Key Points
- Users can limit Deep Research to a predefined list of trusted websites.
- Connected apps can be added as data inputs alongside web sources.
- A new report viewer formats long outputs with clear sections and visible citations.
- Targeted sourcing helps reduce irrelevant content and improve citation reliability.
- Limitations include site blocks, paywalls, and changing web pages.
Targeted Source Controls
OpenAI’s latest update to ChatGPT Deep Research gives users the ability to define a shortlist of domains that the model may draw from during a research run. Rather than casting a broad net across the internet, the system now stays inside a user‑specified set of websites, which can include vendor documentation, official agency pages, or any other trusted sources. This tighter scope is intended to cut down on irrelevant material and make the resulting citations more dependable.
App‑Based Input Integration
The upgrade also allows connected applications to serve as additional inputs. Users can blend internal data from supported apps with web‑based sources, creating a richer context for the research task. While the exact list of supported apps and permission details are not fully spelled out, the feature is positioned as a way to incorporate proprietary information alongside public content.
Dedicated Report Viewer
To improve the readability of long‑form outputs, OpenAI has introduced a dedicated report viewer. The viewer presents the generated content with clear sectioning, visible citations, and a document‑like layout. This design is meant to make verification easier, encouraging users to spot‑check key claims rather than skim through a continuous scroll.
Practical Benefits and Limits
For professionals who rely on repeatable research workflows, the new controls represent a workflow upgrade. By building short source lists per topic and running queries within those boundaries, users can quickly confirm that the output stayed within the allowed domains before sharing it. However, the system still faces real‑world constraints: some sites block automated access, paywalled content remains out of reach, and web pages can change without warning. These factors will influence the overall reliability of the experience.
User Recommendations
OpenAI suggests hands‑on testing as the best way to assess the new features. Running a real task, checking the sources, and exporting only when citations hold up are recommended steps. The combination of targeted source steering, app integration, and a structured viewer aims to reduce the time spent on verification and increase confidence in the generated reports.
Source: digitaltrends.com