Can you trust ChatGPT for Saju?

A Saju chart is computed. The year and month pillars switch at the exact time of the 24 solar terms, the day pillar needs an unbroken day count since 1900, and the hour pillar depends on the late-night rule and the day stem. Conversational AI estimates this calendar arithmetic from memory, so it slips most often on the day pillar and near solar-term boundaries. When the calculation is wrong, every reading built on it — personality, compatibility, luck cycles — describes someone else.

In September 2026 we asked the same question for 36 birth date-times (random 1950–2010 plus solar-term and late-night edge cases). The reference is the SajuLens manseryeok engine, which uses solar-term data. A GPT-family model got all four pillars right in 11 of 36 cases; a Gemini-family model in 1. The year pillar was the most reliable (GPT 35/36) and the day pillar the least (GPT 14/36, Gemini 2/36). The script and conditions are public in our repository, and these numbers are refreshed whenever we re-run with a larger sample.

Flip the order. Fix the four pillars with a manseryeok engine first, paste that chart into the AI, then ask for the interpretation. The "Take to AI" button on a SajuLens result copies the pillars, hidden stems, element share, useful element and luck cycles as one text card. Calculation by the engine, reading by the AI is the safest split.

Enter your birth details below and paste the four pillars the AI gave you. The tool compares each pillar, shows which ones differ and how, and lets you open the full reading with the correct chart.