A draft from your notes is a start, not a finished answer

Start here

The task itself is plain: paste rough notes into an assistant, ask for a draft, and turn something scattered into something readable. The step most people skip is the second half — checking the draft against the notes before anyone else sees it. Vendors describe file upload as a way to work with a document rather than retype it. The ChatGPT file uploads FAQ states the feature exists to help 'combine or analyze information from files and documents to create something new,' a fair description of drafting from notes. That framing matters: the tool is described as a synthesis aid, not a fact-checker of its own output.

What the documents say

Three vendor help pages describe the same input differently. The ChatGPT page states a 512MB hard limit per file and a 2 million token cap for document files in a conversation, figures that only matter if your notes run unusually long. The Claude upload documentation states that PDFs beyond 100 pages are read as text only, visual elements skipped, and that non-PDF documents undergo text extraction with any embedded images left inaccessible to the model. The Gemini Apps file upload page allows up to 10 files per prompt and states a chart generated from an uploaded spreadsheet 'doesn't link or respond to changes in the original data set' once created. None of the three pages describes a mechanism for confirming a draft is faithful to the notes; that check is left to the person asking.

Check this

Read the draft next to the original notes, line by line, rather than skimming for tone. Look for three things: a number in the draft not in the notes, a claim stated more strongly than the notes support, and a detail that quietly disappeared. Then ask the opposite question — what did I not mention that this draft assumes — because a model completing a paragraph often fills a gap with a plausible addition rather than leaving it empty. This is an editorial checklist, not a vendor recommendation; none of the three pages describes verification as part of the workflow.

What holds and what fails

The approach holds for reformatting: turning a bullet list into paragraphs, or a paragraph into an email, rarely invents facts because there is little room to add anything. It fails when the notes are thin and the draft is asked to be complete anyway — the gaps get filled with something confident that is not in the source. It also fails quietly once a document reaches a vendor's size or token limit, because the assistant may be working from a partial upload without saying so.

• Keep the original notes open in a second window while reading any draft.
• Ask for a version with no additions beyond the notes, then compare the two.
• For a long file, confirm it was read in full rather than truncated.

A draft is a rearrangement of what you gave the assistant, plus whatever it decided to add to make the rearrangement read smoothly. The verification step is what turns that rearrangement into something you can actually send.