Upload a file to Gemini and one of three things goes wrong: the upload spins forever, the upload “succeeds” but Gemini says it can’t see the file, or it reads the file but the summary is completely wrong. Each symptom has a different root cause: per-file size cap, unsupported format, missing Drive access, or a scanned PDF with no text layer. Gemini runs a format whitelist first, then a content pre-processing pass, so a failure at either stage can still show a green checkmark while the file is functionally unreadable.
Fastest fix: confirm the file is under 100 MB and in a supported format (PDF, DOCX, TXT, PNG, etc.), then re-attach. If a PDF reads but the summary is garbage, it is almost certainly a scanned (image-only) PDF — run OCR first. The rest of this guide is the full diagnosis order: format -> size -> Drive access -> content quality.
Which bucket are you in?
| Symptom | Most likely cause | Jump to |
|---|---|---|
| ”Unsupported format” / button rejects file | Format not on whitelist | Step 1 |
| Bar reaches 100% then nothing attaches | File over the size cap | Step 2 |
| Attaches but summary is nonsense | Scanned PDF, no text layer | Step 3 |
| ”This file is password protected” | Encrypted PDF | Step 4 |
| ”Connect Google Workspace” / can’t reach Drive file | Drive access not enabled | Step 5 |
| Upload only fails on one file | Special characters in filename | Step 6 |
| Web UI works but truncates a huge doc | Over the page/token limit | Step 7 |
Common causes
By frequency, as of June 2026:
1. File exceeds the size cap (most common)
In the Gemini app (gemini.google.com), the per-file cap is the same on every consumer tier. Upgrading to Google AI Pro or Ultra does not raise the per-file document cap — it raises the context window (1M tokens) and the total video/audio length you can analyze.
| Limit | Free / Google AI Pro / Ultra (Gemini app) | Gemini API (AI Studio) |
|---|---|---|
| Per file (documents, images, code) | 100 MB | 2 GB |
| Per file (video) | 2 GB | 2 GB |
| Files per prompt | up to 10 | depends on context window |
| Project storage | n/a | 20 GB, files auto-deleted after 48 h |
| Workspace accounts | IT-set (often lower) | IT-set |
Video length caps: 5 minutes (Free) extending to 1 hour (Pro/Ultra). Audio: 10 minutes (Free) extending to 3 hours (Pro/Ultra).
How to tell: the upload bar reaches 100% but no “attached” thumbnail appears, or you get a file too large error.
2. Unsupported format
Supported in the Gemini app:
- PDF, DOCX, TXT, MD, RTF, plus Google Docs/Sheets
- XLSX, XLS, CSV, TSV
- PNG, JPEG, WEBP, HEIC
- MP4, MOV, AVI (video)
- MP3, WAV, M4A (audio)
- Common code files (
py,js,ts,go,rs,java, etc.) — and a single code folder or GitHub repo up to 5,000 files / 100 MB
Not supported (will be rejected or fail silently):
- DMG / EXE executables
- Old DOC (Word 95-2003) — convert to DOCX first
- KEY / PAGES / NUMBERS (Apple native) — export to PDF first
- EPUB in most cases — convert to PDF first
Note on ZIP: the Gemini app does accept ZIP archives (up to 10 files / 100 MB inside), but a ZIP cannot contain video or audio, and extracting individual files is still more reliable.
How to tell: the upload button rejects the file or shows unsupported format.
3. Drive file but Drive access is not enabled
“Add from Drive” no longer works on OAuth scope alone. As of 2026 it requires Gemini Apps Activity (Keep Activity) turned on and Google Workspace connected to Gemini Apps. If either is off, the Drive picker either won’t open or the file returns unable to access.
How to tell:
- Clicking a Drive file gives
unable to access - A prompt to “Connect Google Workspace” or “Turn on Gemini Apps Activity” appears
4. Scanned PDF (no text layer)
A PDF that looks like a document but is really scanned page images. Gemini treats each page as an image and tries OCR:
- Short doc (under 10 pages): on-the-fly OCR usually works
- Long doc (over ~30 pages): OCR degrades and the summary becomes nonsense
How to tell: open the PDF in any reader and try to select/copy text. If you can’t select text, it’s a scan.
5. Password-protected PDF
Encrypted PDFs can’t be parsed even if they open fine on your machine.
How to tell: opening the original requires a password.
6. Special characters in the filename
Filenames with emoji, control characters, or extreme unicode can fail at the upload or processing stage.
How to tell: rename to plain ASCII and retry; if it now works, this was it.
7. Over the page / token limit (large PDFs)
Even within the 100 MB size cap, a single PDF is limited to 1,000 pages, and each page costs about 258 tokens. A 1,000-page PDF alone is roughly 258K tokens, so very large documents can blow the context window and get silently truncated — the summary covers only the first chunk.
How to tell: the summary is accurate at the start but ignores the back half of a long document.
8. Workspace IT set stricter limits
Managed work accounts often cap upload size and allowed file types below the public tier.
Shortest path to fix
Step 1: Check format and size
# Size
ls -lh your_file.pdf
# Over 100 MB fails in the Gemini app (any tier)
# Format / real file type (extension can lie)
file your_file.pdf
# Should return "PDF document"Convert unsupported formats first:
- Apple Pages / Numbers -> File -> Export To -> PDF
- Word .doc -> Save As -> .docx
- ZIP -> extract and upload individual files
- EPUB -> convert with Calibre to PDF
To upload in the web app: in the prompt box click Add files, then Upload for local files, More Uploads for code/notebooks, or Add from Drive for Drive files.
Step 2: Compress under the cap
# Compress a PDF with Ghostscript
gs -sDEVICE=pdfwrite -dCompatibilityLevel=1.4 -dPDFSETTINGS=/ebook \
-dNOPAUSE -dQUIET -dBATCH -sOutputFile=compressed.pdf input.pdf
# Image-heavy PDFs (ImageMagick)
convert input.pdf -compress jpeg -quality 50 compressed.pdfPDFs typically compress to 30-50% of original. Still over 100 MB, or over 1,000 pages? Split into batches and upload each, or move to AI Studio (Step 7).
Step 3: OCR scanned PDFs
# ocrmypdf (open source, most reliable)
brew install ocrmypdf
ocrmypdf scanned.pdf searchable.pdf
# Or online: adobe.com/acrobat/online/ocr-pdf.htmlAfter OCR, Gemini reads real text and summary quality jumps. If you can’t OCR locally, you can still nudge the model — start your prompt with: This PDF is image-only. Transcribe the text on each page, then answer. This alone often moves a scan from “I can’t read this” to usable.
Step 4: Decrypt the PDF
# With the known password
qpdf --password=YOUR_PASS --decrypt input.pdf output.pdf
# Or Adobe Acrobat -> File -> Properties -> Security -> No SecurityStep 5: Enable Drive access
1. Settings -> Gemini Apps Activity -> turn it ON
2. In the Drive picker, accept "Connect Google Workspace" when prompted
3. If still blocked: myaccount.google.com -> Security ->
Your connections to third-party apps -> find Gemini -> re-authorize
4. In the OAuth popup, check ALL permissions — don't skip DriveAs a workaround, download the Drive file and use Add files -> Upload instead; local upload doesn’t need Activity enabled.
Step 6: Simplify the filename
mv "My Report v3 - Final!!.pdf" "report_final.pdf"ASCII plus underscores; no spaces, emoji, or special characters.
Step 7: Use AI Studio or the API
If the web app keeps failing or truncates a huge document, use aistudio.google.com: create a new prompt and attach the file. The API path allows up to 2 GB per file (vs 100 MB in the app), stores files free for 48 hours, and gives 20 GB of project storage. Remember the 1,000-page / 50 MB PDF inline limits still apply per document.
Step 8: Workspace accounts — talk to IT
On a managed account:
- IT -> Admin Console -> Apps -> Google Workspace -> Gemini app
- Raise the maximum file size and enable required file types
- Confirm Gemini Apps Activity is allowed by policy
How to confirm it’s fixed
After re-uploading, ask a verification question whose answer is buried deep in the file, for example: What is the exact heading of section 4, and quote its first sentence. If Gemini quotes the real text, the file is fully readable. A vague paraphrase that dodges specifics usually means it still only sees metadata or a partial OCR.
FAQ
Why does Gemini say “I can’t access the file” right after it uploaded successfully? The format whitelist and the content reader are two separate stages. A green checkmark only means the upload landed. If the reader can’t extract text — a scanned PDF, an encrypted file, or a Drive file without access — you get a successful upload that’s still unreadable. Run the diagnosis order: format, size, Drive access, then content quality.
Does Google AI Pro let me upload bigger files? No. The per-file document cap stays at 100 MB on Free, Pro, and Ultra in the Gemini app. Pro/Ultra raise the context window to 1M tokens and extend video to 1 hour and audio to 3 hours, but not the per-file document size. For files above 100 MB, use AI Studio (2 GB per file).
My PDF uploads but the summary only covers the first part. Why? The document is over the 1,000-page limit or large enough to overflow the context window, so it gets truncated. Split it, summarize sections separately, or use AI Studio with a 1M-token model.
Gemini summarizes a scanned contract completely wrong — is it lying?
No, it’s reading page images and the OCR failed. Run ocrmypdf to add a text layer before uploading, or prefix your prompt with This PDF is image-only, transcribe each page first.
Why won’t “Add from Drive” work even though I’m signed in? As of 2026 it needs Gemini Apps Activity turned on and Google Workspace connected, not just sign-in. Enable Activity in Settings, accept the Workspace connection prompt, or just download the file and upload it locally.
Prevention
- Keep critical docs in PDF / DOCX / TXT; avoid Pages / EPUB / proprietary formats
- OCR every scanned PDF with
ocrmypdfbefore uploading - Filenames: ASCII plus underscores only; no spaces or emoji
- Turn on Gemini Apps Activity before relying on Drive uploads
- For large or long-context files, use aistudio.google.com — per-file caps are 20x wider (2 GB vs 100 MB)
Related
- Gemini Workspace integration issue
- Gemini beginner guide
- Gemini brainstorming
- Gemini Deep Research workflow
Tags: #Gemini #Debug #Troubleshooting