When parents ask AI for a tuition centre, the answer barely follows Google reviews
Last updated 7 November 2026
A parent in Tampines types a question into an assistant and gets back a few names. Those names are not a count of what other parents think.
We asked ChatGPT, Gemini and Perplexity 24 questions of that kind, three times each on each assistant: 216 answers. Between them they named 132 tuition centres. The Learning Lab came up most, in 62 answers (29%). On Google Maps it ranks eighth of the 15 by review count. Mind Stretcher has the most Google reviews of the 15, 4,176 across 26 listings, and came up in 50 answers: second of the 15.
The question a parent types
“Where can I find a reputable tuition centre in Tampines, Singapore for my P6 child preparing for the PSLE?”
That is one of the 24. The others asked about PSLE maths, Primary 5 English, secondary science, O-level A-Math, JC H2 Chemistry, Chinese for primary children and an affordable small group for Secondary 2 maths, each in Bishan, Tampines and Jurong. The words were fixed before the run, and a centre counts only when its name is in the answer.
Named by AI, reviewed by parents
The stars do not separate these centres. Eleven of the 15 hold 4.8 or more on Google, so stars alone barely tell them apart. The review counts do separate them, and their order is not the assistants’ order. Ranked side by side, review counts and AI mentions show a weak link (rank correlation 0.44, where 1 is the same order); stars and mentions show no link we could measure (0.06).
Age does not sort them either. We found a reliable opening year for thirteen of the 15; the oldest, Mavis Tutorial Centre, has been teaching since 1986. Line the years up against the answers and you get no link we could measure (rank correlation 0.12).
Each assistant keeps its own list
Nine of the 15 were named by all three assistants. PMC Academy came up 15 times, every one on Gemini. Which assistant a parent opens changes which centres they hear about, before reviews come into it at all.
What it means
For parents. Read an assistant’s list as a set of names to look up, not a verdict. No official body ranks tuition centres in Singapore. The Ministry of Education registers them and says so plainly: “MOE’s registration does not in any way represent an endorsement or accreditation of the quality of the courses offered” (MOE, list of private schools). The checks parents already use still do the work: reviews, a trial lesson, a talk with the teacher.
For centre owners and headmasters. Being reviewed and being named are two separate things, and an assistant does not add up your Google reviews for you. It repeats what it finds written about a centre, on the centre’s own site and elsewhere. One run cannot say which source weighs most. We also ran our free scan on the 15 websites on 6 November 2026: eleven of 15 score under 80 for search readability, and the centre named most is not the one with the cleanest site. A readable site does not put a centre in the answer. It decides whether the assistant can read what the centre says about itself once it looks.
How we did it
24 questions (8 topics in 3 areas), each asked three times on each of three assistants through their APIs, with no location and no history: 216 answers. Models: openai/gpt-6-luna, google/gemini-3.5-flash, perplexity/sonar. Names were kept only when they appear in the stored answer, and counted by code; branch names were merged into one business by hand and checked. Google Maps was read on 7 November 2026: for each centre, every listing in the results for its name and for its name plus “tuition centre”, scrolled to the end, whose website link is the centre’s own domain. Maps shows a different part of a large chain from one read to the next, so we read it twice and counted a listing seen in either read once; where a branch on the centre’s own locations page was still missing, we searched for it by place. Review counts are summed across those listings. Opening years come from the centre’s own site where it gives one, otherwise from an independent page; where the sources disagree or we could only read a search snippet, we leave the year blank. The rank correlation is Spearman’s, over the 15 centres.
What this does not show
One run in November 2026. Assistants change their answers, and the apps a parent uses, signed in and with a location, may answer differently from the API. Fifteen centres is a small set, so the correlations are a reading, not a law. A branch whose Maps listing does not link to the centre’s site is not counted, and ratings move every day. A centre named more often is not a better centre, and one named less is not a worse one; this is why we print no star rating or score per centre. If a figure about your centre is wrong, write to hello@aiqsight.com and we will correct it.
If you run one of these centres
You may show this badge on your own site:
It states a fact about one run, and nothing about next month. To see what an assistant can and cannot read on your site, run the free scan.