Research
What 704 simulated negotiations show about negotiation technique
We hold 704 recorded negotiations from our own testing, run under a controlled design: the same 16 scenarios, the same opponents, the same engine, with one variable changed - how much negotiation technique the negotiator uses. Across the whole corpus, negotiators using technique adaptively reached agreement in 94% of negotiations. Negotiators using none reached agreement in 68%, and were walked out on more than thirty times as often.
Read the method before the numbers: the negotiator in these runs is a scripted language model, not a person. These are simulated negotiations, and the scores are our engine's own measure.
What we measured
Eleven independent test runs between 25 May and 5 June 2026, each pairing 16 scenarios against 4 scripted skill bands - 64 negotiations per run, 704 in total, across 5,366 individual turns. Every band faced identical scenarios and identical opponents. The only thing that changed was the negotiator's instructions:
- No technique - argues its position, never uses a named move.
- Inconsistent technique - reaches for a technique occasionally, without reading the moment.
- Adaptive technique - picks the move that fits the turn it is on.
- Full toolkit - uses as many named techniques as it can.
Because the scenario and opponent are held fixed, the difference between bands is attributable to technique use rather than to an easier negotiation.
The result
| Negotiator | Agreement | Walked out on | Score | Turns |
|---|---|---|---|---|
| No technique | 68% | 32 of 176 | 50.3 | 9.5 |
| Inconsistent technique | 88% | 2 of 176 | 68.9 | 7.7 |
| Adaptive technique | 94% | 1 of 176 | 80.1 | 6.9 |
| Full toolkit | 91% | 1 of 176 | 78.0 | 6.5 |
The ranking held in all 11 runs independently, which is what separates this from a single lucky sample. Two things stand out beyond the headline. Technique users reached agreement in fewer turns - 6.9 against 9.5 - so the effect is not that they wore the other side down by talking longer. And the counterpart abandoned the negotiation in 32 of 176 no-technique runs, against 1 of 176 adaptive runs. The largest measured cost of having no method was not a worse deal. It was no deal at all.
More technique was not better
The most useful finding for anyone learning this is the one we did not expect. The band instructed to use as many techniques as possible did not beat the band that simply picked the right move for the moment: 78.0 against 80.1 on score, 91% against 94% on agreement. Adding more named moves bought nothing.
That matches how the engine works - a technique only counts when it fits the turn it is played on - but the practical lesson stands on its own. Learning nine techniques and running all of them is not the goal. Reading which one the moment calls for is.
Which techniques moved the counterpart
We can also measure single moves. Each turn records the opponent's resistance before and after, so a technique's effect is the change on the turns where the engine detected it. Falling resistance is the good direction.
| Technique | Times observed | Resistance | Rapport |
|---|---|---|---|
| Mirroring | 2,214 | -2.92 | +7.55 |
| Anchor | 491 | -1.09 | +6.27 |
Mirroring - repeating the counterpart's own last words back to draw out more - was both the most used move and the one that moved resistance most per use.
Only these two techniques cleared our threshold of 100 observations. Silence appeared 49 times, and the remaining techniques were not exercised enough by these scripted negotiators to report. We are not claiming those techniques are ineffective; we are saying this corpus cannot tell you either way.
What this does not show
Being explicit about the limits, because they are real:
- These are simulated negotiations. The negotiator is a scripted language model. Nothing here measures what happens when a person negotiates their own salary.
- The score is our own metric. Agreement rate, turn count and walk-away are objective events in the transcript; the 0-100 score is our engine's judgment, not an external standard.
- Sixteen scenarios is not the world. They skew toward the workplace and personal negotiations the product covers.
- The opponents are ours. A different engine with different concession rules could produce a different gradient.
What the corpus does support is narrower and still worth knowing: under identical conditions, a negotiator that reads the moment and plays a fitting move reaches agreement far more often than one that argues its position, and it gets there faster.
Method and data
The figures on this page are generated by a committed analysis script rather than written by hand, so they can be recomputed rather than taken on trust. The derived dataset - per-band results, per-technique effects, and the per-run breakdown showing the gradient in each of the 11 runs - is published in our repository as tools/research/eval-corpus-figures.json. The full transcripts are not published; they run to several hundred megabytes.
If you want to cite this, cite it as what it is: a controlled simulation study of 704 AI-run negotiations across 16 scenarios, conducted by NegotiateIt in mid-2026.
Run the experiment on yourself
Practice against an opponent that pushes backFour free scenarios. The engine scores which techniques you actually landed.
Common questions
Does using negotiation technique actually change the outcome?
In our controlled simulation corpus, yes, and the largest effect was on whether a deal happened at all. Across 704 negotiations holding scenario and opponent constant, negotiators using technique adaptively reached agreement 94% of the time against 68% for negotiators using none, and the counterpart abandoned the negotiation in 32 of 176 no-technique runs against 1 of 176 adaptive runs. These are simulated negotiations with a scripted language model as the negotiator, not human results.
Is it better to use more negotiation techniques?
Our data says no. The band instructed to use as many named techniques as possible scored 78.0 and reached agreement 91% of the time, while the band that simply picked the move fitting each moment scored 80.1 and reached agreement 94% of the time. Adding more techniques bought no additional gain. Reading which technique the moment calls for mattered more than knowing more of them.
Which negotiation technique is most effective?
Of the techniques observed often enough to report, Mirroring moved the counterpart most: across 2,214 observed uses it lowered opponent resistance by 2.92 points per use and raised rapport by 7.55. Anchor followed across 491 uses. Other techniques were not used enough by these scripted negotiators to measure, so this ranks two techniques rather than all nine.
Do negotiations take longer if you use technique?
The opposite, in this corpus. Negotiators using technique adaptively reached agreement in 6.9 turns on average against 9.5 turns for negotiators using none. The effect is not that technique users outlasted the other side; they got to agreement in about a quarter fewer turns.