On September 16, 2026, the economic researchteam at OpenAI published a study examining more than 1.5 million work-related messages sent to ChatGPT by users registered in the United States. The results suggest that some employees are regularly turning to tasks traditionally associated with other professions. This trend may broaden the scope of jobs even before their job titles change, but the study does not measure productivity gains, the quality of results, or the effects on employment and compensation.
A Study on the Actual Professional Uses of ChatGPT
The "Work at the Frontier" report analyzes more than 1.5 million messages classified as professional, sent between April and July 2026 from individual ChatGPT accounts. Participants also had a ChatGPT Business account, which allowed OpenAI to link their messages to a profession or service they had specified during registration.[1]
This methodological clarification is important. The messages do not come from companies’ ChatGPT Business spaces, and the study does not examine all of the work that was done. OpenAI states that it excluded users who refused to allow their data to be used for training, automatically removed personal information before classification, and prevented researchers from reading individual conversations.[1]
The lab then compares the requests to the activities listed in O*NET, the U.S. database of occupations and tasks. A request is considered “cross-functional” when it corresponds more closely to another occupational family than to the occupation reported by the user. The study thus seeks to determine whether AI enables only one-off experimentation or whether certain new activities become part of routine work.
Tasks outside the scope of one's job that become recurring
Among the approximately 6,200 people monitored continuously over the four-month period, the proportion of messages corresponding to cross-functional tasks that had already been used rose from 13.1% in April to 25.9% in July. The metric has nearly doubled, but its design inherently favors an upward trend: each new month increases the number of tasks that can be classified as recurring. OpenAI clarifies this in the report.[1]
A second comparison, however, reinforces the hypothesis that the tool has become established in users’ practices. When a user had used ChatGPT for a cross-functional task during a given month, they returned to that same category the following month in 23.6% of cases, compared with 8.4% among comparable users who had not used it previously. The difference is 15.2 percentage points.[1]
| Observed indicator | Published Results | What it helps us understand |
|---|---|---|
| Analyzed messages | More than 1.5 million | A large sample of professional requests, but limited to a selected group of U.S. users. |
| Cohort followed for four months | Approximately 6,200 users | Analysis of recurrence among individuals with at least ten messages sampled each month. |
| Percentage of cross-functional tasks already in use | 13.1% in April; 25.9% in July | Progress consistent with the integration of new tasks, without indicating a contractual change in the job role. |
| Back next month | 23.6% versus 8.4% | A task that has already been performed using AI reappears more often, but this correlation alone does not establish causation. |
| Average feedback after first use | 18.5% | About one in five user-task combinations is reused the following month. |
Not all expansions of responsibilities are created equal
Return rates vary significantly depending on the nature of the work. Client communications account for 54 percent, advertising or promotional copywriting for 44 percent, and the creation of marketing materials for 37 percent. In contrast, legal research is reused in only 10 percent of cases, while the explanation of financial information and the presentation of professional information account for 15 percent.[1]
This difference may reflect the natural frequency of tasks, but also the perceived level of risk. Producing a first draft of a marketing message is relatively easy to control. Interpreting a legal rule or explaining financial information requires expertise, validation, and more clearly defined accountability. The low repetition rate therefore does not mean that AI is incapable of helping, but rather that its integration depends on the context and the consequences of an error.
| Cross-functional task | Return Rate the Following Month | Read Carefully |
|---|---|---|
| Discuss goods or services with customers | 54 % | Frequently used and easily integrated, provided that the information shared is verified. |
| Write advertising or promotional content | 44 % | Generation speeds up the initial drafts, but does not guarantee compliance or brand consistency. |
| Create marketing materials | 37 % | AI streamlines production without replacing market knowledge and human validation. |
| Explaining Financial Information | 15 % | The risk of error and accuracy requirements may limit delegation. |
| Conduct a legal search | 10 % | The low recurrence rate may reflect the need for up-to-date sources and specialized expertise. |
Does AI really allow us to tap into expertise?
OpenAI notes that cross-domain prompts are slightly shorter, averaging about twenty characters fewer. They are less likely to request explanations, advice, or a detailed format, but they provide more examples and context and more frequently ask for verification. The authors propose the concept of “borrowed expertise”: the user presents a problem and contextual elements, then uses the model to apply knowledge from another domain.[1]
This interpretation should remain cautious. A well-formulated result does not prove that a skill has been acquired. The user may produce an output without understanding the method, assumptions, or limitations. In the short term, this reduces friction between departments. In the longer term, the organization must distinguish between assisted autonomy—which relies on the ability to exercise control—and uncontrolled delegation, which can undermine learning.
The International Labor Organization estimates that one in four workers worldwide is employed in a job that involves some degree of exposure to generative AI. However, it believes that a transformation of tasks is more likely than the complete disappearance of most jobs, particularly because human intervention remains necessary.[2] The OpenAI study provides behavioral insights consistent with this view, without directly measuring employment.
Executive Training
AI & Data Science
s for Managers
Integrate AI into your business strategy. A 360° approach—Technology, Business, and Ethics—designed for decision-makers. Prerequisites: 5 years of managerial experience.
Jobs that are broader in scope, but not necessarily less specialized
An expansion of responsibilities can change the division of labor. A project manager might prepare marketing materials, an analyst might draft an initial legal summary, or a developer might structure a client communication. The job title remains the same, while the mix of activities gradually changes.
This versatility does not mean that specialists will disappear. When issues become complex, regulated, or irreversible, their judgment remains of central importance. AI may shift the point at which they become involved: the specialist no longer performs every step, but instead defines the rules, verifies the assumptions, handles exceptions, and provides final validation.
For companies, the issue is therefore not just about access to ChatGPT. It concerns the design of work: which tasks can be expanded, which ones require authorization, where should human review be incorporated, and how should new responsibilities be recognized? A silent expansion of the scope can create value, but it can also lead to overload or an informal transfer of responsibility.
Skills to Develop Beyond Mastering Prompts
- State the problem and provide sufficient context, without disclosing any confidential information.
- Evaluate the quality of an answer, verify sources, and identify weak assumptions.
- Understand the limits of your own area of expertise and know when to consult a specialist.
- Document the use of AI, human corrections, and the final decision to ensure traceability.
- Collaborate across departments to ensure that the expansion of responsibilities does not lead to software-enabled individual silos.
In the European Union, Article 4 of the Artificial Intelligence Regulation has required, effective February 2, 2025, that providers and deployers take measures to ensure that individuals using AI on their behalf have a sufficient level of proficiency with it. The expected level depends on the knowledge, experience, context, and the individuals involved.[3] Training cannot, therefore, be limited to a few techniques for writing queries.
Privacy, Governance, and Social Dialogue
The study itself raises a question regarding data governance. OpenAI states that the researchers did not read the conversations, that personal information was automatically removed, and that the analysis focused on pseudonymized labels and identifiers. These safeguards reduce exposure, but they do not eliminate concerns regarding transparency, the legal basis, retention periods, and individuals’ rights when professional uses are analyzed.[1][4]
Within organizations, risks also arise when employees copy customer data, contracts, HR information, or trade secrets to an unauthorized service. A usage policy must specify the permitted tools, prohibited data categories, monitoring procedures, and responsibilities. In particular, the CNIL recommends defining the purpose of the processing, the roles of the parties involved, the legal basis, and, when necessary, an impact assessment.[4]
Finally, the redefinition of tasks must be discussed with employees and their representatives. It can enhance autonomy and reduce certain dependencies, but it can also raise expectations without changing working hours, recognition, or compensation. The ILO emphasizes social dialogue to ensure that productivity gains are accompanied by improved working conditions.[2]
What the study does not yet allow us to conclude
- The sample is not representative of the entire U.S. labor force, much less the global labor market.
- The study examines a sample of messages and may overlook repeated usages found in conversations that were not included in the sample.
- The automatic classification of a task based on O*NET does not fully capture the reality of a job, which depends on the industry and the organization.
- There are no direct measures linking the observed practices to quality, productivity, errors, wages, or employment.
- OpenAI is studying the use of its own product, with no independent validation of the results as a whole announced.
Toward professions that are restructured before being renamed
The main takeaway is not that ChatGPT replaces one profession with another. It is that the boundaries between roles may become more fluid. Certain simple, verifiable tasks are being taken on by employees who were not traditionally responsible for them, while high-stakes activities remain more closely associated with specialized expertise.
To assess the true scope of this shift, we will need to monitor the quality of output, incidents, workload, career mobility, and the recognition of skills. If AI expands the range of job roles, organizations will also need to expand training, governance, and accountability mechanisms. The question, therefore, is not merely what employees can now accomplish with a model, but under what conditions these new tasks lead to sustainable professional growth.
Learn more
To learn more about the transformation of professions, the professional applications of AI, and the skills needed to adopt them, be sure to check out these analyses from the aivancity blog.
Sources
[1] OpenAI Economic Research. (2026). Work at the Frontier: How Workers Are Unlocking New Ways of Working. View the report
[2] International Labor Organization (ILO). (2025). Generative AI and Jobs: A 2025 Update. View the report
[3] European Commission. (2026). AI Literacy: Questions and Answers. View source
[4] CNIL. (2024). Recommendations on the Application of the GDPR to the Development of Artificial Intelligence Systems. View the recommendations
[5] OpenAI. (September 16, 2026). How workers are unlocking new ways of working. View source