One source. Every audience version. One prompt.

Matthew Sutherland

🔥

6h • Hire Me / Looking For Hire 💰

Someone this week dropped a 14-page client SOW into Claude and got back:

- A one-page plain-English summary for the client

- A scoped build checklist (.docx) for the dev team

- A milestone breakdown (.xlsx) for ops

One source. One prompt. Real .docx and .xlsx files attached in the response.

The pattern works for anything you have to translate for multiple audiences:

- One messy CRM export, versions for sales, marketing, and finance

- One vendor contract, versions for the client, legal, and procurement

- One SOP, versions for new hire, veteran, manager

- One audit report, versions for internal team, executive, regulator

Two things make it work on Opus 4.7:

1. It reads the source closely. Dense tables, footer terms, attached schedules, scanned pages, fine print. You don't have to pre-extract the section you want. Drop the whole thing in.

2. It produces finished files. Not "here's what your spreadsheet could look like." Actual .docx and .xlsx, ready to open. First pass usually comes back complete.

The prompt structure that holds up:

- Attach the source (PDF, scan, photo, or doc)

- Attach the rules each version must follow (audience profile, length limits, required fields)

- List what you want produced, and what stays the same vs. what changes across versions

- Close with: "List the assumptions you made, the terms you simplified, and anything from the source you left out."

That last line is what lets you trust the output. You review the decisions Claude made instead of comparing every line against the source.

If you've been running this pattern on Sonnet, try it once on Opus 4.7. The files come back more complete. The model swap is worth it when the input is dense and the output has to be finished, not draft.

What's the source document in your work that always gets rewritten three times for three audiences? Drop it in the comments. We can workshop the prompt together.

2 comments