Sona

Journal/Choosing an app

Is AI astrology accurate? What it gets right and wrong

Is AI astrology accurate? Only when the chart is calculated first. Why chatbots get placements wrong, and how to check any app against a free calculator.

A translucent chart wheel in lavender mist with a fine gold measuring arc laid across it, one planet point sitting exactly on the line.

Is AI astrology accurate? It depends on which part you mean. The chart can be exact, if software calculates the planet positions before the AI writes a word. If a chatbot works the positions out from memory, the chart is often wrong, and the reading built on it is wrong with it. The interpretation is a third thing: no app, AI or human, can be accurate there in a scientific sense.

So the useful question is not "is the AI good". It is "where did the positions come from". This page explains how to find out, in about five minutes, for any app or chatbot.

The two halves of a reading

A birth chart is a map of where the Sun, the Moon, and the planets were at the minute you were born, seen from the place you were born. That half is astronomy. It has one right answer, and you can check it.

The other half is interpretation: what a Moon in Scorpio or a Saturn in the seventh house tends to describe. That half is a language. Astrologers disagree inside it, and nobody can prove a sentence of it in a lab.

PartWhat it isCan you check it?
Planet positionsAstronomy, from an ephemerisYes, to the degree
Rising sign and housesAstronomy plus a choice of house systemYes, if you know the system
InterpretationA pattern languageNo. Judge it by how well it helps you
PredictionA claim about the futureNo app can deliver this

Most arguments about AI astrology mix these rows together. Keep them apart and the answer becomes clear.

Why a chatbot can get your chart wrong

An ephemeris is a table of planet positions over time. Chart software looks your birth minute up in it, corrects for your time zone, and does the geometry for your location. It is plain calculation, and computers do it well.

A language model does something different. It predicts likely text. Ask a general chatbot for your chart and, unless it runs a calculation tool, it produces positions that sound right. Some of them are right. The pattern of errors is easy to explain:

  • Slow planets are easy. Pluto stays in one sign for years, and Saturn for about two and a half. A birth year is almost enough.
  • The Sun is mostly easy. It moves about one degree a day, so the date gives the sign, except on the days it changes sign.
  • The Moon is hard. It changes sign about every two and a half days. A guess is often one sign off.
  • The rising sign is hardest. It changes about every two hours and depends on the place and the exact local time, with daylight saving and historical time zones included. Without a calculation, it is a coin toss with twelve sides.

The houses follow the rising sign, so one wrong rising sign moves every planet to the wrong house. The chatbot then writes three confident paragraphs about a seventh-house Venus that you do not have.

That is the real risk. The writing does not get worse when the chart is wrong. It sounds exactly as sure of itself.

Can ChatGPT read my birth chart?

It can read one. The weak point is when it has to make one.

If you give a general chatbot a list of placements that a chart calculator produced, it can explain them, compare them, and answer questions about them. That is text work, and language models are good at it.

If you give it only a birth date, time, and place, the result depends on what happens next. Some assistants can run code or call a tool, and a chart calculated that way can be correct. Others answer from memory. From the outside the two answers look the same, and the assistant does not always tell you which one you got.

Two habits make this safe:

  1. Calculate first. Get your chart from a calculator, then paste the placements into the chat. The model then interprets and does not guess.
  2. Ask where the positions came from. If the answer is not "I calculated them with an ephemeris", treat the placements as unverified until you check them.

A general chatbot also starts from zero in each new conversation, unless you paste the chart again. It reads a chart. It does not know you.

The pattern that works: calculate first, then interpret

The reliable design is the same in every app that gets this right. The software calculates the chart. The AI receives the calculated positions and is asked only to interpret them. It never has to remember where Mars was.

Many apps use the Swiss Ephemeris for the first step. Astrodienst, the company behind astro.com, publishes it. Its documentation (checked November 22, 2026) says it is based on NASA Jet Propulsion Laboratory data, currently the DE441 ephemeris, and that its compressed files reproduce that data to 0.001 arcseconds. That is far finer than any reading needs.

Sona works this way. Its server calculates your chart with the Swiss Ephemeris, through the open-source Kerykeion library, from your birth date, time, and place. The AI companion then reads the calculated positions. It does not work out the chart itself. Houses are Placidus by default, and the iPhone app offers Whole Sign in settings. The companion also keeps memory notes across conversations, so it can connect this week's question to one from last month.

Sona is not the only app with a correct chart. As we note in our comparison of AI astrology apps, the astronomy is the solved part. Dedicated chart apps calculate, and Co-Star states that it uses NASA Jet Propulsion Laboratory data. The difference between apps is what happens after the calculation: who or what writes the words, and how much they know about you.

How to check any app against a free calculator

You do not have to trust an app, including ours. This check takes five minutes.

  1. Find your placements in the app. You need the sign and degree of the Sun, the Moon, and the rising sign (also named the Ascendant).
  2. Open a free calculator. The "Chart Drawing, Ascendant" tool on astro.com is a common reference. Enter the same birth date, time, and city.
  3. Compare the Sun, the Moon, and the Ascendant. The signs must match, and the degrees must be within a degree of each other.
  4. If the Moon or the Ascendant is far off, check the inputs. Look at the birth city, AM and PM, and the time zone. A one-hour daylight saving error can move the rising sign.
  5. If the planets match but the houses differ, check the house system. This is a setting, not an error.

Step 5 confuses many people. Sona and TimePassages default to Placidus, Co-Star defaults to Porphyry, and CHANI and The Pattern use Whole Sign. The same planet can sit in a different house in each, most often when it is near the edge of a house. The houses guide explains what the twelve houses describe, and the birth chart walkthrough shows where each number is on the wheel.

If an app or a chatbot fails step 3 with correct inputs, do not use its readings. Everything after that point describes a different person.

Why your birth time changes the answer

No calculation can supply data that you did not enter. Without a birth time, two things happen.

The Moon can be in one of two signs. It crosses a sign border on some days, and without the hour nobody can say which side you were born on.

The houses cannot be calculated at all. They come from the rising sign, and the rising sign comes from the time. A tool that shows you houses when you entered no time made the time up, usually noon.

An honest tool tells you what is missing. A birth certificate usually has the time. If yours does not, a reading from signs and aspects (the angles between planets) is still a real reading. It is a shorter one.

What "accurate" cannot mean in astrology

A correct chart does not make a reading true. It makes the reading about the right sky.

Interpretation is not measurement. When a reading says a Mars-Saturn square tends to show up as frustration with delay, that is a description in a shared vocabulary. It is not a finding. Its worth is practical: does the description help you notice something about how you act? Sometimes it does. Sometimes it reads like a stranger, and that result is also information. We wrote more on this in astrology for people who don't believe in it.

There are three limits to remember with AI in particular.

Fluent is not the same as right. A model writes with the same confidence about a correct chart and a wrong one. Check the chart, then read.

Agreeable is not the same as perceptive. Some chat tools bend toward what you appear to want. Disagree with one line and see what happens. A good reader, human or AI, can say which placement the line came from.

No app predicts the future. Sona does not, and it says so. Treat any tool that promises dates for a breakup or a job offer with caution. A transit can describe the weather of a month. You decide where to walk.

A working astrologer still brings something that software does not: judgment about which of the forty things in your chart matters this year. If you want that, book a person. If something heavy is going on, talk to a friend or a therapist first. A chart is the wrong tool for that job.

A five-minute test for tonight

Ask the app or the chatbot you use for three numbers: the degree of your Sun, your Moon, and your Ascendant. Write them down. Then enter your birth data at astro.com and compare.

If the numbers match, you know that the readings are about your sky, and you can judge the words on their merits. If they do not match, you learned something more useful than any horoscope: where the wrong answers came from.

Statements about Sona describe the app in November 2026. The Swiss Ephemeris documentation and astro.com were checked on November 22, 2026. House-system defaults for other apps come from each company's own listing or help centre, checked between August and November 2026.

Common questions

Is AI astrology accurate?

The chart can be accurate if software calculates the planet positions from an ephemeris before the AI writes anything. A chatbot that recalls positions from memory often gets placements wrong, most often the Moon and the rising sign. The interpretation is a different matter: it is a way to describe patterns, not a scientific measurement.

Can ChatGPT read my birth chart?

A general chatbot can interpret a chart that you give it. It is less reliable when it has to work out the chart itself, because a language model predicts text and does not look up planet positions unless it runs a calculation tool. Calculate the chart with a chart calculator first, then paste the placements in.

Why did an AI give me the wrong moon sign or rising sign?

The Moon changes sign about every two and a half days, and the rising sign changes about every two hours. Both need an exact calculation from your birth time, place, and time zone. A model that estimates them without a calculation tends to miss, while slow planets are easier to get right.

How do I check if an astrology app calculated my chart correctly?

Enter the same birth date, time, and place into a free calculator such as the chart drawing tool on astro.com. Compare the Sun, Moon, and rising sign first. If the planets match and only the houses differ, the two tools probably use different house systems.

How does Sona calculate a birth chart?

Sona's server calculates the chart with the Swiss Ephemeris, through the open-source Kerykeion library, from your birth date, time, and place. The AI companion then interprets those calculated positions. Sona uses Placidus houses by default, and the iPhone app offers Whole Sign in settings.