I found an odd gap in the Deep Research workflow.
I ran Deep Research inside a normal ChatGPT conversation. After the research completed and the report appeared in the thread, I continued in that same conversation and asked:
“Thoughts?”
ChatGPT knew that Deep Research had just been run, but it could not see the completed report and asked me to provide it again—even though the report was still visibly sitting directly above us in the same conversation.
Here’s the interesting part: we tested whether the report was actually unavailable, and it wasn’t.
ChatGPT searched its available artifacts, found the exact completed report identified as a deep_research_report, opened it, and successfully read the entire report. I didn’t have to export it, upload it, or rerun the research.
So the report itself was still accessible to ChatGPT. The problem was that the connection between the current conversation and its own Deep Research result was not made automatically.
SUGGESTION
Deep Research reports should remain reliably accessible to ChatGPT within the same conversation in which they were generated.
If I run Deep Research and then continue that same thread with something like:
• “Thoughts?”
• “What did the research miss?”
• “Do you agree with its conclusions?”
• “What should we investigate next?”
• “Summarize that report.”
ChatGPT should recognize that I’m referring to the Deep Research report in the current conversation and use it.
I don’t think the entire report necessarily needs to remain permanently loaded in the model’s active context. If retrieving it on demand is more efficient, that’s fine. From the user’s perspective, the implementation shouldn’t matter.
The important behavior is:
Research → report appears in conversation → continue discussing that report in the same conversation
The user shouldn’t have to:
Research → discover ChatGPT can’t see the report → export/save it → upload or otherwise provide it back to ChatGPT → continue discussing it
That workaround is especially backwards because discussing the research with ChatGPT may be what helps the user decide whether the report is worth saving or preserving in the first place.
There have been other reports of Deep Research results becoming inaccessible after follow-up messages. What seemed notable about this case is that the report had not actually disappeared. Once ChatGPT explicitly searched for it, it successfully found and read the existing Deep Research report.
That suggests at least some of these failures may be a same-conversation continuity/retrieval problem rather than the research itself being lost.
EXPECTED BEHAVIOR
If a ChatGPT-generated Deep Research report is still associated with the current conversation, subsequent turns in that conversation should be able to reference it without requiring the user to manually provide the report again.
If it has fallen out of the model’s immediate context, ChatGPT should retrieve the existing report before asking the user to export, paste, save, or upload it.
From the user’s perspective, if the report is right there in the same thread, “What do you think?” should just work.