Most people go into a salary negotiation having rehearsed what they want to say. Almost nobody has rehearsed what the other person is likely to say back.
That asymmetry is the whole problem. You’ve convinced yourself the number is reasonable. You haven’t tested whether you can actually defend it under mild pressure — or whether you’ll fold the second someone says “our budget is capped this cycle.”
AI is genuinely useful here, not because it knows your company’s compensation bands, but because it can simulate the uncomfortable parts of the conversation cheaply, before they matter.
Start With the Number, Not the Script
The worst thing to ask AI is “help me write a negotiation script.” Scripts break the moment the conversation goes somewhere you didn’t expect.
Start differently. Give it your situation — your current salary, what you’re asking for, your role, roughly how long you’ve been there — and ask it to play a skeptical hiring manager or direct line manager who genuinely likes you but has real budget constraints. Then just start the conversation.
Something like: “I’m preparing to ask for a move from £48k to £55k at my current company. I’ve been here two years, I manage one person, and I’m in a product-adjacent ops role. Play my manager — skeptical but fair. I’ll make my case and you push back realistically.”
Claude handles this better than ChatGPT in my experience. It holds a character more consistently over several exchanges and doesn’t immediately capitulate when you make any halfway decent point. ChatGPT tends to validate you too quickly, which defeats the purpose entirely.
The Pushback You Actually Need to Practice
There are four or five responses that come up in almost every negotiation, and most people haven’t thought through their answer to any of them until they’re sitting across from someone saying them:
- “That’s above our band for this level.” — What do you actually say? Do you ask to see the band? Do you ask what it would take to move levels? Do you pivot to other comp?
- “We’re doing a full comp review in Q4, let’s revisit then.” — Is that a stall or a genuine commitment? How do you get it in writing without sounding like you don’t trust them?
- “What are you basing that number on?” — Can you actually answer this with anything more specific than “I feel like I deserve it”?
- “We just gave you a raise six months ago.” — True or not, how do you respond without getting defensive?
- Silence. Just silence after you name your number.
Each of those needs a real answer, not a theoretical one. Saying it out loud — even typed, in a chat window — is meaningfully different from just thinking it. You find out fast whether your reasoning actually holds.
Using AI to Build the Evidence Layer
The part where most people’s negotiation falls apart isn’t confidence, it’s evidence. “I think I’m underpaid” is not an argument. “Levels.fyi and two recent offers I’ve had put this role at £52-58k in this market” is one.
AI can help you structure a case: which categories of evidence actually move the needle in a comp conversation (market data, scope creep since your last review, measurable outputs, retention cost to the business), and which arguments tend to backfire. “I have a higher cost of living now” is the main one to drop — that’s entirely your problem, not theirs, and saying it hands them the frame that this is about your needs rather than your value.
What AI can’t do is give you accurate, current salary benchmarks for your specific role in your specific city. It will try. The numbers will sound plausible and often aren’t. Use Glassdoor, LinkedIn Salary, or Levels.fyi for actual data — then paste that data back in and ask AI to help you frame it persuasively.
The Debrief Is Where It Actually Gets Useful
After the roleplay, ask it to debrief you.
“Based on that conversation, where did my reasoning feel weakest? What pushback did I not have a good answer to? What would a stronger version of my case look like?”
This is the part most people skip because the roleplay feels like enough. It isn’t. The debrief surfaces specific gaps — not “be more confident” advice, but something like “you couldn’t explain why you want this number rather than £53k or £57k, which means you don’t have a real anchor.” That’s fixable before the actual conversation. It isn’t fixable in the moment.
One Thing It Won’t Tell You
AI has no idea how your specific manager negotiates, how your company’s finance cycles actually work, or what mood the room is going to be in. It also can’t tell you whether you have real leverage — whether you’re genuinely replaceable, whether they know you’ve been interviewing, whether headcount pressure makes this a bad month to ask.
Those things matter enormously, and no amount of prep replaces reading the actual situation. The AI prep handles your side of the table. You still have to read theirs.
Most people walk in underprepared on their own side — fuzzy on their number, no practiced response to the first pushback, evidence that collapses under a single “where’d you get that?” question. Fixing that takes about forty-five minutes with a chat window.
Before the conversation, run through the four pushback scenarios above. Out loud if you can, typed if not. Find the one you can’t answer cleanly. That’s the one to work on first.
