MiniFiles.io
All articles
How-to5 min20 April 2026

Clean Up Files Before Uploading to ChatGPT or Claude

When you upload a file to Claude, ChatGPT, Copilot or Gemini, more data is transmitted than you might expect: EXIF metadata in images can include GPS location, device model and timestamps; PDFs often embed revision history, author names, and hidden annotations. Compressing the file locally before uploading strips this data and reduces what you send to an AI provider's servers — without changing the content the AI needs to analyse.

What metadata AI tools receive when you upload a file

A JPEG photo taken on a smartphone typically contains EXIF metadata: GPS coordinates, camera make and model, date and time, software version, and sometimes a thumbnail of the full image. A PDF document can contain: author name, creation date, modification history, software used to create it (Word, Acrobat, LaTeX), embedded comments and annotations, and internal revision data. When you upload these files to an AI assistant, all of this metadata is transmitted along with the actual content. The AI model may not use it, but it is present in the data sent to the provider's infrastructure.

Why file size matters when uploading to AI tools

Most AI tools enforce file size limits: ChatGPT allows up to 512 MB per file, but context window limits mean very large files are truncated or summarised. Claude supports up to 32 MB per file via claude.ai. Gemini 1.5 Pro supports up to 2 GB via API but 50 MB in the consumer interface. In practice, a 20 MB PDF is often better handled by AI than a 150 MB one — the model can process the full content rather than a compressed summary. Reducing file size before uploading improves analysis quality, not just upload speed.

Step-by-step: compress a PDF before uploading to Claude or ChatGPT

Open MiniFiles.io in a browser tab. Switch to the PDF tab if it is not already selected. Drag your PDF onto the drop zone, or click to select it. MiniFiles compresses the PDF locally — images inside the document are downsampled, fonts are subset, and duplicate images are merged. Click Download to save the compressed PDF. Upload the compressed version to Claude or ChatGPT. The text content and document structure are preserved; the file is simply smaller and cleaner.

Step-by-step: strip metadata from images before uploading

Open MiniFiles.io. Drop the image file (JPEG, PNG, WebP) onto the upload zone. Compression runs locally — EXIF metadata including GPS data, device information and timestamps is stripped automatically. Download the result. Upload the stripped image to your AI tool. The visual content is identical; the embedded metadata is gone. This is particularly relevant if the image contains location data you would prefer not to transmit to a third-party service.

When this matters most

This workflow is most relevant in three situations. Uploading images that may contain GPS location data — street photos, property images, personal photos. Uploading internal business documents to a consumer AI tier (without a DPA) — stripping author names, revision history and company metadata reduces the identifiable information transmitted. Uploading medical or legal documents to an AI for analysis — compression reduces file size for faster processing and removes embedded metadata that is not relevant to the analysis.

What local compression cannot do

Compressing a file locally does not encrypt it. If you upload a compressed, metadata-free PDF to ChatGPT, the text content is still transmitted to OpenAI's servers and subject to their data policies. Local compression reduces the metadata footprint and file size, but is not a substitute for proper data governance — using enterprise tiers with DPAs, redacting personal information, or keeping files local entirely when AI analysis is not necessary.

Key takeaways

  • JPEG EXIF metadata includes GPS location, device model, timestamps — transmitted when uploading to AI tools.
  • PDF metadata includes author, revision history, software, annotations — all sent to AI provider on upload.
  • MiniFiles.io strips image EXIF metadata by default and compresses files locally before they leave the device.
  • Compression preserves text content and document structure — AI analysis quality is not affected.
  • Local compression reduces metadata footprint but is not a substitute for DPAs or data governance.

Frequently asked questions

Does compressing a PDF before uploading to Claude affect the AI's ability to read it?

No. PDF compression reduces image resolution, but preserves all text content and document structure. Claude can read and analyse compressed PDFs identically to originals.

What metadata does MiniFiles strip from images?

By default, MiniFiles removes EXIF metadata including GPS coordinates, camera make and model, date and time, software version, and embedded thumbnails. A toggle lets you keep EXIF if you need it.

My PDF is too large to upload to ChatGPT — how do I reduce it?

Open MiniFiles.io, drop the PDF in, and download the compressed version. PDFs with embedded high-resolution images typically compress by 40–70%. The compressed version uploads faster and is processed more reliably by the AI context window.

Is it safe to upload compressed files to AI tools?

Compression reduces the metadata transmitted, but the file content is still sent to the AI provider's servers. For sensitive documents, review the provider's data policy and use enterprise tiers with DPAs for GDPR compliance.

Does this work for files uploaded to Microsoft Copilot?

Yes. The same process applies — compress the file locally in MiniFiles.io before uploading to Copilot in Microsoft 365. Copilot processes files on Microsoft's Azure infrastructure; stripping metadata before upload reduces identifiable information transmitted.

Try it now — free

Compress PNG, JPEG, and PDF files in your browser. No upload, no installation.

Compress files