Do AI Astrology Sites Get Your Chart Right?

Why We Tested Astrology Sites After Testing Chatbots

On 13 October 2026 we asked three AI models to work out ten birth charts from the birth date, time and city alone. They named the right Sun sign every time and the right Moon sign once in 25 answers. A language model asked cold is guessing at positions it has no way to calculate.

Most AI astrology products say they do it the other way round: an ephemeris calculates the chart, and the AI only explains it. That claim can be checked too. So we took five of the same births and entered them, as an ordinary visitor would, into the free chart forms of AI astrology products and popular chart sites, then compared what each page showed with the calculated positions.

How We Tested: Five Births, Public Forms, No Accounts

The five births were picked from the earlier test for their traps. London, 22 July 1971 at 14:40, when Britain kept summer time all year and the Sun sat at 29°06′ Cancer. Sydney, 3 November 1985 at 23:05, under Australian daylight saving. São Paulo, 17 January 1993 at 04:30, during Brazilian summer time, with the Ascendant at 28°34′ Sagittarius. Toronto, 25 October 1998 at 16:00, the afternoon daylight saving ended, with Saturn at 29°59′ Aries and both the Ascendant and Midheaven within a degree of a sign change. Mexico City, 19 June 2005 at 20:45, with the Moon at 0°00′ Sagittarius and a daylight saving rule Mexico has since dropped.

For each site we used the free public form, chose the city from the site's own place list, and kept every default the site filled in, including the time zone. We created no accounts, paid nothing and stopped wherever a site asked for a sign-in, a CAPTCHA or another check. We scored the same nine placements as in the chatbot test: Sun, Moon, Mercury, Venus, Mars, Jupiter, Saturn, Ascendant and Midheaven. A placement counts as right when the page prints the right sign; the degree error is the distance from the calculated position.

Seven sites could be tested this way: natalchart.ai, astrofox.ai, openfate.ai, asksoma.ai, deeporacle.ai, astro-read.com and astrohouse.me, plus cafeastrology.com, a long-running chart site without AI, as a baseline. OpenFate allowed two charts a day without signing in and astrohouse.me three, so they have fewer charts. Six other sites could not be tested on these terms: kaucim.ai and astro-seek.com showed a browser check to our test browser, astroma.co and kundligpt.com asked for a CAPTCHA, astro.com required registration before showing the Ascendant, and aistro.io draws charts only in its app.

The reference positions came from Cosine Astro's chart engine, the same one used for the chatbot test. It looks up each city's historical time zone and daylight saving rules and calculates positions from an astronomical ephemeris.

Results by Site, 17 October 2026

Every site calculated the planets correctly. Across the seven sites, 210 of 210 planet signs were right, 206 planets were within 0.1° of the reference, and the largest planet error anywhere was 0.6°. All 30 Moon signs were right, including the Mexico City Moon at the very start of Sagittarius. Most results matched the reference to a few hundredths of a degree, closer than any page prints them.

The errors were in the Ascendant and Midheaven, and only in four charts. 26 of the 30 charts had all nine placements within one degree. The table lists each site in alphabetical order, with the baseline last.

Every Large Miss Was a Time-Zone Problem

An ephemeris needs universal time. A birth certificate gives local clock time, so a site first has to know the offset in force in that city on that date, including any daylight saving. Get it wrong by an hour and the planets barely move, but the Ascendant and Midheaven shift by about 15° or more. That is exactly what we saw.

astrofox.ai calculated four of its charts to within 0.02°. For Toronto it gave an Ascendant of Pisces 2°14′ against a true 29°21′, and a Midheaven 13.9° off. Rerunning our engine with daylight time instead of standard time reproduces its numbers almost exactly. Canadian daylight saving ended at 2 a.m. that day, so a birth at 4 p.m. was on standard time. The likely cause is today's rule, which runs daylight saving into November, applied to 1998, though we cannot see the site's code.

astro-read.com fills in an offset when you pick a city and says, right under the form, that it does not correct for historical daylight saving and that you should check the offset yourself. For London, São Paulo and Mexico City the offset it filled in was an hour off. The Ascendant moved by 10.5° to 14.5° and the Midheaven by about 16°, which put the Midheaven in the wrong sign in all three and the São Paulo Ascendant in Capricorn instead of Sagittarius. With the correct offset typed in, the same three charts came out within 0.11°. Its engine was right; the default was not.

astrohouse.me showed London's longitude as 0°08′ east instead of west. That moved its Ascendant and Midheaven by about a quarter of a degree, which changed no signs. The one sign the baseline got wrong was the Mexico City Moon, printed at 29°59′ Scorpio when the calculated Moon was six arcseconds into Sagittarius: a rounding-level difference, not a real disagreement.

Different Houses or a Different Zodiac Are Not Errors

Some differences between sites are choices, not mistakes, and we did not count them. deeporacle.ai uses whole-sign houses, where the rising sign is the whole first house; astro-read.com uses them by default. Most of the others use Placidus. The same planet can sit in a different house under the two systems, so a house that differs between sites is often the house system, not a miscalculation.

asksoma.ai draws a Vedic chart in the sidereal zodiac, which in these years is about 23.5° to 24° behind the tropical zodiac used by the other sites. Its London Sun at 5°38′ Cancer is the same point as 29°06′ Cancer in the tropical zodiac, and once we added the ayanamsa printed on its own page, every position matched to within 0.03°. The Ascendant and Midheaven do not depend on the house system at all, which is why we could compare them on every site.

How the Sites Compare With Asking a Chatbot (13 October 2026)

The gap between calculating and guessing is large. Four days earlier, three AI models asked for the same nine placements with no chart got 124 of 225 signs right, 55%. The sites in this test got 266 of 270, 98.5%. The Moon is the clearest case: right in 1 of 25 chatbot answers and in all 30 site charts. The table below repeats the chatbot results from the earlier test; they come from ten births, not five, and from the models' API, not the chat apps.

The two tests measure different things. The chatbots were asked to produce positions they could not calculate. The sites calculated them, and their errors came from the step before the calculation, turning a local birth time into universal time.

The full chatbot test

How to Check the Chart a Site Gives You

Look for the time zone or universal time the site used. openfate.ai, deeporacle.ai, astrohouse.me and astro-read.com print it on the result; on natalchart.ai, astrofox.ai and cafeastrology.com we did not find it next to the positions. If you were born in summer, in a year or country whose clock rules have since changed, or on a day the clocks changed, that line is the one to check first.

Compare your Ascendant and Midheaven on two sites. If they differ by more than a degree or so while the planets agree, one of the sites used a different birth time in universal time, usually through the time zone. If the planets themselves differ by several degrees, check whether one site is sidereal. Differences in house placements alone usually mean a different house system.

Limits of This Site Test

This is a small test: five births, seven sites and a baseline, one run each on 17 October 2026. Two sites allowed only two or three charts without signing in. We tested only what each free form showed, with each site's own place lookup, so coordinates differed slightly between sites. We did not test AI readings, paid reports, apps, or any site behind a sign-in or a check, and sites change, so a rerun could come out differently.

The reference is Cosine Astro's own engine. Where every site agrees with it to hundredths of a degree that is mutual confirmation, not independent proof, so the engine's own check matters: its positions, Ascendant, Midheaven and house cusps are compared with the Swiss Ephemeris on the chart accuracy page. Our methods, the saved result pages and the scoring script are kept on file so every number above can be traced.

Cosine Astro's chart accuracyHow the chart is calculated