Most salary conversations fail before anyone sits down at the table. Not because the numbers are wrong, but because the person asking has never heard their own argument out loud. They rehearse in their head, where every objection gets a polite answer and the manager nods approvingly. Reality is different: the manager checks Slack, asks "and what makes you say that?", and the carefully built case collapses into "well, I just feel like I do a lot."
There are two ways to prepare. The intuitive one — talk it through with a friend, write a few bullet points, hope for the best. And the structured one — a rehearsal with a language model that plays your manager, pushes back on every claim, and forces you to hold a pause instead of filling silence with concessions. This roundup is about the second path: a set of prompts for a talk with your current boss, for answering an offer, and for haggling with a recruiter. Each one is written to be pasted as-is and adapted to your situation.
Two approaches, and what each one costs you
Intuitive prep is fast and free. You know your context better than anyone, and a friendly conversation surfaces things you genuinely care about. The problem: friends are agreeable. They don't attack your weakest claim, and they don't make you sit in silence after you name a number.
Structured LLM rehearsal costs you an evening and some discomfort. In exchange you get a tireless opponent who will ask the same awkward question ten times, a written record of where your argument leaks, and the ability to replay the exact moment you fumbled.
| Intuitive prep | LLM roleplay | |
|---|---|---|
| Cost | One coffee chat | 1–2 hours of practice |
| Feedback quality | Warm, vague | Blunt, specific, repeatable |
| Weak spots surfaced | Rarely | Every round |
| Best for | Low-stakes asks | Offers, counteroffers, promotions |
The honest trade-off: roleplay can make you over-rehearsed and stiff. If your delivery starts sounding like a call-center script, stop — the goal is reflexes, not a monologue.
Prompt 1: Build the case file before you build the ask
Before any roleplay, you need ammunition. This prompt turns the model into a cold auditor of your achievements and forces every claim into evidence.
You are a compensation analyst preparing a promotion case for me.
My role: [title], [X] years at the company, current salary [amount].
Here is a raw dump of what I did this year: [paste notes].
For each item: classify it as (a) measurable business impact,
(b) scope/ownership growth, or (c) just doing my job.
Rewrite (a) and (b) into one-line claims with a metric and a date.
Flag any claim a skeptical manager could dismiss in one sentence,
and tell me what evidence would fix it.
Example use: you paste twelve messy notes. The model returns four strong claims, five "that's your job" items, and three flags like "reduced churn — by how much, over what period?". That flag list is your homework before the real rehearsal.
Prompt 2: The manager who doesn't want to say yes
This is the core rehearsal prompt. The model plays your actual manager — with a budget constraint and a personality — and does not break character.
Roleplay my manager in a salary review meeting.
Context: [company size, team, budget cycle, their style].
Their constraints: no budget increase approved this quarter,
they dislike ultimatums but respond to retention risk.
Rules: stay in character, push back on every claim,
ask "what exactly changed since last year?",
and never offer a number first. After the roleplay,
list my three weakest moments verbatim.
Run it three times. First round: you'll fold in under two minutes. Second: you'll start holding the pause. Third: you'll notice the manager's objections repeat — which means you now know the real objections.
Prompt 3: The pause trainer
Silence is the cheapest negotiation tool and the hardest to use. This prompt isolates it.
Act as a negotiation coach drilling silence.
I will state a number: [target salary].
You respond with one short objection, then stop.
I must reply with at most one sentence, then type [PAUSE].
Score each round: did I fill the silence, apologise, or lower the number unprompted?
Ten rounds.
Watching yourself type "well, I'm flexible on that" after a pause is uncomfortable and extremely instructive. Most people discover they concede within the first three seconds of silence.
Prompt 4: Answering an offer without burning it
A recruiter offer is a different game: the clock is ticking, and enthusiasm matters as much as the number.
I received an offer: base [X], bonus [Y], equity [Z], start date [date].
My market range from public sources: [range].
Draft a counter that: thanks them sincerely, anchors on one number,
justifies it with two concrete facts, and asks for a reply by [date].
Then write the version I should NOT send, and explain why it fails.
Real case: a candidate with a 95k offer and a 108k target used this to draft a counter anchored on scope of the role rather than on personal need. The recruiter moved to 102k plus a signing bonus — not the full ask, but a clean win without a single ultimatum.
Prompt 5: The recruiter who asks your number first
"What are your salary expectations?" is a trap if you answer with a single figure. This prompt builds a deflection that stays polite and keeps information asymmetric.
You are a recruiter who always asks for my expected range in the first call.
I will practise deflecting: ask about the approved band for the role,
restate my flexibility, and give a range only if they give theirs first.
Push me twice. If I name a single number, end the call and tell me why.
Run it until the deflection sounds natural. The skill transfers directly to real calls.
Prompt 6: The counteroffer you're offered to stay
When you resign, a counteroffer often appears. Deciding in the moment is a mistake; deciding in advance is not.
My manager just offered me [X] to stay after my resignation.
Compare it against the new offer: [details].
List what a counteroffer typically does NOT fix — reporting line, scope, team, growth path.
Give me three questions to ask before accepting, and a polite script to decline.
This is where an LLM earns its keep: it separates the money from the reason you started looking in the first place.
Where this breaks down
The model doesn't know your company's politics, your manager's mood, or whether the budget is genuinely frozen. Treat its objections as a stress test, not a prediction. If a rehearsal makes you sound robotic, cut the script and keep only the evidence. And never let the model invent market data — feed it ranges you verified yourself. See also , and Production ML (MLOps): Deploying ML Models to Production — Myths and Reality.
Pick one prompt, run it tonight, and record yourself. The gap between what you planned and what you actually said is the whole negotiation — and it's cheaper to find that gap in a chat window than across a desk. See also , and Return Money for a Lost Parcel.