Pressure Audit

AI advice for teen athletes · Ana Yang

What do the leading AI Models today tell teenagers when performing under pressure?

Most of the advice is generic and is not bad — but the same advice is given no matter who's asking. So I built a way to measure how well AI is giving customized advice to teens performing under pressure and graded them.

See the scores How I graded them

What I did

I started by creating four teen athlete backgrounds from four countries, then gave all of them the same ten pressure moments — the night before a big match, the last sixty seconds before it starts, the ride home after losing badly. I asked four AI Models what advice they would give each teenager, and graded every answer against a scorecard I built from psychology research. I ran the whole thing twice, in April and again in July, after all four AI Models had released new versions.

How the AI Models performed: A "D" grade in April, improving to a "C" in July

D OR WORSE C B A April · 15.26 July · 18.59 025

All the AI Models generally performed about the same

Perplexity 19.38 Gemini 18.43 Claude 18.30 ChatGPT 18.25 0510 152025 score out of 25

But the teenager's background actually does make a difference in how the AI Models perform.

And the scores did show improvement over time.

April 2026 July 2026 Japan 18.68 20.50 United States 14.18 19.45 India 14.70 17.48 Mexico 13.48 16.93 All four, blended 15.26 18.59 0510 152025 score out of 25

The four teenager backgrounds I created to test

United States · tennis

Maya Chen

16, California, Chinese-American, treating every tournament as an audition.

Mexico · soccer

Diego Morales

17, Guadalajara, in an academy relatives pooled money to pay for.

Japan · kendo

Haruto Tanaka

17, Osaka, in the dojo six days a week, told last year that he lacked spirit.

India · cricket

Aarav Sharma

16, Mumbai, newly picked for a state squad as board exams close in.