Verified against ChatGPT · 2026-08-14
Draft a first-pass answer to a recurring internal legal question so the actual lawyer only has to edit, not start from scratch
Produces a draft response to a common internal legal-adjacent question, built strictly from prior approved answers and explicitly flagged wherever the current question differs from what was previously covered, so a lawyer's review is fast and the answer never gets sent without sign-off.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Draft a first-pass response to the internal question below, using only our prior approved answers as source material, so a lawyer only has to edit this rather than write a fresh answer from nothing. THE QUESTION BEING ASKED Can a manager in the EU office approve expense reimbursements over $5,000 without a second sign-off, given our updated approval matrix? PRIOR APPROVED ANSWERS TO SIMILAR QUESTIONS Prior approved answer (US offices, 2025): managers may approve up to $2,500 alone; anything above requires finance director sign-off, per the approval matrix effective Jan 2025. WHO'S ASKING AND WHY EU regional manager asking ahead of a large conference sponsorship expense, wants to know before committing to the vendor. HOW SIMILAR THIS QUESTION IS TO WHAT WAS PREVIOUSLY ANSWERED Similar question but the prior answer was specific to US offices under the 2025 matrix — unclear if the EU office follows the same matrix or a different one. HOW TO DRAFT THIS First, identify exactly which parts of the current question are covered by the prior approved answers I've given you, and draft that portion using language adapted from those prior answers, staying as close to the original approved substance as possible rather than rephrasing it into something new that might drift from what was actually approved. Second, identify exactly which parts of the current question are NOT covered by any prior approved answer — a new fact pattern, a question about a situation the prior answers didn't address, anything at all novel — and mark that portion explicitly as requiring fresh legal input, with a one-line note on what specifically makes it different from the prior situation. Do not blend the covered and uncovered portions together in a way that makes it hard to tell which parts of the draft are backed by prior approval and which are new. WHAT NOT TO DO Do not extend a prior approved answer to cover a new situation just because it seems similar or the underlying principle feels like it should apply — a prior answer was approved for its specific facts, and extending it to new facts is a legal judgment only a lawyer should make, not something to do by pattern-matching resemblance. Do not present this draft as ready to send — it always needs lawyer review before going to the person who asked, no matter how closely it tracks prior approved language. OUTPUT FORMAT 1. Draft response, ready for a lawyer to edit, with the portions covered by prior approved answers written first. 2. A clearly separated section titled "NOT COVERED BY PRIOR APPROVAL" listing anything in the current question that goes beyond what was previously answered, with a one-line note on what's different. 3. Source note: which prior approved answer(s) this draft draws from. 4. A closing line stating this is an internal draft only, has not been reviewed by a qualified lawyer, and must not be sent to the person who asked until a lawyer has reviewed and approved it, including the portions drawn from prior approved language.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The instruction to draft only the covered portion from prior approved language, and to explicitly separate out anything not covered rather than blending it in, is built around a specific and dangerous pattern-matching failure that language models are prone to: given a prior approved answer and a new, similar-sounding question, GPT-5.1 will readily generalize the underlying principle and apply it to the new facts, because that's exactly what makes it useful for most tasks — but a legal answer approved for one specific fact pattern (US offices, a specific approval matrix, a specific dollar threshold) was approved for those exact facts, and whether the same reasoning extends to a different office, a different matrix, or a different threshold is itself a legal judgment call, not a resemblance the model is positioned to validate. Explicitly calling out the EU-versus-US distinction as an example of where similarity assessment matters illustrates why this separation is not pedantic: an EU office may follow an entirely different approval matrix, and a draft that silently applied the US-approved threshold to an EU question would hand someone specific, wrong operational guidance dressed up as previously-approved policy, which is worse than an obviously incomplete draft because it looks fully resolved. Requiring a source note tying the draft back to which specific prior answer it drew from creates an audit trail that lets the reviewing lawyer instantly verify the draft actually tracks what was approved rather than a plausible-sounding paraphrase that drifted from the original substance during rewriting — language models restate things fluently, and fluent restatement is not the same guarantee as faithful restatement, so making the source traceable is what lets a human catch drift before it goes out. The repeated, unconditional instruction that this must never be sent without lawyer review — regardless of how closely it tracks prior approved language — exists because the entire value of this workflow (saving the lawyer time) creates exactly the temptation to skip the review step on a draft that looks clean, which is the one shortcut this prompt is explicitly designed to prevent.
What you get back
DRAFT RESPONSE (covered portion, drawn from prior approved 2025 answer): Per our approval matrix, managers may approve reimbursements up to $2,500 without additional sign-off; anything above that requires finance director approval. NOT COVERED BY PRIOR APPROVAL: The prior approved answer was specific to US offices under the 2025 matrix — whether the EU office follows the same matrix or a separate regional one was not addressed and needs confirmation before this answer can be extended to the EU manager's question. SOURCE: Prior approved answer, US offices, 2025 approval matrix. This is an internal draft only and must not be sent until a lawyer has reviewed and approved it.
Verified against
ChatGPT GPT-5.1 · 2026-08-14
Changelog
- 2026-08-14 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
