How to Upload Knowledge Files to a Custom GPT

Knowledge files let a custom GPT answer from documents you control: a product manual, a policy handbook, a price list. Instead of relying on the model's general training, the GPT searches your uploaded files and grounds its answers in them. This guide covers adding files, writing instructions that force the GPT to use them, and verifying it actually does.

What you need

• A custom GPT you can edit (or create a new one first)
• Your source documents as PDF, DOCX, TXT, CSV, or similar
• Files under the per-file size limit (keep each well under 512 MB and prefer text-heavy docs)

Step 1: Open the GPT in Configure mode

Go to GPTs, open your GPT, click Edit GPT, then the Configure tab. Scroll to the Knowledge section, which sits below Instructions and above Capabilities.

Step 2: Upload your files

Click Upload files in the Knowledge section and select your documents. Clean text PDFs and DOCX files work best. Scanned images of text often fail because there is no selectable text to read, so run those through OCR first or convert them to a text format.

Step 3: Force the GPT to cite the files

By default the model may blend its own knowledge with your files. Add explicit instructions so it prefers and references the documents, and admits when something is not covered.

When answering, search the uploaded knowledge files first.

- Base answers on the documents. Quote the relevant policy or spec where useful.
- If the files do not cover the question, say "That isn't in my documents"
  rather than guessing from general knowledge.
- When a price or rule is involved, state which file it came from.

Step 4: Test with a question only your files can answer

Save the GPT, then ask something specific that lives only in your documents, like a part number or a clause in your return policy. A grounded answer that matches the file means retrieval is working.

Result

Your GPT now answers from your own documents and tells you when a question falls outside them, which makes it dependable for support, onboarding, or internal lookups.