You edited a document, re-uploaded it to a Project or chat, and asked ChatGPT to re-summarize, but the answer still references the old text you deleted. It is rarely one thing: a same-named upload can leave the old copy in place, retrieval can pull the stale vector chunks, and the chat history can pin the earlier answer.
The fastest fix, in order: open the Project’s file list, delete the old copy first, then upload the new file, then start a New chat in that Project. If you re-upload over a same-named file, ChatGPT 2026 shows a Skip / Upload anyway prompt. Skip keeps the old version; Upload anyway adds a second file with the same name. So delete first, or rename the new file (report-v2.pdf) so the retrieval key changes.
Which bucket are you in?
| Symptom | Likely cause | Go to |
|---|---|---|
| Same filename appears twice when you ask ChatGPT to list files | Duplicate uploaded with Upload anyway | Cause 1 / Step 2 |
| New chat gives the new answer; old chat keeps the old one | Chat history pinned the earlier summary | Cause 2 / Step 4 |
| Works after a page refresh and ~60s wait | CDN / client preview cache | Cause 3 / Step 1 |
| Quotes a sentence in neither version | Stale vector chunks from an even older copy | Cause 4 / Step 5 |
| New file isn’t listed at all | Uploaded to the wrong chat / account / workspace | Cause 5 / Step 1 |
Common causes
1. Two copies with the same filename coexist
Since the 2026 Projects update, re-uploading report.pdf into a Project that already has report.pdf pops a conflict prompt: Skip (keep the existing file) or Upload anyway (add a second file with the same name). If you click Upload anyway, the Project now holds two files both named report.pdf, and retrieval can pull the older one, the newer one, or both. ChatGPT mixes content with no warning.
How to spot it: Ask “list every file you see in this Project, one per line.” If the same filename shows up twice, you have a duplicate.
2. Chat history pinned the old answer
In a long chat, an earlier turn already summarized the old file. New retrieval reads the new file, but the model often sticks to the answer it already “established” in the conversation. This is especially common when you ask “is the answer still X?” — the model tends to confirm its own earlier claim instead of re-reading.
How to spot it: The same question in a brand-new chat gives the new answer, but the original chat repeats the old one.
3. CDN or client-side preview cache returns the old file
Briefly after an upload, the in-app file viewer can show the old preview from an edge cache, and retrieval can hit a stale index if you ask within a few seconds of uploading.
How to spot it: Refresh the page, wait about 60 seconds, ask again, and it works.
4. Vector index not refreshed
For Custom GPTs and Projects, files are vectorized into a per-Project index for retrieval. Re-uploading the same filename does not always invalidate the old vector chunks. Depending on the deduplication path, the index can keep stale chunks alongside the new ones.
How to spot it: ChatGPT quotes sentences that are in neither the old nor the new file. Those are remnants from a third, even older version still living in the index.
5. You re-uploaded to the wrong scope
You uploaded the new file to a different chat, to your personal account when the Project lives in a Team/Enterprise workspace, or to a separate Project entirely. The chat you are looking at still sees only the old version. Note that on shared (Team/Enterprise) Projects, deleting a file removes it for everyone, so coordinate before you delete.
How to spot it: The file list in this exact Project does not include the new file at all.
Shortest path to fix
Step 1: List the files ChatGPT actually sees
Open a fresh message and ask:
List every file currently available in this Project, one per line,
with filename and the upload date/time if you can see it. Don't
summarize content yet.This catches duplicates immediately. If the new file is not listed, the upload never reached this scope (see Cause 5) — go re-upload into the correct Project/chat.
Step 2: Delete the old file before uploading the new one
The reliable workflow:
- Open the Project, then open its Files (sources) list.
- Click the old file, then Delete. On a shared Project this removes it for all members.
- Wait 10-30 seconds for the index to update.
- Upload the new file. Because the old one is gone, you will not see the
Skip / Upload anywayprompt. - Start a New chat inside the Project.
- Ask your analysis question.
Deleting first is what avoids the duplicate entirely. If you skip the delete and hit the conflict prompt, choosing Upload anyway is what leaves you with two same-named files.
Step 3: Rename the file to bust the cache
If you cannot delete the old file (shared Project you do not own, archived chat), rename the new file before uploading: report.pdf becomes report-v2.pdf or report-2026-06-15.pdf. A new name means no conflict prompt and a distinct retrieval key. Then reference it by exact name:
Use "report-v2.pdf" to answer. Ignore "report.pdf" if it appears.The unique filename forces a different retrieval path.
Step 4: Open a brand new chat
Project files persist across chats, but per-chat conversation history does not. If the old chat is anchored to the old content (Cause 2), the cleanest reset is New chat inside the Project. The new chat reads the latest file with no anchoring from earlier turns.
Step 5: Force a re-index by removing and re-adding the file
For stubborn vector-index issues, remove the file from the Project entirely (do not just upload over it), wait about 60 seconds, then re-add it. This is more reliable than replacing in place because removal clears the old chunks. Some users also report that toggling a Project’s share setting nudges an index rebuild.
To verify the new version is the one being read:
Quote the first sentence of "report-v2.pdf" verbatim. Then quote
the last sentence. Then quote one sentence I can use to verify
this is the version dated 2026-06-15.Compare against the actual file. If the quotes match, you are on the new version.
How to confirm the fix
Three-step verification, run in a fresh chat:
1. List every file in this Project, one per line, with the upload date if visible.
2. From "<new filename>", quote the first sentence and the last sentence.
3. Quote a sentence I can use to confirm this is the 2026-06-15 version.If the list shows only the new file, the quotes match the file on disk, and the version-marker sentence is the new one, you are safely on the new version. Any mismatch means an old copy is still in the index, so repeat Step 5.
Project file limits worth knowing (as of June 2026)
Hitting a cap can also look like “the new file didn’t take”:
- Per-Project files on Plus: about 20 files per Project. Free is far lower; Pro is effectively unlimited.
- Per-file size: up to 512 MB, but text-heavy files are also capped at roughly 2 million tokens and will be truncated past that.
- Storage: roughly 25 GB per user and 100 GB per organization.
- Upload frequency: Free about 3 files/day; Plus around 80 files per 3 hours.
If an upload silently fails because you are over a cap, the Project keeps showing the old file and you get the exact symptom in this article. Delete stale files to make room.
Prevention
- Use versioned filenames from the start:
report-v1.pdf,report-v2.pdf, orreport-2026-06-15.pdf. Each upload is a distinct file with no name collision and no conflict prompt. - Always delete the prior version when uploading a replacement. Never click
Upload anywayon a same-named file unless you genuinely want both. - After every upload, run the “first sentence / last sentence” quote test before trusting any analysis.
- For Projects you use daily, keep an upload log in the Project instructions:
current canonical files: report-v3.pdf (2026-06-15). The model sees this every turn and prefers the right version. - For one-off analyses, prefer a single-chat upload over a Project. There is no per-Project index to invalidate and the file scope is obvious.
FAQ
Why does ChatGPT keep quoting text that is in neither file? That text is from an even older copy whose vector chunks were never cleared from the per-Project index. Remove the file completely (not “upload over”), wait about a minute, then re-add it (Step 5).
Does re-uploading a file replace the old one?
No. ChatGPT does not version a file in place. A same-named upload triggers the Skip / Upload anyway prompt — Skip keeps the old file untouched, and Upload anyway adds a second file with the same name. To truly replace, delete the old file first, then upload.
I deleted the file but the old answer is still showing. Why? You are almost certainly reading an old chat whose history pinned the earlier summary (Cause 2). Start a New chat in the same Project and ask again — Project files are shared across chats, but each chat’s conversation history is not.
How long until a new upload is searchable? Usually seconds, but allow up to about a minute for the index and the file preview cache to catch up. If it is still wrong after a page refresh plus a 60-second wait, it is a duplicate or scope issue, not a cache lag.
Deleting a file in a shared Project — does it affect teammates? Yes. On Team/Enterprise (shared) Projects, deleting a file removes it for everyone in the Project. It can be re-added later, but coordinate first so you do not pull a file someone else is mid-analysis on.
Related
- ChatGPT project knowledge stale after edit
- ChatGPT project instructions ignored after update
- ChatGPT project files not referenced
- ChatGPT file disappears from conversation
- ChatGPT Project vector search misses
- ChatGPT Project Rename Breaks Link
External references: Projects in ChatGPT (OpenAI Help Center) and the OpenAI status page if uploads are failing platform-wide.
Tags: #ChatGPT #ChatGPT files #Troubleshooting #Debug #file-version