Deep Research is the name given to the new information-gathering capabilities that ChatGPT and Gemini make available to professionals thanks to the emergence of reasoning models.
One of the most resource-intensive activities for an export manager, particularly in terms of time, is gathering information, conducting analyses, comparing competitors’ positioning and offerings, and preparing documents for the benefit of both the company and its local distributors. This is why the introduction of Deep Research capabilities in the leading Generative AI chatbots represents one of the most interesting developments for professionals and companies looking to expand into international markets.
Gemini Deep Research
Gemini, Google’s language model, has recently released its 2.0 version in Italian. In addition to improving the quality and naturalness of its responses, the new version also makes its “Deep Research” feature available within the chatbot.
For example, if users ask it, through a simple prompt, to conduct a competitor analysis and explicitly identify a number of competitors, the resulting document can be extremely detailed. After processing information from hundreds of websites, all of which are referenced with relevant links for further investigation, the report provides an in-depth analysis of each competitor, describing its business model, product or service offering, commercial strategies and feedback from online users.
At the end of the document, which can subsequently be downloaded, a summary table highlights the main findings, making it easier to understand and compare the different market players.
ChatGPT Deep Research
Although available to a limited extent in the Plus version and more extensively through a monthly subscription priced at EUR 200, ChatGPT’s Deep Research feature stands out for the structure and depth of the documents it can produce.
When asked, for example, about a regulation currently in force in a foreign country, it can explain the reasons behind the regulation and its effects, examine the guidelines required for compliance, suggest relevant examples and references, and highlight potential penalties and risks, while consistently citing the sources from which its information and data have been drawn.
Reasoning Models
The development of Deep Research capabilities was accelerated by an event that, last January, prompted OpenAI and Google to make these technologies available to the public sooner than expected.
The catalyst was the explosive success of a Chinese app, DeepSeek, although its momentum initially appeared likely to be short-lived. The quality of its results, the relatively low costs required for training, and its efficiency in managing computing resources attracted considerable attention. However, concerns over its handling of personal data also led the Italian Data Protection Authority to block its download in Italy.
Nevertheless, the DeepSeek phenomenon had a genuine “Sputnik effect” on the world of Generative AI, pushing major technology companies to release more advanced language models — known as reasoning models — which form the basis of Deep Research capabilities.
These models, such as DeepSeek’s R1, are described as “reasoning” models because of their ability to solve problems by breaking them down into successive stages, autonomously gathering data and information from the web while clearly identifying sources, and subsequently providing detailed responses based on the information collected.
Following a similar approach, OpenAI quickly responded to DeepSeek’s success by releasing o3 to the public, after it had previously been available primarily to developers. o3 is a “small” reasoning model, meaning that it is designed to use fewer computational resources when processing information, thereby improving efficiency and response speed compared with the previous o1 model.
What Opportunities Are There for Export Managers?
In addition to preparing market analyses using Deep Research capabilities, reasoning models can support export managers in commercial decision-making through, for example, the advanced development of SWOT analyses.
While previous language models, such as GPT-4o, are more than capable of drafting the contents of an email or a presentation, models such as o1, o3 and Gemini 2.0 are designed to address complex problems in greater depth. They can not only gather information to support their answers, but also provide a structured explanation of the factors and evidence underpinning their conclusions.
Their capabilities can be tested using logic problems, such as the well-known puzzle involving a ferryman who must cross a river with a goat, a cabbage and a wolf without allowing the goat to eat the cabbage or the wolf to attack the goat. What is particularly interesting is not only the clarity of the solution, but also the structured way in which the problem is addressed.
Reasoning models can deliver particularly impressive results when tackling logical, scientific and mathematical problems. They can also process uploaded files, making them suitable for professional applications. In fact, it is the combination of their analytical capabilities with a company’s own internal documentation that can produce more reliable, up-to-date and genuinely useful results.
The Challenge of Deep Research
Even more clearly, the depth and quality of the content produced by reasoning models and their Deep Research capabilities raise an important question: how can technological support be combined with the need to personalise its use through professional expertise and personal experience?
The real challenge for professionals, and especially for export managers, will be not simply to accept the analysis produced by AI, but to verify, adapt and refine it.
Another particularly significant aspect of reasoning-based models is their emphasis on making the steps behind their outputs more understandable and transparent. With earlier language models, users often had to explicitly request an explanation of the method used, through techniques commonly associated with “chain of thought” prompting. This new generation of Generative AI is designed to operate with a greater degree of autonomy, making it even more important to provide users with clear explanations, supporting evidence and verifiable sources.
This is essential to preserve the principle of “human in the loop”: human oversight remains crucial to ensuring that these emerging technologies are used safely, critically and responsibly.