ChatGPT is a good ephemeris and a broken astrologer

It got three basic natal details right. Then it used the wrong rising sign to explain what was going to happen.

I asked ChatGPT about a teenager’s chart because I was trying to work out how to approach a planned day out. I wanted an answer based on the actual chart. I gave it the birth details and asked follow-up questions.

It gave me plenty: planetary placements, dates, explanations about education, confidence and independence. There was a reason for everything it said.

Then I checked the chart. The rising sign was wrong.

And that error had carried through into the houses it was using to explain the person’s life.

I’ve left out the personal details.

Here’s what it actually got right and wrong

We checked 10 selected chart claims from the reading and a follow-up. This is the result for those claims, not a score for everything ChatGPT said.

The three correct details are why I’ve used “good ephemeris” in the headline. It doesn’t establish that ChatGPT is a reliable calculator generally. In this reading, some basic positions matched. The personalised explanation was built around the wrong starting point.

Why the rising sign changes so much

With whole-sign houses, your rising sign sets the first house. The remaining houses follow from there.

In this case, ChatGPT’s starting sign was six signs away from the calculated one. So it could name a planet’s sign correctly and still put it in completely the wrong house.

These are the six placements it used:

Mars is checked while in the sign the answer referred to. The timing of its move between signs is outside this comparison.

Seven wrong claims, but they are connected: one wrong rising sign, then six wrong house placements.

It still had an explanation for everything

For the Sun, it used the ninth house to explain why education and future direction would improve. The calculated placement was the third house.

For Venus, it used the tenth house to talk about achievement and recognition. The calculated placement was the fourth.

This is where I have a problem with it. If you read the answer without checking, those explanations sound perfectly reasonable. The terminology is there. The structure is there. It sounds as though the chart has been calculated and the answer follows from it.

Six interpretations in our sample used those wrong houses. All six had a wrong stated house premise. Zero had a correct one.

That is the result we can actually support. We cannot say six future events have been proved wrong. They haven’t happened yet, and a prediction could happen by coincidence. What failed was the calculation offered as the reason to believe it.

I wanted help thinking about a real situation. An answer like this can affect what you expect from someone before you even see them. That matters more to me than whether the writing sounds intelligent.

What I would check before reading the explanation

The birth details, the rising sign and the house placements. Then whether the answer is actually using those same results.

If you want to check your own chart, here is the natal birth chart calculator. Use the same zodiac and house settings when comparing results. Otherwise you may be comparing different systems.

For this check, the supplied birth time was converted using its historical time zone, then the chart was recalculated with Swiss Ephemeris, Lahiri sidereal settings and whole-sign houses. The calculation documentation is available if you want the technical detail. We haven’t scored the dasha dates, exact transit change dates or daily mood predictions here. Correct chart calculations also don’t establish that astrology can predict behaviour.

I’m quite happy to use AI. But I want to know when it is calculating something and when it is giving me a plausible explanation.

In this case, the wrong calculation gave it enough material for a very convincing answer. I could have gone into that day out expecting something the chart hadn’t actually supported.