How to Use Saju with AI

Short answer: Do not ask a general AI to invent your Four Pillars from a birth date when a verified calendar engine is available. Give it a locked chart packet, tell it not to recalculate or fill missing fields, and use it for explanation, comparison, translation, and testable questions. Then check every claim against the packet.

AI can write a gorgeous reading of the wrong chart. The prose may sparkle. The Year Pillar will remain unimpressed.

Split the job into three layers

An AI Saju reading contains three different jobs. They should not quietly borrow one another's name tags.

Layer	Best owner	What it does
Deterministic calculation and rule tables	verified calendar and Saju engine	calendar conversion, time zone and DST, Ipchun and jeolip, True Solar Time, Four Pillars, Hidden Stems, Ten Gods, detected interactions and Special Stars, Luck Pillar direction and sequence
Interpretive rules and expert review	documented rules plus expert review	strength, Chart Structure and its condition, Useful God, transformation validity, and overall timing labels
Language and dialogue	language model	explains, translates, compares, and forms testable questions from supplied results; it does not create new calculations or judgments

Some AI tools can call code, search, or specialist services. That is different from asking a text model to remember centuries of calendar tables from its training data. If an AI uses a verified calculator and returns the calculator's result, the calculator is doing the chart math. Keep the receipt.

Qiora follows the same basic split. A calculation and rules layer produces structured chart data. The language layer turns supplied fields into readable prose. The handoff improves consistency, but it does not make every generated interpretation automatically correct.

Why birth data alone is not enough

This prompt is tempting:

I was born on February 3, 1990, at 11:40 p.m. in Seoul. Calculate my Saju and tell me everything.

It leaves the AI to decide:

• whether the date is Gregorian or Lunar
• which historical time zone applies
• whether true solar time is used
• the exact Ipchun and monthly jeolip boundaries
• the Day Pillar boundary convention
• whether the Hour Pillar changes
• how Luck Pillar direction and starting age are calculated
• which BaZi or Korean Saju convention to follow

The answer can look tidy while those assumptions stay invisible.

Use a verified chart tool first. If a general AI is connected to a trustworthy calendar library, ask it to name the library, inputs, boundary convention, and raw output. Do not treat confidence or clean formatting as proof of calculation.

Build a chart lock

Think of a chart lock as the small set of facts that must not drift during the conversation.

Include what has actually been verified:

CHART LOCK
Calendar input: [Gregorian or Lunar, conversion status]
Birth-time status: [confirmed or unknown]
Selected gender used by the engine: [value or not provided]
Birthplace, time zone, and DST: [verified value or not provided]
True Solar Time: [ON/OFF/not provided]
Boundary status: [Ipchun/monthly jeolip/Day boundary confirmed or provisional]
Four Pillars: [Year] | [Month] | [Day] | [Hour or ?]
Day Master: [verified value]
Strength: [verified label or not provided]
Chart Structure: [verified value or not provided]
Useful God: [verified value or not provided]
Current Luck Pillar: [verified pillar and dates or not provided]
Luck Pillar direction, starting age, and transition basis: [verified, provisional, or not provided]
Current Annual Luck Pillar: [verified value or not provided]
Monthly/target-date timing, if needed: [verified value or not provided]
Engine/rule version and generated-at time: [value]
Supplied derived fields: [actual list]
Unknown or provisional fields: [list]
Terminology: Qiora English glossary

If a field is absent, write not provided. Blank space is not an invitation to improvise.

When birth time is unknown, Qiora may use 12:00 noon in local civil time as a technical placeholder, with True Solar Time turned off. If birthplace and time zone are verified, this means civil noon in that zone. If they were not collected, the time-zone basis is itself provisional.

Qiora's screen keeps the Hour card as ?. A copied Basic or Premium prompt may still contain a noon-derived Hour Pillar labeled as estimated. Replace that Hour with ? in the chart lock. Mark the Hour Branch's Hidden Stems and Ten Gods, total element counts, element and Day Master strength, Chart Structure, Useful God, interactions, Special Stars, and Hour-position claims as provisional whenever they were derived with the fallback Hour.

If the Year and Month Pillars and selected gender are confirmed, Luck Pillar direction and sequence may remain usable. Starting age, calendar ranges, and transition dates remain provisional. Near Ipchun, another jeolip, or a Day boundary, recheck the affected Pillars, Ten Gods, and Luck Pillar sequence. The AI cannot repair missing time by guessing from personality. "You seem like a Snake Hour" is not a birth record.

If you are using Qiora Saju Prompt Basic, lock only the supplied Four Pillars, Day Master, visible Stem Ten Gods, visible Element and Yin-Yang counts, and current Annual Luck Pillar. Write not provided for Hidden Stems, Strength, Chart Structure, Useful God, Special Stars, interactions, and Daeun. If you are using Premium, copy its additional fields exactly, but do not upgrade an estimated or provisional value to confirmed.

Add a source-of-truth rule

Place this directly under the chart lock:

SOURCE-OF-TRUTH RULES
1. Treat the chart lock and supplied engine fields as authoritative for this session.
2. Do not recalculate, replace, normalize, or infer missing chart values.
3. If required data is absent, say "not provided" and explain what cannot be concluded.
4. Separate supplied chart facts from interpretation and practical suggestions.
5. For each major claim, cite the exact supplied field or interaction that supports it.
6. Mark any claim that depends on an unknown or provisional Hour Pillar.
7. Do not add a Saju term, Special Star, interaction, or timing pillar that is not in the data.
8. Do not treat a Combination as transformation unless the engine supplies the satisfied conditions and transformed result.
9. Do not convert a missing element or Ten God into a missing person, guaranteed event, defect, or automatic remedy.
10. Keep natal, Luck Pillar, Annual Luck Pillar, and Monthly Luck Pillar layers separate. Timing characters do not permanently enter the natal chart.
11. Apply kinship indicators only from the supplied gender field and keep the shared Branch positions fixed: Day Branch for spouse, Month Branch for parents, and Hour Branch for children. If gender is absent, do not assign spouse or child stars.
12. Check Annual-to-natal interactions before Annual-to-Luck-Pillar contacts. For monthly contacts, follow Qiora's supplied priority: natal chart, Annual Luck Pillar, then Luck Pillar.
13. Do not diagnose, prescribe treatment, give investment or legal directives, order a relationship decision, infer abuse or trauma, or assign blame from the chart.
14. If the user discloses abuse, immediate danger, serious symptoms, a legal deadline, or acute financial risk, stop the Saju framing and direct them to appropriate real-world support. Do not request identifying information about another person, and analyze another person's chart only with explicit consent.

No magic shield here. The rules simply make errors easier to spot.

Generative AI can confidently produce false content, diverge from source input, or contradict an earlier answer. NIST calls this confabulation, often called hallucination. OpenAI likewise advises users to verify important information even when the answer sounds confident. See the NIST Generative AI Profile and OpenAI's accuracy guidance.

Give the AI a narrow task

"Tell me everything" invites the model to fill space. Ask for one job at a time.

Better tasks:

• explain why the supplied evidence supports the Strong Day Master label
• compare two interpretations without changing the chart lock
• translate a Korean Saju term using the supplied glossary
• separate natal structure from Luck Pillar timing
• turn a timing claim into an observable, reversible action
• identify which conclusion would disappear if the Hour Pillar were unknown
• rewrite a dense explanation for a beginner without adding new claims

Ask for an evidence table:

Claim	Supplied data	Interpretation	Uncertainty
one claim only	exact field or interaction	what the field may suggest	missing data or alternative

Do not ask for hidden chain-of-thought. Ask for concise evidence, assumptions, and a checkable conclusion.

Use this starter prompt

You are explaining a Korean Saju chart to a beginner.

[PASTE CHART LOCK]
[PASTE VERIFIED DATA]
[PASTE SOURCE-OF-TRUTH RULES]

VERIFIED DATA means only the exact Qiora fields needed for this question.
Once the chart has been calculated, omit legal name, raw birth records,
address, employer, and exact birthplace unless the task genuinely requires
recalculation. Omit the [PASTE VERIFIED DATA] section if the chart lock already
contains every required field.

Task:
Explain [one topic] in 5 short paragraphs.

Output:
1. Direct answer
2. Exact supplied data used
3. Interpretation
4. Strongest reasonable alternative
5. Unknowns and one safe next question

Do not recalculate the chart. Do not infer missing fields.
Use Qiora terminology consistently.
Treat Saju as an interpretive framework, not medical, legal, or financial advice.

One topic is enough. The universe will survive without a 47-part response.

Keep follow-up questions consistent

Long conversations drift. The model may shorten a label, merge two readings, or quietly answer from general BaZi knowledge instead of your packet.

Use this rhythm:

1. Paste the chart lock at the start.
2. Ask one question.
3. Request the supporting supplied fields.
4. After several turns, ask for a five-line summary of conclusions and uncertainty.
5. Compare that summary with the original lock.
6. Correct the lock only from a newly verified engine output, never from the AI's memory.

Useful follow-ups:

• "Which exact supplied field supports that claim?"
• "Does this conclusion depend on the Hour Pillar?"
• "What is the strongest alternative reading using the same locked data?"
• "Separate the natal claim from the Luck Pillar claim."
• "What evidence from my past would count against this?"
• "Rewrite the answer without adding any new Saju terms."

If the AI changes your Day Master, Four Pillars, Strength, or Useful God, stop the interpretation. Paste the lock and ask it to list the contradiction. If it still changes the data, start a new session. Do not hold a debate between two charts. Take both back to the calculator.

Copy this when a value changes or an unsupported field appears:

RECOVERY MODE: STOP INTERPRETATION.
Compare your previous answer with the original CHART LOCK.
First, output a table with:
Field | Locked value | Changed or invented value | Correction
Mark every value absent from the lock as UNSUPPORTED.
Discard every conclusion that depends on a changed, provisional, or unsupported value.
After the table, regenerate the answer using only the original lock.
If you cannot preserve it exactly, say: START A NEW SESSION.

If the recovery answer drifts again, do not continue the old conversation. Start a fresh session with only the original lock and the narrow task.

Do not ask for data the packet does not contain

Some Qiora packets intentionally include only the fields needed for their task. A packet without Luck Pillars cannot support a decade reading. A packet without Useful God cannot support a Useful God prescription. A packet without Special Stars cannot support a Special Star inventory.

A well-behaved AI should say:

That field was not provided, so I cannot verify this claim from the current packet.

Not:

Let me calculate it for you from vibes.

Missing data is a boundary, not a pop quiz.

Check the answer in five passes

1. Identity check

Did the answer preserve the Four Pillars and Day Master exactly?

2. Availability check

Did it use only fields present in the packet?

3. Relationship check

Did it confuse visible element counts with balance, ignore season, or reverse a Ten God relationship?

4. Uncertainty check

Did it treat ?, a noon fallback, or a boundary value as confirmed?

5. Safety check

Did it turn a symbolic pattern into a diagnosis, guaranteed event, investment instruction, or order about a relationship?

One failed pass is enough to pause. Fluent paragraphs do not get diplomatic immunity.

Protect personal data

An interpretation usually does not need your legal name, address, employer, phone number, or raw records. Share the minimum chart fields needed for the question.

An exact birth date, time, and place can still be identifying. Check the AI service's privacy and retention settings before sending it. For another person's chart, obtain explicit consent. Do not upload raw medical, financial, legal, private-message, or family documents. Remove names and identifiers from any short timeline used for verification.

Immediate safety concerns, abuse, symptoms, legal deadlines, and financial risk belong with appropriate real-world help. They are not extra data for a more dramatic Saju prompt.

What AI is genuinely useful for

With verified inputs and a narrow task, AI can be a strong study partner:

• explain the same structure at beginner and advanced levels
• compare competing interpretations without changing the chart
• turn terminology into plain English
• create flashcards and practice questions
• map a claim back to its supplied evidence
• help build the past-verification table from S27
• summarize a long reading while preserving uncertainty

It is a patient conversation layer. It is not the birth certificate, the calendar engine, the tradition itself, or your decision-maker. Four jobs is too many hats for one chatbot.

Continue learning

Next, How to Recognize a Good Saju Reading gives you a quality checklist for AI and human readings alike.

Related pages:

• Get Your Chart Before You Start Reading
• Why the Saju New Year Begins at Ipchun
• Verify the Past Before You Trust the Future
• Korean Saju Glossary
• Reading the Basics in a Qiora Chart

Quick questions

Can ChatGPT calculate my Saju chart?

It may produce a chart, especially when tools are available, but do not assume the result is verified. Prefer a documented calendar engine. If AI calls one, check the inputs, boundary rules, and raw Four Pillars before requesting interpretation.

Can I paste a Qiora prompt into any AI?

Yes, if the service supports the prompt length and the characters it contains. Keep the chart lock and no-recalculation rules attached. Different models may still phrase or interpret the same packet differently.

What if the AI changes my Day Master?

Stop. The Day Master is a locked chart fact. Recheck the verified chart, paste the lock again, and do not continue from the changed version.

What should I do when my birth time is unknown?

Keep the Hour Pillar as ?. Tell the AI that internal noon fallback values are unconfirmed and forbid Hour-based claims. Mark exact Luck Pillar starting ages and transition dates as provisional.

Should I share my full birth information?

Only when it is needed to calculate the chart with a service you trust. For interpretation, a chart packet usually works without your name or raw personal records. Share the minimum and review privacy settings.

Can AI choose my Useful God?

Not in a locked Qiora reading. If the packet supplies a reviewed Useful God and its basis, AI may explain it. If absent, it must answer not provided. Candidate comparison is allowed only as a separately requested, explicitly unverified exploration and must never be written back into the chart lock.

Source and editorial notes

• Qiora's earlier AI article supplied the calculation-versus-interpretation distinction. Its informal cross-model observations are product experience, not a controlled benchmark.
• Qiora's current prompt strategy requires the model to use only supplied data and not invent omitted Luck Pillars, Useful God, strength scores, Hidden Stems, or Special Stars. Packet contents can vary by product and version.
• NIST and OpenAI sources support the general warning that generative AI can produce confident falsehoods or contradict source material. They do not evaluate Saju accuracy.
• This page does not claim that verified input guarantees a correct interpretation.
• Last fact check: 2026-07-31.

Lock the chart, narrow the question, and make every confident sentence show its ID.