TL;DR
You get one real shot at the counter-offer call, and most people walk in having never said their number out loud. Use an AI chatbot as a free sparring recruiter: feed it your offer, your floor, and real market data, then have it push back across five rounds and flag exactly where you hedge. The prompt below does this. Use ChatGPT’s voice mode (free, ~2 hours/day) if you want to drill the call out loud, or Claude if you want longer, more consistent role-play. Bring real numbers — anchor 10-15% above your real target, not above the offer.
Why rehearsal beats research
You already know to research the band. The gap is verbal: the recruiter calls back, asks “So, what are you thinking?”, and you have a fraction of a second to name a number cleanly. Anchor too low and you lose money you cannot recover — a $5,000 base shortfall compounds through every raise and bonus for years. Pause too long or over-explain and the recruiter walks you down.
The fix is reps. An AI recruiter does not get tired, does not judge you, and will run the same uncomfortable objection at 11 p.m. until your reply stops wobbling. That is the whole value: the real call should not be the first time you hear your own ask.
Pick the right tool for the drill
Both major chatbots role-play a recruiter well. The difference is how you want to practice (as of June 2026):
| Tool | Best for | Voice | Free tier |
|---|---|---|---|
| ChatGPT (GPT-5.5) | Spoken drilling — say the words out loud | Yes, Advanced Voice on free tier (~2 hrs/day) | Free; Plus $20/mo for higher limits |
| Claude (Sonnet 4.6 / Opus 4.7) | Long, consistent role-play; stays in character | Beta voice on all plans + mobile (~20-30 convos/day free) | Free; Pro $20/mo |
| Gemini 3.1 Pro | If you already live in Google Workspace | Yes (mobile) | Google AI Pro $19.99/mo |
Practical call: do your text reps in Claude (it holds the recruiter persona without drifting into a pep talk), then run the final dress rehearsal in ChatGPT voice mode so you hear yourself answer. If you want the full on-site comparison, see ChatGPT vs Claude vs Gemini.
What to feed the AI
The model has no idea what a senior PM in Berlin actually earns. Tell it “senior PM, Berlin” with no numbers and it will invent a range, and you will rehearse against fiction. Ground it:
- Role, level, location
- The actual offer: base, equity, sign-on, bonus, benefits
- Your minimum acceptable total comp (your walk-away floor)
- Market data with sources — Levels.fyi is the gold standard for tech total comp; Glassdoor is a useful reference but its reported numbers can run 15-20% off verified market, so treat it as a pattern, not an anchor
- Your current comp and your reason for moving
- Anything you cannot move on (start date, location, remote)
If you are in a pay-transparency state — 16 states plus Washington D.C. require posted salary ranges as of 2026, including California, New York, Colorado, Illinois, and Washington — the job ad’s posted range is real leverage. Quote it back.
The prompt
Play a recruiter pushing back on my counter-offer. Stay in character; do not coach me until the final summary.
Role and level: [line]
Location: [line]
Offer details: [base / equity / sign-on / bonus / benefits]
My counter (total comp): [number]
My minimum: [number]
Market data with sources: [list]
Current comp: [line]
Things I cannot move on: [list]
Rules:
- Push back firmly but professionally, like a senior recruiter who likes me but has a budget.
- Use real objection patterns: "we have a band," "that's at our ceiling," "we'd need to take it to the comp committee," "what if we add equity instead of base?", "the posted range tops out lower than that."
- After each round, in one line, name what you observed in my reply: hesitation, over-explaining, hedging, dropping my number unprompted.
- Run 5 rounds. Then summarize: my strongest move, my weakest move, and the one sentence I should rehearse before the real call.For a competing offer, add: Now play a recruiter who knows I have a competing offer — push back by questioning that offer's credibility and timeline.
How to anchor (with real numbers)
The single rule: do not anchor at your real target, because negotiations gravitate toward the middle. As of June 2026, the standard guidance is to counter 10-15% above the number you actually want, so the “compromise” lands on your target.
| Situation | Counter above the offer |
|---|---|
| Offer is below market, normal leverage | 10-20% |
| Lower leverage, no competing offer | 5-10% (push benefits and bonus instead) |
| High-demand role with a competing offer | 15-25% |
| Clear lowball, well below your researched range | 20-30% |
Example: offer is $80,000, your research supports $85,000-$90,000. Counter at $90,000-$92,000. The employer “wins” by walking you to ~$87,000-$88,000 — still well above where they opened. If you had countered at $85,000, the middle would have dropped you below market. (See Harvard’s Program on Negotiation for the underlying anchoring research.)
Reading the AI’s output
A useful run looks like this:
- The pushback uses real objection patterns, not a generic “we can’t do that.”
- Each observation points to a specific phrase you used — “you said ‘I think’ before your number” — not vague “you sounded uncertain.”
- You hold your minimum for all 5 rounds. If you cave, the fix is returning to data, not emotion.
- The final summary names one short, memorizable sentence to drill. That sentence is your rehearsal priority.
If the AI just agrees with you or folds after one push, it is not stress-testing anything. Tell it to be harder: Be a tougher recruiter — push twice before conceding anything, and never concede base on the first ask.
Common mistakes
- Anchoring low to fill the silence. Silence is your tool, not the recruiter’s. Name your number and stop talking.
- No data behind the ask. “I think I’m worth more” loses every time; “Levels shows the median for this level here at $X” does not.
- Talking past the number. You already said it. Every extra sentence invites a “well, actually.”
- Justifying with need. Recruiters do not negotiate your rent. They negotiate market value and budget.
- Treating the recruiter as the enemy. They often want you to win inside their band — make their internal case easy.
- Forgetting the non-base levers. Sign-on, equity refresh, vacation, start date, and title all move when base is stuck.
FAQ
What if they say “this is final”? They rarely mean it on the first pass. Acknowledge it, restate the value you bring, and ask a sideways question: “Is there flexibility on sign-on or equity even if base is fixed?” That reopens the door without restarting the fight.
Should I disclose a competing offer? Only if it is real and only with specifics (company tier, total comp, timeline). A vague “I have other things going on” reads as a bluff and weakens you.
Is voice practice worth it over text? Yes, for the final rep. Text proves you know what to say; voice proves you can say it cleanly under a half-second of pressure. ChatGPT’s free Advanced Voice Mode (about 2 hours a day) is enough for a dress rehearsal.
What about written negotiation? Same prep, different cadence. Email lets you draft and delete, so practice tight written replies separately — and never put your floor in writing.
Which AI gives the most realistic recruiter? Both ChatGPT and Claude are strong. Claude tends to hold the persona longer without slipping into coach mode, which makes its pushback feel more like a real adversarial call.
Related
- Salary expectation phrasing AI — the first-pass phrasing before the counter
- Salary negotiation prompts — additional negotiation prompt variants
- Recruiter reply prep AI — replying to recruiter messages
- Mock interview AI — the same simulation pattern for interviews
- Self introduction — the interview opening that leads to the offer
- Job description analysis — JD details you can use in negotiation
Tags: #AI writing #Job search #Workflow