I want to report a serious limitation with ChatGPT Projects that makes the feature ineffective for large projects.
The issue is not simply memory. The issue is access and retrieval.
I created a large Project containing multiple long-running chats for a creative continuity project. These chats contain detailed worldbuilding, canon changes, character relationships, timelines, and specific lore. The whole point of putting them into a Project was to let ChatGPT help me find and use information across them.
However, Project chats do not function like searchable source material. ChatGPT may reference some project memory or summarized context, but it cannot reliably search the full text of selected Project chats, retrieve exact details, quote relevant passages, or cite where the information came from. There also does not appear to be a user-facing search bar across all chats inside a Project.
This makes Projects very limited for large projects. They organize chats visually, but they do not give users or ChatGPT the retrieval tools needed to actually work across those chats.
The frustrating part is that ChatGPT can search the external web and can analyze uploaded files, but it cannot reliably search my own selected internal chats, even when I have intentionally placed those chats into the same Project. That means users are forced to manually find, copy, summarize, or export information they already created inside ChatGPT.
The problem is not that ChatGPT fails to remember everything permanently. Users do not need ChatGPT to memorize every detail. Users need ChatGPT to be able to search a defined body of their own information when asked.
At minimum, there needs to be either full ChatGPT-assisted project retrieval or a separate user-facing search tool.
The ideal solution is for ChatGPT to be able to search selected Project chats directly, retrieve exact passages, and cite where details came from. What Projects need is explicit retrieval functionality:
- Search across all chats in a Project.
- Search within selected Project chats.
- Let ChatGPT retrieve exact passages from Project chats on command.
- Let ChatGPT cite or link to the specific chat/message where a detail came from.
- Let users convert a chat into a Project source/reference file.
- Let users pin canonical facts or create project-level indexes/ledgers from selected chats.
- Show what project context was retrieved when ChatGPT answers.
For large writing, research, coding, planning, or worldbuilding projects, memory is not enough. Users do not need ChatGPT to remember every detail permanently. They need ChatGPT to be able to search a defined body of their own information when asked.
However, if that is technically difficult or limited by model access, then users should at least have a normal search bar that is separate from ChatGPT’s memory and retrieval limitations.
That search bar should allow users to search inside:
-
A single chat.
-
All chats in a Project.
-
Selected chats within a Project.
-
Archived chats that belong to a Project.
-
Uploaded Project files and sources.
This search tool should return exact message matches, not summarized memory. It should let users jump directly to the matching message, copy relevant passages, and identify which chat the result came from.
I would accept this even if ChatGPT itself cannot access or search the full transcript directly. The current issue is that neither the user nor ChatGPT has reliable granular access to long Project chats. If ChatGPT’s internal retrieval is limited, then the user needs an independent search interface to manually find the information themselves.
Right now, ChatGPT is effectively the only search tool for old conversations, but ChatGPT’s access to Project chats is selective and incomplete. That creates a bad workflow: users are expected to ask ChatGPT to find details, while ChatGPT may not actually have direct access to the full text.
For large writing, research, coding, planning, or worldbuilding projects, memory is not enough. Users need retrieval. Right now, Projects feel like containers, not archives. They group work together, but they do not provide the access layer needed for serious long-term projects.
Example:
I have a long-running creative project with multiple chats inside one Project. I asked ChatGPT to retrieve details about specific created planets, races, guardians, predators, and canon changes from those Project chats. These details had been discussed extensively in prior chats. ChatGPT was able to retrieve some broad names and fragments, but it could not reliably retrieve the detailed information, quote the relevant parts, or tell me exactly where in the Project chats the information came from.
This means that even after putting the relevant chats into a Project, I still have to manually remember which old chat contains which detail. But there is no proper search bar for me to search those Project chats myself. So the feature creates the expectation of continuity without giving either the user or ChatGPT the tools required to maintain it.