establish that AI caused those employment changes.

The ILO’s August 2026 skills report describes changing combinations of cognitive, socioemotional and digital skills as workplaces adopt AI. A useful career plan considers how to work with the technology as well as which tasks are harder to automate.

“Less exposed to generative AI” is also narrower than “secure.” A role can face weak demand, outsourcing, robotics, budget cuts or poor working conditions even when a chatbot cannot do its core work.

Start with a real working week

Pick two roles you could realistically pursue. Find several current employer descriptions for each, and ask people doing the work what consumes their time. Break each role into five to eight recurring tasks. “Customer success” is a title; handling a difficult renewal, preparing a report and fixing an account issue are different tasks.

For each task, write the output, the tools used and who checks it. Estimate whether it is a small, medium or large part of the week. Treat those estimates as questions to validate—not precise exposure percentages.

Use these six questions for every task

1. Is it mainly digital and repeatable? Standard summaries, routine drafts and predictable information processing are plausible automation or assistance targets. Ask what happens when the input is incomplete or wrong.
2. Does it require physical work in an unpredictable setting? Working around an unfamiliar building, equipment or a person’s changing needs is different from generating text. Physical work can still be affected by other forms of automation.
3. Who carries the accountability? Identify who must approve the result, respond to an error or satisfy a professional requirement. A tool that produces a draft does not automatically take over responsibility.
4. Does trust change the outcome? Negotiation, care, conflict resolution and long-running customer relationships involve more than polite wording. Ask whether the employer values that work or mainly measures transaction volume.
5. How easy is it to verify the output? A routine answer with a simple check is different from a consequential judgment that needs context and investigation. Verification itself may become a larger part of the job.
6. Is there a viable local path? Check pay, training time, licensing, travel, physical requirements and actual openings. Lower exposure does not help if the path is inaccessible or the job does not fit your life.

Worked comparison: two sales roles

Illustrative example—not an occupational ranking: Role A spends much of the week researching public contacts, writing first-touch messages and updating records. Role B combines customer-site visits, diagnosing an operational problem, negotiating with several stakeholders and owning implementation follow-through.

Both roles may use AI to prepare research and correspondence. Role A has more immediately digitizable tasks in this example. Role B has a larger physical and relationship component, but it may require travel, technical learning or an established network. The better choice still depends on the real employer and your circumstances.

Do not turn that example into “all field sales is safe” or “all SDR work disappears.” Teams distribute responsibilities differently, and employers may use productivity gains to change expectations rather than remove a whole role.

A worksheet you can copy

• Role and employer: Which actual opening am I evaluating?
• Core tasks: What are the five most important activities?
• AI assistance: What could a tool draft, classify or summarize?
• Human work: What requires presence, trust, verification or accountable judgment?
• Evidence: What did a current employer or worker confirm, and what am I assuming?
• Transition cost: What training, credentials, time and money would I need?
• Next experiment: What small project, conversation or application would test my fit?

Questions to ask in an interview

Ask which tasks have changed in the last year, what tools the team actually uses, who reviews AI-assisted output and how success is measured. Follow with: “What would you still need this person to own if the routine parts became much faster?” A concrete answer is more useful than an employer simply describing itself as AI-first.

Check the occupation’s BLS Occupational Outlook Handbook entry for a separate view of training, pay and projected demand. A projection is not a guarantee, and national data does not replace your local search.

Choose a testable next step

Keep transferable strengths, then add one skill that complements the work you want: quality checking, customer discovery, data interpretation or safe use of the tools common to that team. Try a small, non-confidential project before paying for a major career change.

Once you have a shortlist, check the employer’s hiring evidence. For remote roles, read our location and legitimacy checks. A promising career category and a genuine vacancy are separate questions.