AI-powered knowledge work: Deep research with ChatGPT from OpenAI: What are the advantages and limitations?

In-depth research: OpenAI opens access and changes the landscape of knowledge work

OpenAI has taken a remarkable step with the gradual opening of its Deep Research feature, a move that has the potential to fundamentally change how we acquire and process knowledge. What was once reserved for an exclusive group of Pro users is now available to a wider audience, including subscribers to ChatGPT Plus, Team, Education, and Enterprise plans. This expansion of access, albeit with monthly usage limits, signals not only the increasing maturity of this technology but also OpenAI's strategic ambition to take a leading role in the highly competitive field of AI-powered information systems. This move comes at a time of intensified competition from companies like Perplexity, Google, xAI, and Microsoft, all striving to develop the next generation of knowledge work tools.

Background and Functioning of Deep Research

Genesis and core functionality

Deep Research arose from the need to overcome the limitations of conventional search methods and usher in a new era of knowledge acquisition. It was conceived as a kind of "AI agent" capable of autonomously conducting complex, multi-stage research. At its core, it's about not only finding information, but also understanding, analyzing, and presenting it in a structured format. Deep Research utilizes a highly advanced version of OpenAI's o3 model, specifically optimized for the demanding tasks of web browsing and data analysis.

Unlike traditional chatbot modes, such as those used in GPT-4o, Deep Research is designed to operate over extended periods – typically between five and thirty minutes per query. During this time, it systematically scours hundreds of online sources, extracts relevant information, interprets its meaning within the context of the question posed, and synthesizes the results into a coherent report. This process goes far beyond simply retrieving search results; it involves actively engaging with the material, identifying patterns, inconsistencies, and relevant connections.

Technological Foundations

Deep Research's capabilities are based on a combination of various advanced AI technologies. A key aspect is "reasoning," the ability to draw logical conclusions and understand complex issues. This enables the system to independently develop and adapt search strategies, critically evaluate sources, and assess the relevance of information within the context of the specific question being asked.

Furthermore, Deep Research is capable of executing Python code, opening the door to direct data analysis. This capability is particularly valuable when it comes to processing large datasets, conducting statistical analyses, or performing complex calculations. Another important feature is the ability to process user-defined files. Users can provide the system with documents, spreadsheets, or other file formats, which can then be incorporated into the research. This makes it possible, for example, to integrate internal reports, research data, or specific documentation into the analysis, thereby broadening the research context.

A crucial difference from previous models lies in the training approach. Deep Research was trained using reinforcement learning, focusing on real-world tasks requiring browser and tool usage. This approach differs fundamentally from the purely text-based training method common in many earlier language models. By training on real-world research tasks, Deep Research learned to effectively navigate the dynamic and often unstructured information space of the internet.

Extended access and terms of use

New user groups and query limits

Expanding access to Deep Research to broader user groups marks a significant step in the democratization of this technology. Originally available exclusively to Pro users with a monthly subscription of $200, access was extended on February 25, 2025, to the following user groups:

Plus users (US$20/month)

10 deep research queries per month. This allows a wide range of users to experience the basic benefits of in-depth research without having to bear the high costs of a Pro subscription.

Team/Enterprise/Education

10 queries per user per month. This policy aims to enable access for organizations and educational institutions and to promote the collaborative use of deep research in teams.

Pro users

The monthly query limit has been increased from 100 to 120 queries. This represents a welcome increase in capacity for power users who regularly conduct extensive research.

Resource-intensive processing: The balance between precision and efficiency

These tiered usage limits reflect the resource intensity of Deep Research. Each query involves significant computational effort, as the model operates autonomously for up to 30 minutes, developing search strategies, evaluating sources, and triangulating results. Limiting the number of queries therefore serves to manage system resources efficiently and ensure consistently high service quality for all users.

Technical improvements as part of the expansion

In parallel with the expansion of the user base, technical improvements were also implemented, further increasing the functionality and user-friendliness of Deep Research:

1. Embedded images with quotes

Visual content from web sources is now directly integrated into reports and accompanied by appropriate source information. This enriches the reports with visual information and facilitates the understanding of complex topics, particularly in fields such as science, technology, and design.

2. Improved document analysis

Deep Research now has an even better understanding of uploaded files, especially PDFs and spreadsheets. This is particularly beneficial in specialized contexts where users frequently work with complex documents. The improved analytical capabilities allow for more precise extraction of information from these documents and its integration into research results.

3. Increased transparency

Every report produced by Deep Research includes detailed source citations and a summary of the research steps taken. This increases the traceability of the research process and allows users to better assess the credibility of the results. Transparency is a crucial aspect of building trust in AI-powered knowledge work and promoting the responsible use of this technology.

Performance and practical applications

Benchmark results and performance comparisons

Deep Research's performance has been proven in various internal and external tests. In direct comparisons with other models, including GPT-4o and Claude 3.5, Deep Research significantly outperformed them in various benchmarks:

Humanity's Last Exam (CAIS/Scale AI)

In this demanding benchmark, which tests the general knowledge and problem-solving capabilities of AI systems, Deep Research achieved an accuracy of 26.6%. By comparison, GPT-4o and Claude 3.5 only achieved 9%. This result underscores Deep Research's superior ability to understand complex questions and deliver precise answers.

GAIA benchmark

In the GAIA benchmark, which tests the ability of AI systems to answer questions in various fields of knowledge, Deep Research took the lead in 43 out of 50 task categories. This demonstrates the broad applicability and high performance of Deep Research across different domains.

Reprogramming research

In a specific use case in biomedical research, Deep Research was successfully used to analyze over 200 cell reprogramming studies in less than 30 minutes. This task, which would traditionally have taken days or even weeks, was accomplished in a very short time using Deep Research. This demonstrates the enormous potential of the technology to accelerate research processes.

Competitive landscape and strategic positioning

Competing solutions and unique selling points

OpenAI deliberately positions Deep Research as an answer to the growing competition in the field of AI-powered knowledge work. Several alternative solutions exist on the market that offer similar functionalities but differ in certain aspects:

Google Deep Research

Integrated into Gemini Advanced (also available for $20/month). Google offers a comparable solution with Gemini Advanced, which also relies on deep research functionalities. The competition between OpenAI and Google drives innovation in this area and leads to a continuous improvement in the available technologies.

xAI DeepSearch

Exclusively for Grok users (starting at $8/month). xAI, Elon Musk's company, offers another alternative with DeepSearch, but this is tied to a Grok subscription. This demonstrates that different players in the AI ​​market are pursuing different strategies to position and market their technologies.

Microsoft Think Deeper

Available for free, but without web browsing functionality. Microsoft offers a free solution called Think Deeper, but its functionality is limited because it cannot access the internet. This highlights that web browsing capability is a crucial differentiator for deep research tools.

A key difference between the various solutions lies in their "agent capability." While Microsoft's ThinkDeeper is limited to static datasets, the systems from OpenAI and Google are able to independently search the web and dynamically access new information. This ability to autonomously gather and process information is a central advantage of deep research and distinguishes it from simpler search tools.

Perplexity Deep Research

Perplexity Deep Research presents itself as a free, AI-powered research platform that provides users with fast and interactive access to extensive, up-to-date information sources. Unlike conventional search tools, Perplexity places particular emphasis on the transparent presentation of source information and the ability to answer complex questions in context. Through the use of advanced algorithms, the platform dynamically extracts relevant data from the web, meeting the user's information needs in real time. This combination of autonomous web research and precise results presentation makes Perplexity Deep Research an attractive tool – especially for users who value not only speed but also well-founded and comprehensible information. Furthermore, the platform's interactive nature allows for the direct clarification of follow-up questions through dialogue, thus supporting an iterative research process.

Economic implications and market strategy

OpenAI's pricing strategy, with a Plus subscription for $20 and a Pro subscription for $200, is a strategic move to appeal to a broad user base while retaining high-performance users. The more affordable Plus option allows a wider audience to learn about and utilize the benefits of deep research, while the Pro subscription is tailored to professional users who conduct extensive research and require advanced functionalities.

Analysts like Paul Schell of ABI Research see this development as a clear trend toward the "democratization of agent-based AI." The wider availability of deep research and similar technologies has the potential to fundamentally transform knowledge work and open up new opportunities for companies and individuals. At the same time, this development also holds disruptive effects for traditional knowledge workers, whose tasks could increasingly be taken over by AI systems. The ability to collaborate effectively with AI-supported tools and critically evaluate their results will be a key competency for knowledge workers in the future.

Security and risk management

Hallucination rates and error susceptibility

Despite the impressive capabilities of deep research, it is important to consider the limitations and potential risks of this technology. OpenAI itself acknowledges that deep research can draw incorrect conclusions or fail to properly evaluate authority sources in 3–5% of cases. These “hallucinations” or errors can have various causes, such as deficiencies in the training dataset, algorithmic weaknesses, or the inherent complexity of the information being processed.

An internal whitepaper from OpenAI specifically warns of the following potential sources of error:

Misinterpretation of regulatory guidelines

Deep research may struggle to correctly interpret and apply complex laws, regulations, or compliance guidelines. This can be particularly problematic in highly regulated industries such as finance or healthcare.

Insufficient distinction between facts and rumors

In the dynamic information space of the internet, it is often difficult to distinguish between established facts and unconfirmed rumors or opinions. Deep Research may, in some cases, struggle to reliably make this distinction and potentially include false or misleading information in its reports.

Limitations of uncertainty communication

AI systems often struggle to explicitly communicate uncertainties and probabilities in their statements. Deep Research might, in some cases, give the impression that its results are absolutely certain and error-free, even though this is not always the case in reality.

Safety measures and quality assurance

To minimize risks and ensure the safety of deep research, OpenAI has taken various measures:

1. Red-teaming campaigns

External security experts and "red teams" were tasked with systematically searching for vulnerabilities and potential for misuse in Deep Research. These tests covered 12 different risk categories, including data privacy, dissemination of dangerous advice, discrimination, and manipulation. The results of these campaigns helped OpenAI identify vulnerabilities and improve its security measures.

2. Automated evaluations

OpenAI relies on automated evaluation systems to continuously monitor the quality and safety of deep research. According to the company, these systems achieve an accuracy of 93% in detecting unwanted content, such as hate speech, propaganda, or harmful information.

3. Sandboxing

Python code execution within Deep Research takes place in isolated "sandbox" environments. This prevents potentially malicious code from gaining access to the overall system or causing unwanted side effects. Sandboxing is a common security technique used to minimize the risk of malware or system compromise.

Future developments and open questions

Planned features and enhancements

OpenAI has already announced that Deep Research will be further developed and expanded with new features in the coming months. The following enhancements are planned for the second quarter of 2025:

Multimodal reports

The integration of data visualizations and generated images into Deep Research reports. This is intended to further increase the comprehensibility and informative value of the reports and enable users to grasp complex information at a glance.

API access

The provision of an application programming interface (API) for selected enterprise partners. This would allow companies to integrate deep research directly into their own systems and applications and adapt the technology for specific use cases. However, OpenAI emphasizes that the API release will only occur once the "persuasion risks" have been sufficiently clarified. This indicates that OpenAI takes the potential risks of deep research, particularly regarding manipulation and disinformation, very seriously.

Dynamic query limits

The introduction of usage-based scaling for teams. This could mean that teams that use deep research extensively would receive more flexible query limits or be able to book additional capacity. Dynamic adjustment of usage limits would make it easier for organizations to optimally integrate deep research into their workflows.

Unresolved challenges and research needs

Despite the impressive progress, open questions and challenges remain regarding deep research and AI-supported knowledge work in general. Critics, for example, question whether current citation mechanisms meet scientific standards. A case study from scientific literature analysis shows that while deep research correctly cited relevant studies in 87% of cases when analyzing Oct4 protein modifications, it included outdated or irrelevant sources in 13% of cases. This example illustrates that quality assurance and critical appraisal of AI system results must continue to play a crucial role.

The question remains how the wider availability of deep research will affect the world of work and the role of knowledge workers. Will deep research truly transform "weeks of work into minutes," as Kevin Weil predicts? Or will it prove to be just another AI tool with limited practical use? The answer to these questions will depend largely on how companies and individuals adapt this technology and integrate it into their workflows. What is certain, however, is that the era of agent-based research has begun and will fundamentally change the way we acquire and process knowledge.

A turning point in AI-supported knowledge work

The opening of Deep Research to a wider audience marks a turning point in AI-powered knowledge work. The tool offers researchers, analysts, and knowledge workers across various fields unprecedented efficiency gains and new opportunities for knowledge acquisition. At the same time, important questions remain regarding quality assurance, ethical responsibility, and the impact on the world of work. OpenAI's decision not to offer Deep Research via an API for the time being underscores the company's cautious approach to potential misuse risks and the need to develop the technology responsibly. For organizations, the integration of such tools is increasingly becoming a competitive advantage, provided they simultaneously develop the necessary skills for critically evaluating the results and using this technology responsibly. The coming months and years will show whether Deep Research truly has the potential to fundamentally transform knowledge work and usher in a new era of AI-powered knowledge acquisition.

Xpert.Digital possesses in-depth knowledge across various industries. This allows us to develop tailored strategies precisely aligned with the requirements and challenges of your specific market segment. By continuously analyzing market trends and monitoring industry developments, we can act proactively and offer innovative solutions. The combination of experience and expertise generates added value and provides our clients with a decisive competitive advantage.

More information here:

• Benefit from Xpert.Digital's 5 areas of expertise in one package – starting from just €500/month

☑️ SME support in strategy, consulting, planning and implementation

☑️ Creation or realignment of the digital strategy and digitization

☑️ Expansion and optimization of international sales processes

☑️ Global & Digital B2B trading platforms

☑️ Pioneer Business Development

 

I would be happy to serve as your personal advisor.

You can contact me by filling out the contact form below or simply call me on +49 7348 4088 965 .

I'm looking forward to our joint project.

 

 

Xpert.Digital is a hub for industry focusing on digitalization, mechanical engineering, logistics/intralogistics and photovoltaics.

With our 360° Business Development solution, we support renowned companies from new business to after-sales.

Market intelligence, smarketing, marketing automation, content development, PR, mail campaigns, personalized social media and lead nurturing are part of our digital tools.

You can find more information at: www.xpert.digital - www.xpert.solar - www.xpert.plus