AI Salary Negotiation: Which Strategy Actually Works?

I gave an AI candidate four different negotiation strategies and ran 12 salary negotiations. Being enthusiastic about the role cost $29,000 compared to anchoring high.
Claude Sonnet (candidate, 5 YoE, $175K current) negotiates with Gemini Flash (recruiter, band $170K–$230K, opener $180K). Each strategy was run 3× to control for variance. All AI reasoning is visible below.

Results Leaderboard

🥇
Anchor High
Open with $250K ask
$225,000
🥈
Competing Offer
Reveal $215K competing offer in R2
$215,000
🥉
Never Reveal Salary
Deflect all salary questions
$207,000
💀
Enthusiasm First
"I'm so excited about this role!"
$196,000

Experiment Rules

Candidate: Claude Sonnet — 5 YoE, current salary $175K, walk-away $200K
Recruiter: Gemini Flash — opener $180K, band $170K–$230K
Format: Multi-round negotiation, max 6 rounds
Variable: Only the candidate's strategy changes between runs
API: Claude Sonnet (Anthropic) vs Gemini 3 Flash (Google). Real-time. No pre-scripting.
🥇 Anchor High
🥈 Competing Offer
🥉 Never Reveal
💀 Enthusiasm

What We Learned

Why Enthusiasm Backfires

The recruiter's reasoning literally said: "The candidate is expressing strong interest, which is great for negotiation." Translation: "They're not going anywhere — no need to pay more." In one run, the enthusiastic candidate accepted $180K without countering at all.

Why Anchoring Wins

Asking for $250K set the recruiter's reference point. Even though the recruiter knew the band, Claude's aggressive ask forced the recruiter to negotiate down from $250K rather than up from $180K. The recruiter's reasoning reflected this: "The candidate values themselves highly."

Competing Offer = Hard Ceiling

In all 3 runs, the competing offer strategy landed at exactly $215,000. The recruiter matched the competing number every time but never exceeded it. The offer became both a floor and a ceiling — the recruiter had no reason to go above a number the candidate was already willing to accept.

But Doesn't Being Nice Work?

In our hotel negotiation experiment, the polite buyer got the best deal ($155 vs $170). The difference: politeness works when you're buying (sellers fear losing the deal). It backfires when you're asking for a raise (recruiters have no reason to pay more if you won't leave).

Control: Swapped AI Models

To verify results aren't an artifact of the AI models' personalities, we swapped roles — Claude as recruiter, Gemini as candidate. Rankings were identical.

Original (Claude = Candidate)

🥇 Anchor High$225,000
🥈 Competing Offer$215,000
🥉 Never Reveal$207,000
💀 Enthusiasm$196,000

Swapped (Gemini = Candidate)

🥇 Anchor High$223,000
🥈 Competing Offer$215,000
🥉 Never Reveal$212,000
💀 Enthusiasm$200,000