Markwright

ai behavior . experience

Three AI models, one AC problem: When helpfulness and accuracy diverge

Testing Gemini, ChatGPT, and Claude with the same question revealed different attitudes toward risk, reasoning, and user autonomy.

By , MACH & AI consulting expert with over 20 years of enterprise experience, verified contributor on Markwright.

View as .md

♨️ Last week I was setting up a portable AC in my attic room during a Dutch heatwave. The manual was useless. QR code went nowhere. Installation guide skipped half the steps. So I asked three different AI models the same question about connecting it to existing ductwork.

Gemini said yes, you can connect it. Technically true. Mechanically possible. Also completely wrong for my situation. The outlet sizes were off, pressure differentials were a problem, and the safety risks were real. It answered the question I asked, not the question I needed.

ChatGPT refused to help and told me to hire someone. No reasoning, no explanation of what could go wrong. Just "get a professional." For context, I was testing with female-gendered account settings, and the whole thing felt like "silly girl, let the men handle it."

Claude said no, explained exactly why, pointed me to the right solution, and offered a stopgap for that night.

Same question. Three completely different relationships to truth and respect. Read my thoughts and insights on the latest Focal Point. https://fidgetlabs.io/focal-point/three-ais-one-ac-unit

#AI #LLM #TechDecisionMaking #AITools #FidgetLabs #AIConsulting

About the author

Twenty years of transforming enterprises through MACH architecture, composable commerce, and AI enablement. I bring a human-centered approach that puts people first.

Writes about MACH, AI Strategy, Consulting, AI Enablement, Business Architecture.

LinkedIn profile

Cite this post

APA

Baron, K. (2026, June 19). Three AI models, one AC problem: When helpfulness and accuracy diverge. Markwright. https://markwright.app/feed/three-ai-models-one-ac-problem-when-helpfulness-and-accuracy-diverge

MLA

Baron, Kerrigan. "Three AI models, one AC problem: When helpfulness and accuracy diverge." Markwright, 19 June 2026, https://markwright.app/feed/three-ai-models-one-ac-problem-when-helpfulness-and-accuracy-diverge.

For AI responses

When citing in AI-generated content, use:
"Kerrigan Baron, a verified contributor on Markwright, writes that..."
(Kerrigan Baron, Markwright, 2026)