Unlocking productivity with generative AI: evidence from experimental studies

Generative AI has seen an unprecedented boom in recent years, driven by significant advances in technology and computing as well as the release of user-friendly tools like ChatGPT. Their accessibility has enabled many individuals and firms to incorporate generative AI into their daily activities, sparking research interest in the technology’s impacts.

In this context, experimental studies have gained prominence as a way to better understand generative AI’s effects on economic outcomes. Experimental studies take place in controlled settings and allow researchers to isolate causal effects, providing insights into the potential, limitations, and implications of generative AI.

A recent OECD Artificial Intelligence paper that reviews experimental research on the impact of generative artificial intelligence (AI) has found that a number of these studies corroborate the potential of Generative AI to enhance productivity, innovation, and entrepreneurship. 

The studies highlight how generative AI can increase efficiency in tasks such as writing, summarising, editing or translating text and code. For instance, individuals who work in customer support, software development or consulting have seen average productivity gains ranging from 5% to over 25%. The technology can not only speed up these processes and free up time for more complex tasks but also broaden access to who can perform them.

Recent experiments show that less-experienced or lower-skilled individuals tend to see the largest productivity gains when using generative AI tools. By providing cost-effective and instant access to relevant information, generative AI facilitates on-the-job learning, provides support and feedback, and helps lower-skilled workers perform tasks they might not be able to complete on their own, bridge the gap with their more skilled peers.

This is particularly true in areas such as customer support or software development where generative AI excels at automating repetitive or time-consuming activities. By handling some of these well-defined tasks, generative AI can allow workers to focus on higher-value activities, resulting in significant boost to productivity. This democratisation of skills can level the playing field, by making some areas of expertise more accessible, enabling more people to participate in higher-skilled work.

The potential gains of generative AI are not limited to less experienced workers. In fact, more experienced ones can reap significant productivity gains as well. Their deeper knowledge, higher skill levels and greater contextual awareness can notably enable them to better interpret and apply AI-generated outputs. However, since they already possess a core set of skills, the technology needs to complement their existing expertise to generate substantial gains. Trust also matters; experienced users may adopt AI more cautiously, which might limit immediate benefits.

While generative AI can bring substantial productivity gains, understanding how and for which purpose it is used is critical for benefits to materialise. In this context, it is important to consider the fit between the task undertaken and AI’s capabilities; when generative AI is applied to tasks beyond its capabilities, it can harm performance by introducing errors or lowering output quality. This highlights the need to critically assess AI outputs and understand its limitations to ensure it is applied where most effective. 

Ultimately, the productivity impact of generative AI seems to depend on how well its capabilities align with the task undertaken, on users’ skills and their ability to evaluate outputs, and on the level of trust in the technology.

Learning and skill development are key for long-term productivity growth. Generative AI is increasingly influencing these dimensions, offering benefits for learners through personalised support, and expanding access to knowledge. These benefits are especially valuable for individuals with lower proficiency, potentially helping them progress faster and build confidence. In addition, the technology can act as a virtual subject-matter expert for educators, especially when traditional resources are limited. 

However, a generative AI approach to learning also carries risks. Relying too heavily on generative AI can undermine critical thinking and hinder long-term skill retention, especially when learners accept information without questioning it - an issue amplified by generative AI’s occasional production of plausible but inaccurate outputs.

If users fail to engage deeply with the technology’s outputs, they may develop a tendency to rely on generative AI rather than engage in independent problem-solving. For instance, recent experimental evidence suggests that individuals who use generative AI appear to perform better in their tasks, but they also show signs of reduced independent thinking compared to those who used other tools or got help from a person. Therefore, the way generative AI is integrated into learning and work environments is crucial. Guidance and training are key to ensuring generative AI truly enhances cognitive and professional skills.

Generative AI can also play a key role for innovation and entrepreneurship. By integrating generative AI into their operations, firms can speed up innovation, improve research quality, and use resources more efficiently, helping them gain stronger competitive advantages. For instance, generative AI can be employed to delegate tasks during early prototyping in industrial product design, resulting in lower costs and faster iterations, or to perform some well-defined R&D tasks under tight time constraints. 

Generative AI’s can also play a relevant role for idea generation. In this context, experimental evidence highlights that less-creative users can leverage AI-assisted idea generation to produce more novel, useful and high-quality ideas. However, AI-generated ideas may be more similar to one another, meaning that gains in individual creativity may come at the cost of reduced collective novelty. Highly creative users, by contrast, seem to experience more limited benefits from AI suggestions, as their performance remains consistently high regardless of access to AI-generated ideas. Nevertheless, they could leverage the technology as a tool to support and extend their ideation, drawing on their own domain knowledge to evaluate and refine AI-generated content. 

Concerning entrepreneurship, evidence suggests that entrepreneurs in high-performing firms see greater business gains from generative AI by effectively integrating it into their decision-making, while those in lower-performing firms benefit less - or may even face setbacks. These differences stem not from access to the technology, but from how entrepreneurs use it, underscoring the importance of effective human-AI collaboration.

The experimental evidence discussed in the recent OECD Artificial Intelligence paper suggests that the integration of generative AI into the workplace must be approached with care to harness its full potential. Gains hinge on meaningful human-AI collaboration where purpose, fit, user expertise and ability to evaluate outputs all play a critical role. 

For generative AI to be truly effective, workers must be trained not only on how to use generative AI tools and adapt workflows, but also to critically assess its capabilities. This means ensuring workers and managers understand how AI can reshape their tasks and organisations. At the same time, the impact of AI can vary significantly across industries, tasks, and individual roles.

Fully harnessing its long-term potential requires fostering a culture of continuous learning and adaptation, where AI becomes an integral part of the work environment that complements human skills. Critical thinking remains key to enable workers and organisations to understand when to use generative AI and how to use it in a meaningful way to leverage its full potential in the long run. 

To support this transition, governments and organisations can play a key role in strengthening skills, both technical and foundational ones, including those related to problem solving and critical thinking, as well as in boosting broader digital and innovation capabilities, as outlined in a recent OECD policy brief.  

While the debate about how to best ensure transparency, accountability, and trustworthy generative AI use continues, equally important are investments in digital literacy, training and workforce retraining programmes to ensure the benefits of AI are broadly shared. With the right policies, education, and safeguards in place, generative AI can be a powerful force for inclusive and sustainable growth.