OpenAI reveals how deep research transforms inquiry
Presented by researchers Isa Fulford and Edward Sun during an OpenAI forum event, the new tool is powered by a fine-tuned version of OpenAI's upcoming O3 model and leverages advanced reasoning and browsing capabilities.
"Deep research is an agent in ChatGPT that can do work for you independently," Fulford explained.
"You give it a prompt, and it will find, analyse, and synthesise hundreds of online sources to create a comprehensive report at the level of a research analyst."
The system is intended to help users across a range of sectors—from academia and medicine to business and software development. "Members are finding that deep research asks clarifying questions to refine research before it even starts," said Fulford. "We think that deep research can accomplish in tens of minutes what would take a human many hours."
The model represents a major step forward in OpenAI's work with reasoning systems, building on reinforcement learning techniques introduced in its earlier models. Fulford explained how the company developed the tool: "We launched O1 in September of last year. This was the first model that we released in this new paradigm of training where models are trained to think before answering… and we called this text where the model is thinking, 'chain of thought'."
This method of structured, internal reasoning proved effective not only in tasks such as maths and coding, but also in navigating complex real-world information environments. "Around a year ago internally, we were seeing really great success… and we wondered if we could apply these same methods but for tasks that are more similar to what a large number of users do in their daily lives and jobs," Fulford said.
Sun detailed how the tool works by combining reasoning with specialised capabilities like web browsing and code execution. "The browser tool helps the model to aggregate or synthesise real-time data, and the Python tool is helping the model to process this data," he explained. The system dynamically alternates between reasoning and action, using reinforcement learning to improve over time.
One striking example involved analysing medal data from the 2020 Tokyo Olympics. "You can see how the model interleaved reasoning with actual tool calls to search for information, refine the data, and process it programmatically," Sun said.
Unlike older approaches that rely on a single-pass search or instruction-following, deep research iteratively refines its answers. "We train the model with end-to-end reinforcement learning," Sun added. "We directly optimise the model to actively learn from the feedback, both positive and negative."
OpenAI tested the model extensively against both public and internal benchmarks. According to Fulford, "the model pairing deep research scored a new high of 26.6%" on the Humanities Last Exam, an expert-level evaluation spanning over 100 subjects.
On another benchmark, GAIA, the tool also achieved a state-of-the-art result for multi-step web browsing and reasoning.
The model also underwent safety evaluations prior to release. "We did extensive red teaming with external testers, and then also went through preparedness and governance reviews that we always do at OpenAI," Fulford said.
Despite strong results, the researchers acknowledged current limitations. "It still may hallucinate facts or infer things incorrectly," Fulford said.
"Sometimes it struggles to distinguish between authoritative sources and rumours."
Use cases continue to emerge in unexpected domains. "People might be using the model a lot for coding. And that's been a really big use case," Fulford observed. Other domains include scientific and medical research, where professionals have begun verifying the model's output against their own expertise.
Users are also adapting their behaviour to suit the model. "We've seen interesting user behaviour where people put a lot of effort into refining their prompts using O1 or another model," Fulford said. "And then only after really refining that instruction, they'll send it to deep research… which makes sense if you're going to wait a long time for an output."
Currently, deep research is available to users on the Plus, Pro, Teams, Enterprise and EDU plans.
"We're very excited to release a smaller, cheaper model to the free tier," Fulford confirmed. The team also plans to improve personalisation and explore ways to let users incorporate subscription services or private data into the research process.
"This showcases how the model can effectively break down a complex task, gather information from various sources, and structure the response coherently for the user," Sun said in closing.
OpenAI's forum audience, composed of members across academia, government, and business, left the event with a clear sense that deep research marks a meaningful step toward AI systems capable of handling work currently done by skilled analysts.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Techday NZ
3 hours ago
- Techday NZ
Goodbye SEO, hello GEO: How AI is reshaping brand discovery
As language models replace traditional search, brands must master generative engine optimisation to stay visible In May, news emerged that sent ripples through the tech world: Google searches in Apple's Safari had apparently dropped for the first time in 22 years, according to testimony made by Apple's Eddy Cue during Google's antitrust trial. While later Google countered that "query growth" is up, the contradiction itself reveals a shift that's already underway. Whether the reality is declining searches or changing search patterns, the trend points in one direction: people are increasingly turning to LLMs for answers. Ask ChatGPT about the best car brands for families, and you'll get a curated list. Search for laptop recommendations, and the AI serves up specific models with reasoning. But notice which brands make those lists and which don't. This represents more than a technological preference. The rapid transition from traditional search engines to AI-powered language models represents a complete restructuring of the discovery layer between brands and customers. And many businesses haven't noticed they're already losing. This trend has given rise to what is being called generative engine optimisation (GEO), the new discipline of optimising content for AI-powered responses rather than traditional search rankings. The visibility challenge The shift from traditional search engines to AI-powered language models represents a complete reshaping of how consumers discover, evaluate, and engage with brands. Those that fail to recognise and respond to this shift risk becoming invisible at the moment of decision. Brands that are not surfaced in LLM-generated responses will see a significant decline in visibility, resulting in downstream impacts on customer acquisition, brand relevance, and market competitiveness. Importantly, this isn't a reflection of product quality, but of digital discoverability: if a brand isn't mentioned, trusted, or properly structured in the sources LLMs rely on, it may simply not exist in the eyes of the AI, or the customer. This transition marks a new brand-customer dynamic, where visibility is no longer about search engine rankings alone, but about being contextually relevant, credibly cited, and machine-readable. Brands that embrace this reality early by adapting content, enhancing structured data, and embedding themselves in trusted digital ecosystems, will establish a lasting competitive edge. Those that delay will not merely fall behind, they risk being excluded from the AI-powered discovery layer entirely. Who wins and who loses While the shift toward AI-mediated discovery poses challenges across all industries, the impact won't be uniform. The degree to which brands face existential risk from this transition depends largely on the nature of their products, customer relationships, and purchase drivers. Understanding these differences is critical for prioritising response strategies and resource allocation. Some purchases are driven by emotion or identity rather than logic, price, or features. Luxury fashion and high-end cars fall into this category, and as a result, are less likely to be impacted by shifts in consumer decision-making. In contrast, products like utilities are essential, broadly interchangeable, and often chosen based on price or promotional offers rather than brand loyalty. This divide will deepen as AI agents begin to play larger roles in decision-making and action pathways. When decision-making occurs without humans in the loop, brand presence such as mental availability becomes less relevant. Instead, the agent's choices will be driven primarily by product attributes, features, and consumer reviews. Consider a future where your AI assistant automatically switches your energy provider based on quarterly rate comparisons, or books travel based on optimal price-to-convenience ratios. As the way people search and discover products changes, brands that adapt to this new environment will be best placed to succeed. This means actively engaging with how AI systems interpret and present information. Mastery of tools like prompt engineering, investment in training partnerships, or the development of custom GPTs can all help confirm products and services are accurately and favourably surfaced in AI-mediated environments. Brands that fail to understand how their audiences now phrase questions, conduct research, or engage with offerings will miss opportunities to evolve. Without these insights, businesses won't be able to redesign digital experiences, particularly websites, to support discovery and decision-making in an AI-driven Brands Must Do Where they once poured much marketing effort into search engine optimisation (SEO), brands must now rethink their digital presence for GEO, optimising not just for humans, but for the machines mediating those interactions. A brand's digital presence and content need to be more conversational and context-rich, rather than just keyword-dense. At BRX, we specialise in AI-driven marketing and have developed strategies to help drive GEO success: Audit your brand's presence across trusted sources. Are you mentioned on CHOICE, Canstar, Reddit, Whirlpool? If not, why not? Update your structured data. Use schema markup and keep pricing, availability, and product specifications always accurate. Drive positive reviews. Be a part of brand dialogue that matters to your category like Trustpilot, Google, TripAdvisor. Optimise for natural language. Rewrite content to reflect how people actually ask questions in LLMs, not keyword-stuffing. Format your language for skim readers and scanners with a "too long; didn't read." Create content that simplifies decisions. Think comparison tables, FAQs, expert explainers, and product fit guides. These steps form the foundation of AI-ready brand presence, but success requires more than checking boxes. The brands that will really thrive in this environment will also share some common characteristics. At BRX, we think brands that win will be those who: Sit outside the dynamic of price and features Create conversational content that answers real questions Prioritise structured clarity in product and service info Maintain consistency across official and third-party sources Earn trust through verified experiences, not just promises Provide highly specific answers to hyper-individual questions In this new environment, the marketer's job is no longer just to shape consumer perception but to influence the AI's perception of their brand. To "win" in AI-generated conversations, content flooding is no longer just a visibility tactic, it's a visibility imperative. Brands that dominate the content space are disproportionately represented in LLM outputs. Flooding without a customer-centric lens, however, risks damaging the post-click experience. The challenge is balancing volume with quality. The emerging reality is that websites are increasingly serving LLMs first, customers second. This creates a tension: you may be designing an experience that LLMs find useful, but human users do not. Time to act AI systems now influence how customers discover brands, and brands need to get out ahead of this. As AI and AI agents will be increasingly making decisions without human input, brands need to audit their AI visibility, create conversational content, and balance quantity with quality in their digital presence. The first step is simple but critical: audit what AI systems currently say about your brand. Ask multiple LLMs about your product category and see if you appear. If you don't, investigate the sources they cite and understand what ecosystems are influencing their responses. Modern LLMs with research functions provide sources that serve as helpful research starting points. Use this intelligence to understand where you need to appear to start showing up in LLM results. Given the complexity and rapid evolution of AI systems, partnering with specialists who understand this landscape can accelerate your progress. BRX helps brands navigate GEO with AI-native strategies that deliver measurable improvements in AI visibility and engagement. The brands that recognise this shift early and master GEO won't just maintain their market position; they'll capture market share from competitors who remain focused on traditional search optimisation. In a world where AI increasingly mediates brand discovery, being invisible to artificial intelligence means being invisible to customers." This emphasises the value proposition of working with experts who understand the complexity and fast-moving nature of AI, rather than trying to figure it out alone.
Scoop
9 hours ago
- Scoop
AI Adoption Surges In NZ Workplaces But Training Fails To Keep Up
New research from recruitment specialists Cultivate and AI advisory reveals that while most Aotearoa workers feel optimistic about artificial intelligence (AI), critical gaps in employer support and training could leave many behind as AI adoption accelerates. The study of 829 white-collar workers, conducted in May 2025, provides a comprehensive view of how AI is impacting New Zealand's workplaces, and what lies ahead. Drawing on insights from a wide cross-section of industries, functions, and seniority levels, the research highlights that while large-scale disruption is still emerging, AI is already influencing how many workers deliver their roles, transforming processes and lifting productivity. 76% of workers report feeling positive about AI in their roles, driven largely by its power to reduce repetitive tasks and assist with communication, creativity, and information management. That positivity is building, with 62% saying they are more positive about AI today than they were 12 months ago. Yet this enthusiasm hasn't been matched by formal support, with only 13% of workers saying they've received company-led AI training. Over half have had no training at all. Despite this, 44% of workers are already using AI tools at least daily at work. 'This has been largely a bottom-up adoption story so far' says Tony Pownall, Co-Founder of Cultivate. 'Workers are embracing AI tools to make their jobs easier, but many are doing it without employer-led guidance, safeguards, or structured learning. That presents both a risk and a massive opportunity.' Among the most used tools are general AI assistants like ChatGPT, Claude and Perplexity (used by 82% of respondents), followed by integrated productivity tools like Copilot and Gemini (53%). However, only 14% reported using custom-built or specialist AI solutions, a further sign that adoption remains largely employee-led rather than organisation-driven. Attitudes to AI and uptake at work appears consistent across gender, ethnicity, and age. However, those in Professional Services, Information & Communication Technology, and Utilities & Energy sectors were most likely to have integrated AI into their workflows. By role, those in technology, consulting, and marketing functions reported the highest rates of AI use. Although only 10% of respondents have seen job losses due to AI so far, nearly half (45%) expect their team's headcount to decrease in the next three years. Two-thirds believe at least part of their current role will be automated, and 15% expect their job to disappear entirely. Casual and temporary workers are already feeling this shift more acutely. They're the most likely to have witnessed AI-related job cuts and to anticipate further reductions. Nearly a quarter expect their entire role to be automated within three years. 'We were surprised to see such optimism given the level of anticipated role change,' says Tony Pownall. 'Our view is that AI users are discovering how AI can remove the mundane and make room for more meaningful, higher value work opportunities. That shift in mindset could be key to navigating what's ahead.' AI is already reshaping the hiring landscape too. According to the research, 42% of hiring leaders have already adjusted what they look for in candidates, and 60% say they'd pay a premium for demonstrated AI capability. 'Employers aren't just looking for people who can use AI, they want people who can think critically about it,' explains Trina Jones, Co-Founder of Cultivate. 'Our data shows the most valued skill is AI awareness, an understanding of what AI can and can't do in a work context. The top behavioural trait is critical thinking, which is the ability to question AI outputs rather than accept them blindly.' Responsibility for AI readiness now sits with both employers and workers. While daily AI use is common, nearly a third of workers still use AI monthly or less, and 36% say their organisation is not yet prepared to maximise AI's potential. Lack of time, limited internal capability, and absence of AI champions were cited as the biggest barriers to organisational preparedness. This signals a significant opportunity, and an urgent need, to close the gap between AI's availability and its effective use. 'This research is a wake-up call for everyone,' says Justin Flitter, Founder of 'AI isn't just coming, it's already transforming how we work. Not knowing where to start is no longer an excuse. Those who succeed will be the ones who act now, by upskilling, experimenting, and embedding AI into daily practice.' 'AI literacy is the new digital literacy,' concludes Jones. 'To stay relevant all workers need to take ownership of their learning and be able to show how they integrate AI safely and creatively, alongside sound judgement.' About the research This research was conducted in May 2025 and is based on responses from 829 individuals currently employed in white-collar roles across Aotearoa. Respondents came from a broad cross-section of industries, functions, and seniority levels. The survey primarily engaged prior job seekers and hiring managers within Cultivate's network, alongside members of online communities coordinated by Diversity Works, MyCareerBrand, Weave Leadership, and Data was collected via an online questionnaire, and for clarity, all percentages cited in the findings have been rounded to whole numbers. Summary of research findings: 'AI at Work: How AI Is Reshaping Roles in Aotearoa' 76% of workers feel positive about AI in the workplace, driven by its ability to reduce admin, assist communication, and support creative tasks. Only 13% of workers have received any company-led AI training, despite 44% already using AI tools daily. 57% of workers have had no AI training at all. The most used AI tools are general assistants like ChatGPT, Claude, and Perplexity (used by 82%), followed by productivity tools like Copilot and Gemini (53%). 10% have already witnessed AI-driven job losses in their team, while 45% expect headcount to drop in the next three years. Two-thirds believe some of their role will be automated, and 15% expect their job to disappear entirely. Temporary and casual workers are most affected.24% believe their entire role will be automated, and 61% expect their team to shrink. 42% of leaders have adjusted their hiring criteria due to AI, and 60% say they'd pay more for candidates with proven AI capability. The top skill leaders now value: AI awareness – understanding what AI can and can't do in context (72%). The top behaviour valued: Critical thinking – the ability to question AI outputs (69%). Over-reliance on AI without critical analysis is the #1 concern among leaders (70%). The most cited benefit of AI by leaders is productivity gains through reduced admin and improved service (81%). 36% of workers say their organisation is not yet prepared to maximise the benefits of AI. The top barriers to organisational readiness: lack of time, internal champions, and confidence in ROI. About Cultivate Cultivate is a New Zealand-based recruitment agency specialising in white-collar roles across a range of sectors, including permanent, temporary, and contract placements. Known for its deep industry expertise and inclusive, people-first approach, Cultivate partners with organisations of all sizes, from fast-growing startups to established enterprises, to deliver tailored hiring solutions. The team is committed to building capable, adaptable and representative workforces, while also helping job seekers find opportunities that align with their skills, values, and career goals. Cultivate was named SEEK's Medium Recruitment Agency of the Year in 2023 and 2024. Learn more at
RNZ News
a day ago
- RNZ News
NZ's new AI strategy is long on 'economic opportunity' but short on managing ethical and social risk
By By Andrew Lensen* of Photo: Supplied/Callaghan Innovation The government's newly unveiled National AI Strategy is all about what its title said: "Investing with Confidence". It tells businesses that Aotearoa New Zealand is open for AI use, and that our "light touch" approach won't get in their way. The question now is whether the claims made for AI by Minister of Science, Innovation and Technology Shane Reti - that it will help boost productivity and enable the economy to grow by billions of dollars - can be justified. Generative AI - the kind powering ChatGPT, CoPilot, and Google's video generator Veo 3 - is certainly earning money. In its latest funding round in April, OpenAI was valued at US$300 billion . Nvidia, which makes the hardware that powers AI technology, just became the first publicly traded company to surpass a $4 trillion market valuation . It'd be great if New Zealand could get a slice of that pie. New Zealand doesn't have the capacity to build new generative AI systems, however. That takes tens of thousands of NVIDIA's chips, costing many millions of dollars that only big tech companies or large nation states can afford. What New Zealand can do is build new systems and services around these models, either by fine-tuning them or using them as part of a bigger software system or service. The government isn't offering any new money to help companies do this. Its AI strategy is about reducing barriers, providing regulatory guidance, building capacity, and ensuring adaptation happens responsibly. But there aren't many barriers to begin with. The regulatory guidance contained in the strategy essentially said "we won't regulate". Existing laws are said to be "technology-neutral" and therefore sufficient. As for building capacity, the country's tertiary sector is more under-funded than ever, with universities cutting courses and staff. Humanities research into AI ethics is also ineligible for government funding as it doesn't contribute to economic growth. The issue of responsible adoption is perhaps of most concern. The 42-page " Responsible AI Guidance for Businesses " document, released alongside the strategy, contains useful material on issues such as detecting bias, measuring model accuracy, and human oversight. But it is just that - guidance - and entirely voluntary. This puts New Zealand among the most relaxed nations when it comes to AI regulation, along with Japan and Singapore . At the other end is the European Union, which enacted its comprehensive AI Act in 2024, and has stood fast against lobbying to delay legislative rollout. The relaxed approach is interesting in light of New Zealand being ranked third-to-last out of 47 countries in a recent survey of trust in AI . In another survey from last year, 66 percent of New Zealanders reported being nervous about the impacts of AI . Some of the nervousness can be explained by AI being a new technology with well documented examples of inappropriate use, intentional or not. Deepfakes as a form of cyberbullying have become a major concern. Even the ACT Party, not generally in favour of more regulation, wants to criminalise the creation and sharing of non-consensual, sexually explicit deepfakes. Generative image, video, and music creation is reducing the demand for creative workers, even though it is their very work that was used to train the AI models. But there are other, more subtle issues, too. AI systems learn from data. If that data is biased, then those systems will learn to be biased, too. New Zealanders are right to be anxious about the prospect of private sector companies denying them jobs, entry to supermarkets , or a bank loan because of something in their pasts. Because modern deep learning models are so complex and impenetrable, it can be impossible to determine how an AI system made a decision. And what of the potential for AI to be used online to mislead voters and discredit the democratic process, as the New York Times has reported, may have occurred already in at least 50 cases. The strategy is essentially silent on all of these issues. It also doesn't mention Te Tiriti o Waitangi/Treaty of Waitangi. Even Google's AI summary tells me this is the nation's founding document, laying the groundwork for Māori and the Crown to coexist. AI, like any data-driven system, has the potential to disproportionately disadvantage Māori if it involves systems from overseas designed (and trained) for other populations. Allowing these systems to be imported and deployed in Aotearoa New Zealand in sensitive applications - healthcare or justice, for example - without any regulation or oversight risks worsening inequalities even further. What's the alternative? The EU offers some useful answers. It has taken the approach of categorising AI uses based on risk : This feels like a mature approach New Zealand might emulate. It wouldn't stymie productivity much - unless companies were doing something risky. In which case, the 66 percent of New Zealanders who are nervous about AI might well agree it's worth slowing down and getting it right. Andrew Lensen is a Senior Lecturer in Artificial Intelligence at Te Herenga Waka - Victoria University of Wellington -This story was originally published on The Conversation.
