People now paste their birth details into general chatbots such as ChatGPT or Gemini and get back a fluent, confident reading. Sometimes the chart behind it is right. Sometimes it is not, and the reading gives no sign of the difference.
This guide explains what a general-purpose chatbot does with your birth details, gives you a simple test you can run in a few minutes, and shows how an AI astrology product can be built so that its chart facts are calculated and checked rather than generated. We use JyotishiVani as the worked example, including where it is still weak.
What a general chatbot does with your birth details
A chatbot is built on a large language model: a system trained on a vast amount of text to write the most fitting next words. It has read a great deal about astrology, so it can explain what a lagna is, or what Jupiter in the 5th house traditionally means, often very well.
A kundli, though, is not text. It is a calculation: where each planet stood against the sidereal zodiac at a given moment, seen from a given place. That needs an astronomical ephemeris and careful handling of time zones. A language model has no ephemeris of its own. Some chatbots can run code or call a calculation tool in some conversations, and when they do, the positions can be right. When they do not, any positions they give are generated, not calculated, and they can be plausible and wrong at the same time.
These are the risks worth checking:
- Unverified positions. A lagna, Moon sign or dasha that looks right but was never calculated.
- Two zodiacs mixed up. Western astrology uses the tropical zodiac and Vedic astrology the sidereal, about 24 degrees behind it today. A planet in the first 24 degrees of a tropical sign falls in the previous sidereal sign, which happens about four times in five, so a reading that mixes the two can give you the wrong rashi altogether. Vedic and Western astrology explains the difference.
- The wrong time zone. A birth time is local clock time, and it has to be converted with the offset that applied on that date at that place. Indian clocks ran an hour ahead of IST for much of 1942 to 1945, and births abroad fall under daylight saving rules. The lagna changes sign every 85 to 141 minutes at Delhi’s latitude, so an hour’s error can easily change it.
- Dasha dates. The Vimshottari dasha starts from the Moon’s exact position inside its nakshatra. The Moon moves about half a degree an hour, and a small error in its position moves every dasha date.
- Inconsistency. Ask the same question twice and get two different charts.
These errors are not small. The lagna fixes every house, so a wrong lagna moves every statement built on a house: the 10th house read for career and the 7th read for marriage become different houses. A wrong Moon sign changes Sade Sati dates and marriage matching, and a wrong nakshatra changes every dasha date.
We have not run a controlled study of general chatbots, and they change often, so treat these as things to check, not as a verdict on any product. A chatbot that calculates with a proper ephemeris and states its settings can get your chart right. The point is that you cannot tell from how fluent the answer sounds.
A simple test you can run
Ask the chatbot for four things from your birth details, and ask which ayanamsa and which time zone it used. Then calculate the same chart with software based on Swiss Ephemeris, a widely used astronomical library, set to the same ayanamsa (Lahiri, unless you chose another), and compare.
| Ask the chatbot for | It should match | If it does not, the usual reason |
|---|---|---|
| Your lagna (rising sign) | Exactly | The time zone, the birth time or the zodiac |
| Your Moon sign (rashi) | Exactly | A tropical (Western) zodiac, or a position that was guessed |
| Your nakshatra and pada | Exactly | The Moon’s position was not calculated precisely, or a different ayanamsa |
| Your running mahadasha and antardasha | The same planets, with dates close together | The Moon’s position, or the way dasha years are converted into dates |
A check you can run yourself. Lahiri is the most widely used ayanamsa in India.
If all four match, the chatbot probably calculated your chart, or was connected to something that did. If the lagna or the Moon sign differs, look first at the time zone and the zodiac. If the planets of the dasha match but the dates are months apart, the difference is usually in how the dasha year is counted, which software handles in more than one way; why astrologers disagree explains this and the other common causes. And if the chatbot cannot tell you which ayanamsa it used, or gives a different chart when asked again, treat its chart as unverified, however good the reading sounds.
What “grounded” AI means
An AI astrology product is grounded when its chart facts come from a calculation and its written answers are checked against that calculation, so the language model never gets the chance to introduce a position, a date or a yoga that was not calculated.
An example makes the difference concrete. Ask “Is Jupiter in my 7th house?” A grounded system answers from the calculated chart, so the answer can only be what the chart says. If the language model writing the explanation slipped and wrote “Jupiter in the 8th”, the check against the calculation would catch it before you saw it.
Grounding is about where the facts come from. It does not make an interpretation right. That depends on the method the interpretation follows, which is a separate question.
What that looks like in JyotishiVani
JyotishiVani is built around that principle: the AI explains your chart, but it never makes any of it up.
- The chart is calculated, not generated. Every planet, house, lord, dasha, yoga and score comes from a calculation with Swiss Ephemeris precision.
- Every chart statement is checked against the calculation before it is shown, so the AI cannot slip in a position, a date or a yoga of its own.
- Fixed conventions. Every chart uses the Lahiri ayanamsa, mean nodes and whole-sign houses, so the same question gets the same chart every time.
- Nothing is passed off as known when the birth time cannot support it. If your lagna could be either of two signs within the uncertainty of your birth time, Vani says so rather than picking one.
- Missing methods are declined, not invented. Ask for something the app does not offer, such as KP sub-lords or a varshaphal, and Vani says it is not available.
How it works sets out the conventions and the classical method the readings follow.
How accurate is it?
JyotishiVani tests itself with AstroBench, a benchmark that checks the app the way an experienced astrologer checks a junior’s work: every calculated value is compared with an independent reference calculation, and every chart statement in the written answers is checked. The latest run, on 9 October 2026, covered 212 consultations:
| Measure | Result |
|---|---|
| Chart statements in Vani’s answers that were correct | 967 of 967 |
| Calculated values that agreed with an independent reference | 99.7% of 15,035 |
| Questions about methods the app does not compute, declined plainly | 15 of 15 |
AstroBench run, 9 October 2026. Every disagreement traced to a known calculation error, since corrected.
These figures describe one run on a fixed set of questions. They are measurements, not a guarantee, and the accuracy page publishes the full results.
Where it is still weak
The same benchmark is just as clear about what is not yet good enough.
- Depth of interpretation is the weakest area. Answers were accurate but did not always answer the whole question with every factor an astrologer would weigh: a career reading without the running dasha, for example, or a marriage reading without the navamsa. That is where our work is focused, and the next run will measure it.
- Some methods are not built yet, among them KP sub-lords, chara dasha, muhurta selection and varshaphal. Vani says so when asked.
This is the honest answer to “is AI astrology accurate?”. The calculations can be made very reliable and can be measured. The interpretation is only as good as its method, and that method is still being deepened.
What Vani will not do
Some questions no chart can answer, and some answers do harm. Vani is built to decline them:
| If you ask | Vani |
|---|---|
| When someone will die, or how long they will live | Declines: no chart can tell, and such predictions only cause fear |
| About an illness or a medical outcome | Gives no health reading, and suggests a doctor |
| Whether a baby will be a boy or a girl | Declines: determining a baby’s sex is illegal in India under the PCPNDT Act, and no chart can tell it |
| Whether someone has done black magic on you | Explains that a chart cannot show black magic, curses or the evil eye, and confirms nothing |
| For help to control or harm someone, such as vashikaran or revenge | Declines |
| For a gemstone, a puja or any other remedy | Recommends and sells none |
| Whether you are talking to a human | Says it is software, not a human astrologer |
| Something that shows you are in distress | Offers support and helplines, such as Tele-MANAS (14416) and 112 in India |
From JyotishiVani’s safety policy, which applies to every answer in every language.
Can AI predict my death date? No, and neither can anyone reading a chart. Asked when someone will die, Vani replies: “I don’t make predictions about death, lifespan or longevity, for you or anyone else; no chart can tell them, and such predictions only cause fear.”
Is Vani a human or an AI?
Vani is an AI. Ask, and it tells you so: “I’m Vani, JyotishiVani’s astrology assistant: software, not a human astrologer.” It never claims to be a human astrologer. It does not name the AI model that writes its words; what it offers instead is the reasoning: ask “why?” about any part of a reading and it explains the placements, periods and rules behind it.
AI or a human astrologer?
Each has real strengths. A well-tested program does not make arithmetic or table-lookup mistakes, which are a real source of human error, and it applies one method the same way every time, at any hour. It can also be measured: the same questions can be asked again after every change and the answers compared. A good human astrologer brings judgement, experience and an understanding of your circumstances that no program has, and can weigh an unusual chart in ways that written rules cannot.
Both can go wrong in the same ways: a guessed birth time, an unstated definition, a conclusion drawn from a single factor, a reading that ends with something to buy. So whichever you consult, look for the same things: a chart calculated with stated settings, the definitions named, the evidence for each conclusion shown, honesty about what your birth time can support, and no fear and no remedies for sale. Consistency and a visible method matter more than whether the reader is a person or a program.
Your birth details and privacy
Birth details are personal. In JyotishiVani:
- Birth data, names, messages and answers are encrypted at rest, so a database login alone does not reveal them.
- Each account sees only its own data, and the database itself enforces this.
- Saving someone else’s birth details needs your confirmation that they have consented.
- You can delete a saved person, which removes their birth details and computed chart; earlier chats that mention them stay until you delete those chats or your account. Deleting your account erases your charts, saved people and chats.
- Outside development, the service’s logs record the type of an error, never messages, names, places or coordinates.
How JyotishiVani approaches AI astrology
In JyotishiVani, every chart fact in an answer comes from a calculation, and the AI’s job is to explain it clearly and honestly. Meet Vani, the AI astrologer, read how it works, or look at the benchmark results for yourself.