News
Which Jobs Can AI Help With? A Business Owner’s Guide to the Research
A study of 200,000 AI conversations points to useful tasks, not inevitable job losses. Here is how to choose a practical, reviewable business pilot.
A useful starting point for business AI is smaller than a job title: one recurring task, a clear input and a result someone can check. Research on real AI conversations provides a reason to examine that work carefully, without treating a list of occupations as a redundancy plan.
AIS Dispatch reviewed the current Working with AI manuscript and its accompanying materials. The research belongs to its authors; the pilot framework below is AIS analysis for business owners, not a result from the study.
What the researchers actually measured
The study examines 200,000 anonymized U.S. Bing Copilot conversations from January through September 2024. It connects activities within those conversations to occupational tasks. Creating, processing and communicating information feature prominently in the findings.
The paper was first submitted in July 2025; this article uses version 7, dated October 23, 2026. Its applicability measures describe overlap between AI use and work activities. They are not probabilities of job replacement, a count of layoffs or evidence that an entire occupation can operate without people. One product’s conversation sample also does not represent every workplace.
AIS analysis: Choose a task you can inspect
Ask your team where they repeatedly turn existing information into a first draft. A useful candidate might be summarizing approved meeting notes, preparing a response from an established FAQ or organizing an inquiry before a person follows up. These are examples to test, not promises that a tool will handle them reliably.
Write the job in one sentence: “Given these approved inputs, prepare this output for this person to review.” If you cannot define the source material, the output or the reviewer, narrow the assignment. A simple form rule or reminder may solve the problem without generative AI.
Set the boundaries before the pilot
Decide what the assistant may read and what it may change. Drafting a reply is different from sending it; summarizing a request is different from promising a price or approving an exception. Keep consequential actions behind a named person’s approval until the workflow has earned a broader scope.
Use material you are permitted to share and check the tool’s data terms and access settings first. Give the assistant a stop rule for missing information, contradictory instructions and requests outside the approved task. An unanswered question should become a handoff, not an invented answer.
Run a small, honest comparison
Choose a manageable set of ordinary and difficult examples. Have the same qualified reviewer assess the existing process and the AI-assisted version against the same standard. Record useful outputs, missed details, correction time and cases sent back to a person.
A faster first draft is not a better process if reviewing it consumes the time saved. Keep the inputs, draft, corrections and final decision together so you can see whether the assistance is improving. Stop or narrow the pilot when it creates unreliable work or exposes information it should not use.
Make the next step operational
Bring one repeated task, its current owner and a few approved examples to a planning conversation. That gives a partner something concrete to evaluate: the handoff, the controls and the result the business needs.
The opportunity is to improve a defined piece of work while keeping responsibility visible. It is not necessary to predict which job disappears to find a useful place to start.
Sources and reporting
AIS Empire’s coverage draws on the original reporting and statements below.
- Working with AI: Measuring the Applicability of Generative AI to Occupations — version 7
Current manuscript dated October 23, 2026. The observations are from January–September 2024, not a 2026 employment survey. Applicability is not a forecast of job losses.
- Working with AI — publication record
Primary publication page and research summary. The initial publication is dated July 2025; the current manuscript is linked separately above.
- Working with AI — accompanying data and version notes
The project repository documents version 1.1 and the updated treatment of physical tasks. AIS does not reproduce an older viral occupation ranking as a current replacement forecast.
