Loading…
How to check whether an AI fortune reading used your actual chart
By Mindy Park
Compare an AI reading with your supplied chart using a practical checklist for labels, missing information and unsupported claims.
To check whether an AI fortune reading follows your actual chart, compare its factual labels with the chart you supplied before judging the personality description. A response that repeats your name or birthday can still use the wrong pillars, invent a missing hour or offer advice that is unrelated to the chart.
This check establishes consistency with your input. It cannot reveal the model’s hidden process, verify that the original calculation was correct or prove a prediction. You are looking for an explanation that you can trace back to visible information.
Keep a reference copy of the input
Save the chart you used and note the date, local time, birthplace and calculation settings. If birth time was unknown, write that beside the chart. Remove names or other details you do not need to share.
For Korean Saju, start with the displayed year, month, day and hour pillars. Our guide to reading a chart explains the layout. If you supplied a Western chart instead, keep its original placements and settings; do not translate a rising sign into a Saju Day Master.
Audit the labels before the story
| What to compare | What a mismatch can look like | What to ask next |
|---|---|---|
| Chart tradition | A Saju reading suddenly discusses natal houses | Which supplied chart feature supports that sentence? |
| Day Master | The response names a different day stem | Please copy the day stem from the reference before interpreting it. |
| Known and unknown inputs | An hour pillar appears despite unknown time | Was that calculated from an assumption? If so, label or remove it. |
| Pillar positions | A year-pillar character is described as the day pillar | Which column are you referring to? |
| Changes across the answer | A chart label changes between sections | Recheck every occurrence against the same reference. |
The Day Master explanation helps with one of the easiest labels to locate. Matching it is a useful first check, not a complete audit of the reading.
Ask for evidence one sentence at a time
Choose a concrete claim from the response: “Your chart contains this element in this position.” Ask the model to identify the exact supplied character. Compare it yourself. Then ask which part of the explanation is a traditional interpretation and which part is an observation about your actual life.
For a hypothetical example, imagine an input chart with no hour pillar. A response that calls your hour pillar especially influential has added information it was not given. The correction is to remove the unsupported interpretation or disclose a real calculation assumption. Do not fill the gap by guessing a birth time.
Why checking a reference matters
OpenAI’s 2024 SimpleQA dataset audit found 94.4% agreement when a third AI trainer independently answered a random sample of 1,000 questions. The benchmark report concerns reference-answer quality, not fortune readings or chatbot accuracy. It illustrates that even a reference set deserves checking; an authoritative-looking label is not enough.
If two charts disagree, preserve both inputs and methods before deciding what the AI should use. Asking the same chatbot to reassure you is not an independent calculation check. Agreement with a supplied chart can coexist with an error in that chart.
A prompt you can reuse
Compare this reading with the chart I supplied. First list any mismatched labels or information that was not provided. For each interpretation, identify the chart feature it refers to. Preserve unknown inputs as unknown. Distinguish traditional interpretation from an independently verifiable fact. If the chart is unreadable, ask me to transcribe it instead of guessing.
Treat the resulting audit as another answer to inspect. Check the quoted labels against your reference; do not accept the audit simply because it appears in a table.
Decide what to keep
Correct mismatches before extending the conversation. Keep explanations that clarify terminology, label unresolved points, and set aside claims that cannot be connected to the input. Our guide to what Saju cannot do sets out why consistency and predictive proof are different standards.
If you want a starting reference for the Korean system, you can generate a Saju chart and inspect its pillars. The goal is to understand what the reading is based on, while keeping your decisions in your own hands.