AI Metadata Cleaner for ChatGPT | BeforeShare
AI Metadata Cleaner by BeforeShare checks readable AI prompts, generator fields and supported provenance containers alongside ordinary file metadata before sharing. It brings the existing beforeshare.com inspection and cleanup engine into a workspace in ChatGPT.
Available in the public ChatGPT plugin directory. Install BeforeShare, then ask ChatGPT to open the metadata workspace. You can also use the website tools below.
By TuliAI · Page updated:
What can it inspect and clean?
Photos: JPG, PNG and WebP
Inspect supported AI prompt and generator fields, C2PA containers, EXIF, GPS location and readable metadata. Clean a new copy without re-encoding image pixels.
PDF documents
Inspect and remove supported author, document properties and Info/XMP metadata. Visible pages, annotations and attachments need their own review. Digital signatures may be invalidated.
Word documents: DOCX
Inspect and remove supported author properties, comments and review traces. Tracked changes resolve to the current visible document. Names in visible text remain.
How the workspace works
- Install BeforeShare in ChatGPT and start a chat asking to inspect or remove metadata.
- Choose your original files inside the workspace. Conversation attachments are not fetched automatically.
- Inspect supported fields, create cleaned copies, and re-check those copies before sharing.
Try a PDF example without using a private document
Our synthetic PDF contains a hidden author property, a name written on the page and a reviewer annotation. The published experiment removes the author property while retaining the visible name and annotation. This is one controlled example, not a guarantee for every PDF.
- Download the synthetic original below. Open the website PDF cleaner and select that file from your device; you do not need to attach it to a conversation.
- Inspect the detected properties, then create and download a cleaned copy. Keep the original so you can compare both versions.
- Open the downloaded PDF and review its visible content. Metadata cleanup does not redact the name on the page or remove the annotation. Use the study to check the measured result and its limits.
Example requests
- Inspect readable AI prompts, C2PA and hidden GPS in my image before I share it.
- Remove author details from PDF metadata while keeping its pages.
- Clean comments and tracked changes from my Word document.
- Clean several files, then use the website to create a local ZIP.
Privacy and limits
Inspection and cleaning run locally in the workspace. BeforeShare receives no file contents or inspected metadata. Your originals remain intact. File selection and cleanup require your action; ChatGPT cannot read the local findings. Saving a verified output through ChatGPT is optional and asks for your agreement before each transfer. ChatGPT then receives that output and filename under its own policies. Use the website for fully local downloads and ZIPs; ChatGPT saves batch copies individually.
Supported files: JPG, PNG, WebP, PDF and DOCX. Limits: 25 MB per file, 20 files and 100 MB per batch. This is metadata cleanup, not visible-content redaction, malware scanning or an anonymity guarantee. Videos, audio, legacy DOC files and pixel-level watermarks such as SynthID are outside scope.
Plugin privacy policy →When the result or save action is unclear
The app or its save action is unavailable
Use the website cleaner and select your original again. Switching between the website and the ChatGPT workspace does not transfer your selected files. In a supported host, saving through ChatGPT sends the selected output and its filename after your confirmation; choosing a local-only workflow means using the website.
No supported metadata is found
That describes the fields this tool can read in the selected copy. It does not establish who created the content or rule out pixel watermarks, visible names or information held by a platform. If verification is incomplete, keep the original and review the warning before deciding whether to share.