How AFileFix's Chat with PDF works

Upload your PDF. Our server extracts the text from it. With each question you ask, the document text plus your question is sent to Claude (Anthropic's AI). Claude reads the document and answers with page citations like (p. 4) pointing back to where it found the information.

The honest framing: the AI is reading the text of the document, not viewing the pages the way a human would. For text-heavy PDFs this works well. For scanned PDFs where the text is stored as images, results vary — and you'll know within seconds whether your scanned PDF has extractable text.

Where does my document go?

Plain-English walkthrough of what happens to your PDF:

• Your PDF uploads to our server over HTTPS.
• Our server extracts the text from the PDF and holds it in memory only — never written to disk.
• With each question you ask, the document text and your question are sent to Anthropic's Claude API.
• Anthropic retains API data for up to 30 days for safety review purposes, then permanently deletes it.
• Anthropic does not train its models on commercial API data — this is contractual, not opt-out. See Privacy for the full AI handling architecture.
• When you close the chat tab, the document text on our server is gone.

This is meaningfully different from uploading a document directly to Claude.ai or ChatGPT on a consumer plan, where your data may be used for model training unless you've manually opted out. AFileFix routes through the commercial API where training on user data is prohibited by contract.

How do I chat with a PDF?

1. Upload your PDF. Drop your PDF onto the upload area above (or click to browse). Up to 25 MB.
2. Wait for text extraction. A few seconds for small PDFs, up to a minute for large ones. You'll see a status indicator while it runs.
3. Ask a question in plain English in the chat panel — anything about the document's content.
4. Read the answer with citations. The AI responds with page citations like (p. 4). Click any citation to jump to that page in the document viewer and verify the source.

What can I ask Chat with PDF?

Works well:

• Summarize: "What are the main points of this contract?"
• Extract specific info: "What's the renewal date?" "Who are the parties?"
• Compare sections: "Does Section 4 contradict Section 12?"
• Find specific content: "Where does it talk about indemnification?"
• Translate or rephrase: "What does Section 7 say in plain English?"
• Compute from numbers in the document: "If the late fee is 5% per month, what's the penalty on $10,000 after 90 days?"

Doesn't work well:

• Questions about general world knowledge the AI can't verify against your document (current events, prices outside the document).
• Anything not present in the PDF — the AI may guess instead of saying "the document doesn't mention this."
• Image-heavy PDFs without extractable text.

How do I get reliable answers from the AI?

Best practices:

• Use the page citations. Every AI answer includes citations like (p. 4) — click them to verify the source in the document. For anything important (contracts, legal, financial, medical), treat AI answers as a starting point and verify the cited pages yourself.
• Ask specific questions, not open-ended ones. "What's the late fee percentage in Section 7?" produces more reliable answers than "Tell me about the contract." Specificity narrows the AI's search and reduces guessing.
• Check numbers carefully. Text extraction can scramble specific numbers, especially in tables. If a number matters, verify it in the source.

Common failure modes:

• Tables — text extraction sometimes scrambles cell relationships.
• Very long PDFs — the AI may miss content from pages it didn't retrieve.
• Questions the document doesn't cover — the AI may guess rather than saying "the document doesn't address this."
• Ambiguous questions — clarify and ask again.

The AI is a research aid, not a final answer for high-stakes content.

What types of PDFs work best?

Best: Text-based PDFs from any source. Contracts, reports, research papers, manuals, ebooks. The text extraction is clean and the AI has everything it needs.

Modest: Scanned PDFs — text extraction may produce sparse results depending on whether the scanned PDF has been OCR'd. You'll know within seconds whether your PDF has extractable text.

Not supported:

• Password-protected PDFs — use Unlock PDF first.
• PDFs over 25 MB — use Split PDF first to break it into smaller files.
• PDFs with content over ~200K characters — the text is truncated with a warning.

Frequently asked questions