In October 2026 we asked ChatGPT, Perplexity and Google's AI Overviews 21 questions an applicant or parent might ask about UK universities. Then we checked 198 of the facts they gave against the universities' own pages.
About two in three were right. One in four was wrong, out of date or stretched. The rest couldn't be checked at all.
That's good enough for a first look and not good enough for a UCAS choice. These are the checks we'd run before acting on an answer. None of them takes more than a couple of minutes.
1. Find the year next to every number
Fees, grades and league-table positions change every year, and AI answers mix the years up constantly.
The most common version we saw: a student starting in 2027 being quoted £9,790 a year for an English university. That's the 2026/27 fee. For 2027 entry it's £10,050. One answer even put the higher figure down as the "Year 2" fee, which would only be true for someone who started in 2026.
Check: if a figure in the answer has no year, assume it might be last year's. Look it up on the university's fees or course page, and make sure the page says 2027 entry (or whichever year you're applying for).
2. Click the source, and see whose website it is
Most AI answers show where they got their information. Look at the name of the site, not just the headline.
In our checks, answers that drew on a university's current official page were right 79% of the time. Answers that drew on other websites were right 44% of the time. We saw international fee bands for Manchester taken from a student-housing company, and graduate figures for Cardiff where the link actually went to Cardiff Metropolitan, a different university.
Check: is the source the university itself, UCAS, gov.uk or Discover Uni? If it's a blog, a forum or a company selling something, find the same fact on the university's own site.
3. Even the university's own page can be old
This one surprised us. When an answer came from an out-of-date page on the university's own website, it was right only 7% of the time.
Old pages don't get deleted. Newcastle's employability page still leads with a figure for 2020/21 graduates. St Andrews' facts page still quotes last year's Guardian ranking. AI tools find these pages and repeat them as if they were current.
Check: look for a date on the page. Start from the course page for your entry year and follow the university's own links from there, rather than trusting whichever page the AI landed on.
4. Is that the normal offer, or the reduced one?
Many universities make lower offers to some applicants, often called contextual, adjusted or widening-access offers. AI answers often give the reduced grades as if they were the normal requirement.
We saw it with Edinburgh twice and with Hull York Medical School. We also saw grade ranges come out garbled, such as "AAA*", which isn't a real combination of A-level grades. On one platform "A*A*A" lost its asterisks and appeared as "AAA".
Check: on the course page, find the line that says standard or typical offer. Read the contextual offer separately and check whether you qualify for it.
5. Ask it twice
During our early tests, ChatGPT gave Manchester's Computer Science entry grades as A*A*A at about 11am. Asked a similar question less than an hour later, it said AAA. The real offer is A*AA.
Check: ask the same question again in a new chat, or on a different AI tool. If the numbers don't match, neither answer is safe until you've checked the university's page.
6. "Which is better?" answers are often forum opinion
When we asked which of four universities was better for economics, eight of Google's nine sources were Quora, Reddit, Instagram or The Student Room. The ninth was a student-housing company's top-10 list. One Quora thread dated from 2016.
Student opinion is useful. But an AI summary of forum posts isn't a ranking, and it can sound far more certain than the posts it came from.
Check: for "better", look at the subject tables in the current Complete University Guide, Guardian University Guide and Times Good University Guide, all 2027 editions. Then read what current students say, and decide what "better" means for you.
7. Watch where you're asking from
When we asked ChatGPT an open question ("which universities should my daughter apply to for medicine?") from Scotland, it listed five Scottish medical schools first. It did this when signed out too. A student in Birmingham asking the same words would probably get a different list.
That isn't wrong, exactly. But it means an AI shortlist reflects where you are, not just which courses suit you.
Check: name the universities you're interested in, and ask about those. You'll get answers you can compare and check.
8. Big money claims need two sources
The costliest single error in our study: Perplexity told an applicant from England that St Andrews only charges tuition for three of its four years. It doesn't. That's a different university's policy, and the mistake was worth about £9,800. The page Perplexity cited actually said the opposite.
Check: for anything about fees, funding or scholarships, read the university's fees page yourself, and if you're unsure, email its admissions or funding team. They answer these questions every day.
The short version
- Every figure has a year, and it's the year you're applying for
- The source is the university, UCAS, gov.uk or Discover Uni
- The university page itself is current, not an old news story
- You know which is the standard offer and which is the reduced one
- Two separate answers agree
- "Better" claims are checked against current league tables
- You named the universities instead of asking for a shortlist
- Fees and funding confirmed on the university's own fees page
Where to check
- Entry requirements: the course page on the university's website, for your entry year
- Fees in England: gov.uk (the 2027/28 cap is £10,050)
- Fees in Scotland: SAAS for Scottish students; the university's fees page for everyone else
- Graduate jobs and salaries: Discover Uni, which uses official data
- Deadlines: UCAS. The main deadline for 2027 entry is 13 January 2027, 6pm UK time
FAQ
Is ChatGPT accurate about university entry requirements? Often, not always. In our checks ChatGPT was right on 72% of the facts it gave, the best of the three tools we tested. Its errors included giving one course's offer a grade too high and quoting last year's fees as this year's.
Which AI tool is most reliable for university research? In our study, where the answer came from mattered more than which tool gave it. Answers that drew on a university's current page were right 79% of the time; answers that drew on an old page on the same university's site, 7%.
Should I trust AI for UCAS choices at all? It's a quick way to build a list of questions and places to look. It isn't a substitute for the course page, an open day, or your school or college adviser.
From the study. Every example on this page comes from our UK universities AI answer study: 198 facts checked against official sources, with report cards for 31 universities. The data is open on Zenodo.