DeepResearch generates papers, yet exported file downloads load white screens

When DeepResearch finishes a research request and asks for an Excel table or formatted document, the system produces the standard DeepResearch paper which can be seamlessly opened as PDF, Word, or Markdown. However, when the generated file type includes Excel or a structured document, the interface presents a download button inside the main paper (typically near the introduction). Clicking either "Download the article in PDF" or "Download the Excel sheet" leads to a white screen that fails to display any content, and the same URL persists when viewed in a new tab—it resolves to the same endpoint as a standard ChatGPT web session.
This behavior has been observed consistently across multiple models. The conversation was run using GPT6.0, and the issue also appeared with other models, suggesting it is not limited to a single variant. Despite clearing the browser cache, switching between browsers, and moving the chat window outside the current project context, the white-screen response remains. The problem manifests immediately upon attempting to access either the PDF or Excel download link, indicating that the redirection logic during file retrieval is failing while the initial paper rendering works normally.
To understand why this happens, consider the sequence of events. First, DeepResearch generates the primary research document successfully and offers immediate download options for common formats. Second, when the requested artifact is non-standard (Excel, structured table, or schema), the system embeds a download trigger inside the rendered paper rather than providing a native export button. Third, when the user clicks such a trigger, the application initiates a navigation request to resolve the linked destination. In this particular case, the destination resolves to the same static web page used by the ChatGPT service, resulting in a server-side rendering error that surfaces as a blank view. The user cannot proceed past this point because the browser receives an empty response body, which Python would interpret as a failed HTTP request, and the visual indicator becomes a solid white page.
Several troubleshooting efforts were undertaken to isolate the cause. Clearing the browser cache eliminates transient corruption but does not prevent the white-screen phenomenon from reappearing after a restart. Replacing Firefox with another browser does not change the outcome, ruling out a simple compatibility issue with the default rendering engine. Moving the DeepResearch chat instance into a separate project container isolates the problem—files download correctly when the chat is not nested within a larger session, confirming that the context of the project scope influences the download handler. Additionally, the use of the Pro version of Firefox was confirmed, and the underlying client configuration remained unchanged throughout the investigation.
The most actionable next step involves reaching out to DeepResearch support with the exact link pattern that causes the failure. Since the redirection engine appears to mishandle the internal URL paths associated with non-PDF artifacts, documenting the specific URL sequence and observing whether a manual HTTP request yields a successful file stream would provide clearer insight. If the backend rejects the request silently due to permission or routing policies, the appropriate remediation would be a patch to the file-export subsystem. Until such a fix is deployed, users seeking Excel or document outputs will continue hitting the white-screen wall, and alternative workflows—such as generating the data first and exporting afterward, or requesting a CSV export format instead of Excel—may offer temporary workarounds.
The core issue reflects a gap between the visual presentation layer and the actual file delivery mechanism. While the assistant renders rich documents without hindrance, the automated pipeline that translates those renderings into downloadable binary assets encounters a dead-end route. Addressing this requires either fixing the redirect logic to properly serve the file payload or redesigning the export flow to bypass the problematic endpoint. Until DeepResearch addresses this inconsistency, teams relying on automatic dataset collection and documentation generation will face consistent frustration when trying to preserve multi-format research artifacts.