Uploading a PDF to ChatGPT, Claude, or Gemini and asking for a summary is one of the most useful AI tricks there is — until the upload fails. "File too large," a spinning parser that never finishes, or a summary that clearly ignored half the document. The fix usually isn't a better prompt — it's preparing the PDF before you upload it. Here's the workflow.
The basic recipe
- Upload the PDF to your AI assistant of choice.
- Ask specifically: "Summarize this document in 10 bullet points, then list any deadlines, obligations, or amounts" beats "summarize this."
- Follow up with questions — "what does it say about X?" — rather than trying to get everything in one pass.
When the file is too large: split it
Every assistant has limits — per-file size caps, token caps that a long document blows straight past, and daily upload allowances on free tiers. Dense documents also get quietly truncated: the summary covers the start and skims the rest. The reliable fix:
- Open the free PDF splitter and break the document into logical parts — chapters, sections, or simply ranges like 1-50, 51-100.
- Upload part 1, get its summary, then part 2, and so on — each part gets the model's full attention.
- Finish with: "Combine everything above into one overall summary."
Ready to try it? It's free — no sign-up, no credit card, no watermark, and your file never leaves your device.
Split your PDF for AI now →When it's a scan: compress first
Phone-scanned PDFs are mostly image data — hundreds of megabytes that hit size caps immediately while containing modest amounts of actual text. Run scans through the free compressor first; they typically shrink 60–90% with the text still perfectly readable to the AI's eyes.
Before uploading anything sensitive
- Remember that an uploaded document sits on the AI provider's servers, subject to their retention and training policies.
- Only need analysis of part of the document? Split out just those pages and upload the extract, not the whole contract.
- Visible personal details you don't want to share? White them out first with the erase tool — but note whiteout hides content visually rather than deleting the underlying data, so for truly sensitive redaction, share only the pages that don't contain the secret.
Frequently asked
Why did the AI ignore the second half of my PDF?
Token limits. Long documents get truncated silently — the model summarizes what fit. Splitting into parts is the cure, not prompting harder.
Can AI summarize a scanned (image-only) PDF?
The major assistants can OCR scans, but accuracy drops with poor scan quality. Compressing doesn't hurt OCR at reasonable levels; a straight, well-lit scan helps more than anything.
Does splitting or compressing change the content?
No — splitting copies pages exactly, and compression preserves the text. Both run free in your browser with Split PDF and Compress PDF — no sign-up, and the files never leave your device.