A new study from Microsoft shows that generative AI is most used by people in knowledge-heavy work like writing, research, sales, and teaching. It is used far less by those in physical, care-based, or manual jobs. For thought leaders, this means that selling AI needs to ignore ‘job replacement’. Instead, it should focus on guiding organisations through augmentation, workflow redesign, and responsible transition. Now, as audiences turn to AI assistants for answers, our storytelling must adapt. We’re speaking to people—and to the algorithms that shape what they see.
Exploring the Microsoft WorkLab study
Microsoft’s recent study, Working with AI: Measuring the Applicability of Generative AI to Occupations, provides a comprehensive look at how generative AI is shaping work. The study drew on over 200,000 anonymised Bing Copilot conversations. It looked at the tasks where AI is being used, and the success of its assistance, measured by the ‘thumbs up’ given to the interaction.
This highlights a key issue in the use of AI: the difference between what AI delivers, and what the user intended. In 40% of interactions, the user did not achieve what they wanted from the AI. This suggests that AI is not necessarily completing tasks for users, and that—at present at least—its main role is augmentation rather than direct replacement.
So far, the study is relatively uncontroversial. The next step in the research, however, moves into the realms of speculation. The researchers tried to map these tasks onto occupations to identify the jobs in which generative AI is ‘most applicable’. The main problem with this is the definitions of both ‘applicable’ and the tasks associated with individual jobs.
It is not always easy to define jobs, or associate particular tasks with them. The top three jobs on the list—that is, the jobs for which Microsoft suggests AI is ‘most applicable’—are translators and interpreters, mathematicians, and historians. Does this mean that translators, mathematicians and historians are most likely to be using Copilot? It seems more likely that a lot of people who are not expert in those fields are using Copilot to fill the gaps in their knowledge. They need a quick and superficial answer to their questions—and that is probably what they are getting, hence the high rate of satisfaction.
Drawing out the lessons for thought leadership in the AI era
Criticisms aside, it is worth moving beyond the list of job titles, and instead looking at the tasks that people are using AI to support, and the categories of tasks that are clearly not AI-friendly.
AI is most applicable to knowledge work centred on information, writing, and communication. Copilot is being used to provide answers that might otherwise have required research, or where accuracy is important (for example, in mathematics). Its use is bypassing the need for an expert such as a translator, and instead allowing people to get on with their work quickly and easily. Effectively, it is automating the work of researching and writing across professions and organisations.
AI is far less used in roles requiring physical dexterity, human presence, or care. The revolution in robotics in factories came many years ago, and AI doesn’t really have much to add there. We might also speculate that people in these jobs simply don’t have time or resources for AI. As one user on Reddit said, hospitals in many low-income countries don’t even have computers, let alone AI.
This provides useful ideas about how organisations are likely to buy, adopt, and reconfigure around AI. Selling into companies undergoing these transformations will need technical knowledge, but it will also require a nuanced understanding about augmentation, culture, and business outcomes. The key messages should be:
- Emphasise that AI transformation is not about automating or replacing whole functions
Instead it is about reshaping the tasks that employees perform every day. It must be presented as a collaborator, not a competitor. People may also need to be trained and supported in its use.
- Focus conversations on using AI for specific tasks
The Microsoft study sets the tone here, once you look past the job roles. Instead of talking about ‘AI in [sector or role]’, thought leaders should identify specific tasks where AI will add value, such as drafting reports or summarising research. This will map its use to specific workflows, not roles.
- Recognise and acknowledge professional identity
On Reddit, professionals in fields where AI was ‘highly applicable’ bristled at the thought of it being used to do creative or expert tasks. Their professional identity was bound up in their credibility, judgement and creativity. Thought leaders must acknowledge that—and show that wise and ethical use of AI will not dent that.
Rethinking your narrative when AI joins the audience
As AI assistants become part of how people learn, research, and make decisions, the way we tell stories needs to shift too. It’s no longer just about reaching people—it’s about shaping the ideas that their AI copilots will surface, summarise, and amplify. The best thought leaders will be those who can speak to both: the human mind and the machine filter standing between ideas and influence.