The best file-analysis prompt is not “summarize this.” It explains what decision the file should inform, who will read the answer, and what a useful result looks like. That gives the model an outcome to work toward and gives you a standard for reviewing the answer.
Choose the right home for the file
Attach a file directly when it matters to one conversation. Put it in a Workspace when several chats will need the same source material: a client project, thesis, product launch, policy library, or internal knowledge base.
A Workspace keeps related chats, folders, and ready files together. That shared context boundary is useful when the work evolves over days or weeks, because you do not have to reattach the same project brief every time. Keep unrelated work in a different Workspace so a reusable library remains clear and purposeful.
Start with the outcome, not the upload
After uploading, give LLMBase a concrete task. Ask for a decision memo, a table of changes, an explanation for a non-specialist audience, a list of risks to investigate, or a structured set of next actions.
For a long document, name the sections, dates, or entities that matter most. For a spreadsheet, identify the worksheet, columns, units, and time range. If two files should be compared, say what counts as a material difference before asking for the comparison.
For example:
Compare the renewal terms in these two agreements. Make a table with the clause, the 2025 position, the 2026 position, the material change, and the source section. Flag unclear wording rather than filling in a conclusion.
That instruction is more reliable than a broad request because it defines both the scope and the output contract.
Ask for evidence, not only conclusions
Good follow-up questions make an answer reviewable:
- Quote the relevant passage or identify the row behind each conclusion.
- Separate facts in the file from suggested next actions.
- Call out missing data, assumptions, conflicting figures, and unreadable sections.
- State which file, page, slide, sheet, or section each answer came from.
- Compare every source against the same criteria.
If the result will influence a decision, ask for a confidence label and the most important check a human should make next. The point is not to create false precision; it is to keep the conversation anchored to the material you supplied.
Treat spreadsheets differently from prose
A spreadsheet often hides its meaning in headers, dates, units, and calculation logic. Tell the model which table is authoritative and whether blank cells, merged headings, or repeated labels have a specific meaning. Ask it to reproduce the values it used before drawing a conclusion.
Uploaded workbooks are handled as documents and data. Do not assume that formulas, macros, external connections, or embedded programmes will be executed. When a calculation matters, describe the formula you expect, check the source values, and verify the output in the workbook or another appropriate review tool.
Make the output reusable
Request the format before the final answer: a short email, meeting brief, Markdown table, CSV outline, report structure, or presentation plan. A clear format turns a chat response into something a colleague can evaluate and use.
Try this prompt:
Using only the attached files, prepare a one-page briefing for [audience]. Lead with the decision required, list the evidence for each recommendation, identify missing information, and end with three next actions. Cite the source file and section for every factual claim.
Review the result against the source files before sharing it. A file upload becomes genuinely valuable when it supports a repeatable process: understand the material, test the interpretation, and move the work forward with a record of why.
