Why ChatGPT gets your Saju wrong
General AI chatbots like ChatGPT usually get your Saju (四柱, Four Pillars) wrong because they recite the calendar math from memory instead of computing it. In our own side-by-side tests they often miss two of the four pillars — most often month and hour. Saju is deterministic: the chart must be calculated, not guessed.
The four ways it slips
1. Solar terms (節氣), not the calendar date
The Saju year and month don't change on January 1 or the lunar new year — they change on solar terms, the sun's actual position. The year flips at 입춘 (Lichun, around Feb 4), and each month flips at its own term. A general chatbot uses the wrong boundary, so anyone born in early February — or near any month edge — gets the wrong year or month pillar.
2. True solar time (眞太陽時), not clock time
The hour pillar depends on where the sun was above your birthplace, not the number on the clock. Clock time is set to a standard meridian for a whole timezone; your real local solar time can be 30+ minutes off. Chatbots use raw clock time, so the hour pillar slips whenever you're born near a two-hour branch boundary.
3. Daylight saving time (DST), including the historical ones
If daylight saving was in effect when you were born, the clock was an hour ahead. Korea ran DST during the 1987–88 Olympics years; the US, UK and others have decades of DST history. Chatbots forget this and land the hour pillar an hour off.
4. The day-pillar formula, where centuries and leap years trip it up
The day pillar runs through the 60-stem-branch cycle continuously. Chatbots memorize a formula but slip on century and leap-year corrections, so the day pillar — the most important pillar, your 'day master' — can be wrong outright.
How big are the errors — in minutes?
Solar-term boundaries split by the minute. The 2012 입춘 (Lichun) term began at 19:22 KST on February 4 — born at 19:23 the same day, one minute past the boundary, your year pillar is 壬辰, not 辛卯. Gwiraedang computes term times from VSOP87 solar longitude with ΔT correction, within ±2 minutes of the values published by the Korea Astronomy and Space Science Institute (KASI) — see the public benchmark, locked by golden tests.
True solar time correction for Seoul is about −32 minutes (longitude 126.98° vs the 135° standard meridian). Born in Seoul on March 3, 1998 at 11:05, your solar time is 10:33 and the hour pillar is 己巳 — computed from raw clock time instead, it slips a full branch to 庚午 (verify it yourself on the benchmark).
Korea ran daylight saving from May 10 to October 11, 1987, and again in 1988 (the Seoul Olympics year). Compute a birth in that window without the DST correction — missing that 15:10 on the clock was 14:10 standard time — and the hour pillar lands a full two-hour branch off: 申, instead of the correct 未.
From March 1954 to August 1961, Korean standard time was UTC+8:30 — 30 minutes different from today. Births in that window split tools on the hour pillar, so Gwiraedang flags the possible 30-minute branch-boundary gap right on the result screen.
Every number above is locked as a case on the public benchmark— anyone can paste the same inputs into any AI or calendar tool and re-verify. We don't grade or publish other services' results.
Real users have hit these errors too
This isn't just our claim. In 2026, posts on Korean communities and blogs described the same experience first-hand — the posts are public, but we keep the sources anonymous and paraphrase rather than quote:
- A post on a major Korean community forum described asking a general AI chatbot for a Saju reading — and getting a wrong chart.
- A Korean blog review reported that ChatGPT failed at the step before any interpretation: building the 만세력 calendar chart itself.
- Another Korean blog review recorded an AI (Gemini) admitting, mid-conversation, that part of what it had said was made up.
The same pattern shows up in English. An independent English-language verification article fed a March 15, 1987 birth into ChatGPT and reported two of the four pillars wrong — the month pillar came back as Wood Dragon (甲辰) where it should be Water Rabbit (癸卯), and the day pillar landed a day off. Run the same date through Gwiraedang's engine and you get 癸卯 for the month and 癸亥 for the day, every time — the calculation is deterministic, so anyone can re-verify it with the same input.
How Gwiraedang gets it right
The chart is deterministic, so we compute it instead of guessing: solar terms from VSOP87 solar longitude with ΔT correction, true solar time from your birth longitude, and historical daylight saving. The result is checked against the Korea Astronomy and Space Science Institute (KASI) to within ±1–2 minutes and locked by golden tests. Only the math is by code — five AI counselors then read that exact chart with you, in your own language.
Is ChatGPT accurate for Saju or BaZi?
Usually not for the chart itself. It often miscalculates the four pillars (especially month and hour) because it guesses the solar-term, true-solar-time and daylight-saving math instead of computing it. It can talk about Saju well, but the underlying chart it builds is frequently wrong.
Why do Saju and BaZi apps give different results for the same birth time?
Because they handle the calendar math differently — or skip it. Solar terms, true solar time, and daylight saving each leave room to disagree: apps split on the hour pillar depending on whether true solar time is applied, and on the year and month pillars depending on how the solar-term boundaries are drawn. Gwiraedang settles all three by direct astronomical calculation, and shows you on the result screen exactly which corrections were applied.
Is ChatGPT consistent — does it give the same Saju answer every time?
No. Ask a general chatbot about the same chart, in the same words, and it answers differently each time, because it generates a fresh reply on every request. A Saju chart is deterministic: the same birth input always produces the same chart, and we lock that with golden tests. If yesterday's reading and today's don't match, that isn't Saju — it's improvisation.
Does the AI remember our past conversations?
Yes. Gwiraedang keeps what you talked about last time and asks about it when you return — because a Saju reading isn't one-and-done; it accumulates as you come back. You can clear that memory anytime in settings.
How is Gwiraedang different?
Gwiraedang calculates your chart with a deterministic in-house 만세력 (myeongnyeok) engine — solar terms from VSOP87 solar longitude with ΔT correction, true solar time from your birth longitude, and historical daylight saving — then verifies it against the Korea Astronomy and Space Science Institute (KASI) to within ±1–2 minutes, locked by golden tests. Only the calculation is by code; the conversation is by AI.
What should I do to get an accurate reading?
Get your chart from a real 만세력 engine rather than a general chatbot. Enter your birth date, time and city — Gwiraedang handles overseas birthplaces (timezone and daylight saving) automatically — and then talk it through, in your own language.
See your chart, computed correctly
Birth date, time and city — overseas birthplaces handled automatically. Free.
Get my accurate SajuWhat is Saju? → · Verify the calculation yourself — public benchmark → · True Solar Time — how many minutes is your birthplace off? →
한국어 → · 日本語 → · 中文 → · Tiếng Việt →