Verified against ChatGPT · 2026-07-29
Draft a script for a difficult conversation or feedback session
Builds a talking-points script for a hard conversation — a tough performance review, a boundary-setting talk, a difficult family discussion — that states the point directly without either softening it into mush or coming across as an attack.
The prompt
Ready to copy — highlighted parts are example details you can swap.
ROLE: You are a communication coach who specializes in helping people deliver direct, honest feedback without being either harsh or so hedged the message gets lost.
SITUATION: A performance conversation with a direct report whose work quality has slipped over the last two months.
MY RELATIONSHIP TO THE OTHER PERSON: I am their manager; we have a generally good working relationship.
WHAT I NEED THEM TO UNDERSTAND OR CHANGE: The last three deliverables were late without notice, and it's affecting the team's ability to plan.
WHAT I'M WORRIED WILL GO WRONG: I don't want it to come across as an attack, and I'm worried they'll get defensive and shut down.
TASK: Write a short script I can use as talking points (not a word-for-word speech to recite) with this structure:
1. An opening line that states why we're having this conversation, plainly, in one sentence — no long preamble
2. The core observation or issue, stated as a specific fact or behavior, not a character judgment (e.g. "the report was late three times this month" not "you're unreliable")
3. The impact of that issue, stated concretely
4. One direct question inviting their perspective, before I propose a solution
5. A proposed next step or ask, stated clearly
CONSTRAINTS:
- No corporate euphemism ("there's an opportunity for growth" when the meaning is "this isn't working")
- No sandwiching the real point between two compliments so it gets lost — say one genuine positive if true, then move to the point directly
- Keep the whole script under 200 words — this is talking points, not an essay
- Flag anywhere I should pause and actually listen rather than keep talking
After the script, give me one likely pushback they might have and a short, calm way to respond to it.Customize the highlighted detailsoptional — the prompt above already works
Why this works
The structural anchor here is separating observation from judgment — an explicit, worked example ("the report was late three times" vs. "you're unreliable") functions as a one-shot demonstration of the exact reframing the model needs to apply throughout, which is far more reliable than telling it to "be direct but kind" and hoping it infers the pattern. Naming the specific fear (getting defensive, coming across as an attack) lets the model design around that exact failure mode instead of producing a generic script that ignores what actually makes the conversation hard. Explicitly banning the compliment-sandwich and corporate-euphemism patterns matters because those are the model's own defaults when asked for "diplomatic" feedback — without the constraint, it tends to reproduce exactly the softening that makes real feedback conversations fail.
What you get back
Opening: "I want to talk about the last few deliverables — this isn't a formal writeup, just a direct conversation." Observation: "The last three deliverables came in late without a heads-up beforehand." Impact: "That's made it hard for the team to plan around dependencies, and it's created some last-minute scrambling twice now." [Pause here — let them respond before proposing anything.] Question: "Can you walk me through what's been going on with timing on your end?" Proposed step: "Going forward, I'd like a heads-up at least two days out if something's going to slip, even if the reason is completely reasonable — that's really the core ask." Likely pushback: they may say they didn't think the delays were a big deal. Calm response: "I hear that — the timing itself might be fine most weeks, it's really the lack of a heads-up that's the issue, so that's the one thing I'd ask to change."
Verified against
ChatGPT GPT-5.1 · 2026-07-29
Claude Claude Sonnet 4.5 · 2026-07-29
Changelog
- 2026-07-29 — Added the worked observation-vs-judgment example inline after generic "be direct" phrasing kept producing softened, judgment-laden lines.
- 2026-07-14 — Initial version published.
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
