OpenAI Leveled Up ChatGPT's Deep Research With GPT-5.2, More Controls And A Built-In Document Viewer

The intensifying war among large language models (LLMs) has reached a fever pitch.

After OpenAI ChatGPT's arrival, companies like Google, Meta, Anthropic and many others race towards delivering not just smarter chatbots but truly agentic systems capable of independent, multi-step work. OpenAI has long held a unique position in this battle, thanks to its early lead in consumer adoption and its relentless push toward practical, workflow-integrated AI.

While competitors have rolled out their own deep-dive research tools, often with broad web access or specialized integrations, OpenAI's latest moves solidify its edge by emphasizing control, trust, and customization in an era where hallucinations, outdated info, and unreliable sources remain persistent pain points.

This has now evolved dramatically.

ChatGPT's 'Deep Research began as a powerful but somewhat autonomous research agent, first introduced back in February 2025. Now, OpenAI has transformed it into a highly steerable, user-directed powerhouse that prioritizes authenticated and personalized data over unrestricted scraping.

At the heart of this shift is upgrading the Deep Research feature in ChatGPT with the advanced GPT-5.2 model for faster, more precise reasoning and synthesis.

Deep Research itself is an agentic capability that goes far beyond standard queries.

It autonomously conducts multi-step investigations across the web (or restricted sources), analyzing and synthesizing information from hundreds of pages to produce comprehensive, citation-backed reports.

Deep Research is designed to tackle complex tasks that would take humans hours. It functions like a virtual research analyst: given a prompt, it devises a plan, browses relevant sites, cross-references data, and compiles structured findings with clear citations for verification.

The recent enhancements, rolled out to Plus and Pro subscribers (with broader access following), take this to a new level of usability and reliability.

Users can now explicitly steer the process by curating and approving specific sources upfront.

Whether that's a list of trusted websites, official vendor documentation, regulatory agency pages, industry databases, or paid datasets.

This restriction dramatically reduces the risk of pulling from dubious or irrelevant corners of the internet, ensuring outputs stay grounded in credible, on-topic material.

Beyond websites, Deep Research integrates seamlessly with connected apps in ChatGPT, pulling in private files, enterprise data from services like Google Drive or SharePoint, or specialized tools such as financial platforms. All apps in the ecosystem are compatible, allowing for authenticated, real-time context that blends internal knowledge with external research.

Interactivity has also been supercharged.

During a research run, which can take up to 30 minutes for thorough queries, users get real-time progress tracking, the ability to interrupt with follow-up questions, add new sources mid-process, or refine the direction on the fly. This turns what was once a "set it and forget it" operation into a collaborative, guided session.

Once complete, reports appear in a sleek full-screen viewer separate from the main chat, complete with a left-side table of contents for easy navigation and a right-side list of sources for quick citation checks.

Exports are straightforward in formats like PDF, Word, or Markdown, making it simple to share or incorporate into professional workflows.

These updates address key criticisms of earlier AI research tools: lack of transparency, over-reliance on unvetted web content, and limited adaptability to specific domains like healthcare, finance, legal, or technical analysis.

By empowering users to define boundaries, prioritizing certain domains, incorporating proprietary data, or enforcing enterprise-grade permissions, OpenAI has made Deep Research not just powerful, but trustworthy and repeatable for high-stakes decisions.

In a crowded LLM landscape, this blend of cutting-edge model intelligence with genuine user agency positions ChatGPT as the go-to for professionals who need research that's both exhaustive and auditable.

While the enhanced Deep Research does pack a huge punch, it isn't flawless.

Greater control means more setup, and for quick tasks, that added friction can slow things down. The most advanced features remain behind paid tiers, limiting access. And while curated sources reduce hallucinations, they don’t eliminate bias, gaps, or outdated information. Then, there is the fact that privacy and governance also remain real considerations when integrating enterprise data.

And no matter how sophisticated the system becomes, it still depends on human judgment to interpret and validate its findings.

Deep Research may be a powerful multiplier, but it's not a replacement for expertise.