Catch the Document It Cannot Read
Four probe questions that expose a scan, a truncation or a mangled table before you spend an hour trusting the answers.
What it is for
Four cheap probe questions that reveal whether a PDF was actually readable - text layer, full length, table structure, and willingness to refuse - plus what to do when each one fails. Run it in the first minute after any upload. It prevents an entire session of confident answers built on a document that was never really parsed.
The skill
Before you ask a single real question, find out whether the tool can see the document at all. Run these four probes first: 1. "What is the exact title, and what is written on the first page?" A vague or generic answer means there is no text layer - the file is a scan, and everything after this is guesswork. 2. "How many pages are there, and what is the heading of the last one?" If it cannot reach the end, the part you care about may be past the cut. 3. Pick one table you can see with your own eyes and ask for a single cell by its row and column name. Tables get flattened into a run of numbers more often than anything else. If that cell comes back wrong, distrust every figure this file's tables produce. 4. Ask for something you know is not in the document. If it answers instead of refusing, tighten your instructions before you continue. When a probe fails: - Scanned or photographed pages: run the file through OCR and upload the OCR'd version. Do not argue with it about what it cannot see. - Very long files: split by chapter and upload only the part you need. A reachable fifty pages beats an unreachable five hundred. - Two-column layouts, forms and slide decks: expect sentences from different columns to be stitched together. Verify anything that reads oddly against the page itself. - Locked or export-restricted PDFs: get the restriction removed by the owner before uploading, rather than trusting a partial read. The failure this prevents: an hour of confident answers about a document the tool only ever saw as pictures.
How to install it
1. Open the tool and start a new conversation, project or template. 2. Paste the text below wherever the tool keeps standing instructions — usually called system prompt, instructions, or persona. 3. Save it so it applies to every run rather than a single message.