Many applicants no longer start choosing a degree on a university website or with the school's student advisor. They start in a chat window. That changes what your programme pages have to cope with, because the first reader is now a language model summarising them for an 18-year-old. Below we follow one applicant through an imagined sequence, and look at where the institution's content works and where it drops out. The setting is Norway, but the pattern will be familiar anywhere.
The scenario: Sara asks an AI assistant
Sara is made up. She is in her final year of upper secondary school on the general studies track, takes science subjects, likes biology and wants to work with people. It is February, and she has not decided.
First question, on a Tuesday evening: "What can I study if I like biology and want to work with people?"
The assistant replies with a tidy list: biomedical laboratory science, nutrition, nursing, pharmacy, medicine, public health. Each profession gets two lines on what the job involves. No institution is mentioned yet. That answer draws on general knowledge about professions, and most of it is correct.
Second question: "Where can I study biomedical laboratory science in Innlandet or Trøndelag, and what are the entry points?"
Now the model starts pulling from the web. It names two institutions and gives entry point thresholds. One is from an earlier year, because that was the number in plain text on a page the model found. The other comes from a news story about "popular programmes", not from the institution itself. The assistant does not distinguish between the quota for first-time school leavers and the ordinary quota. A third institution that offers the programme is not mentioned, because its admission requirements are in a PDF and the programme page is mostly images and a video.
Third question: "Do I need specific subjects to get in?"
This is where it goes wrong in a way Sara does not notice. The assistant says "most programmes require science subjects" and lists subject combinations that resemble the requirements without matching them. One institution's programme page has a section on "recommended prior knowledge" above the section on admission requirements, and the model mixes the two.
Fourth question: "What kind of job do I get afterwards, and what does it pay?"
The answer is general and reasonable. But the institutions' own figures on where graduates work, which several of them have, are not used. They sit in a report that the programme page does not link to.
Sara ends up with a shortlist of two places. She visits the websites to check what the assistant said, or she does not. Either way, the chat has already decided which places she looks at more closely.
What the scenario shows
In every case the right answer existed somewhere, but it was hard to find, read or tell apart from something else. Three patterns keep coming back:
- An outdated number beats the right number when the old one is in plain text and the new one is in a table loaded with JavaScript, in a PDF or behind a tab.
- Third-party sources fill the gap when the institution's own page does not answer the question. A newspaper article or a forum becomes the source.
- Near-requirements get mixed up with requirements when pages have an unclear structure and no clear headings.
How big is this?
We do not know exactly how many Norwegian applicants use AI when choosing what to study. As far as we have found, no Norwegian survey measures that specifically. But several figures point the same way.
The Norwegian Media Authority's Children and Media 2026 report, published in August, shows that half of all 9 to 18-year-olds use AI services such as ChatGPT, and the share is highest among the oldest. Among those who use AI, 78 percent say they use it to search for or find information. That is the age group this year's and next year's applicants belong to.
Among students, use is close to universal. The UK HEPI Student Generative AI Survey 2026 found that 95 percent of full-time undergraduates use AI in at least one way. The survey covers British students and their studies, not how they chose a course, but it says something about the habits applicants bring with them.
AI has also moved into ordinary search. Google launched AI Mode in Norway in October 2025. A search for "biomedical laboratory science entry points" can then return a summarised answer at the top, and the applicant may never click through to the source.
Meanwhile, competition for applicants is real. Norway's national admissions service received 150,856 applicants to universities and university colleges in 2026, up 6 percent, spread across 1,395 programmes. The application deadline was 15 April, and admissions were completed on 20 July (HK-dir). Much of the exploring happens in the months before the deadline, and that is when AI answers shape the shortlist.
What educational institutions can do
Make the programme page the reference
Every programme should have one page that answers the questions Sara asked, in plain text with clear headings:
- Admission requirements, with exact subject codes or names, clearly separated from recommended prior knowledge.
- Entry point thresholds from the latest admission round, labelled with year and quota.
- Number of places, campus, duration and mode of study.
- What graduates go on to do, ideally with figures from your own graduate surveys and a link to the source.
- Application deadline and where to apply.
Do not put admission requirements and entry points in PDFs, images or tabs that need a click. People and models alike read what is at the top and clearly stated.
Update on fixed dates
Entry point thresholds come out in July. The web team's annual plan should say that every programme page is updated straight after admissions, and that old figures are labelled with the year. The same goes for admission requirements that change. A page that says "entry points: 48.2" with no year invites mistakes.
Mark up programmes with structured data
Schema.org has the types EducationalOccupationalProgram and Course, with properties for admission requirements (programPrerequisites), application deadline (applicationDeadline), duration (timeToComplete), qualification (educationalCredentialAwarded) and occupation (occupationalCategory), among others. This makes the facts easier for machines to read.
Be realistic about the effect. Google says in its documentation on AI features in Search that there are no additional requirements or special schema markup for appearing in AI Overviews or AI Mode, and that standard SEO best practice applies. Structured data is good craft, and no shortcut. What matters most is still that the text on the page is correct and easy to find.
Write FAQs in applicants' own words
Applicants ask "will I get in with 45 points?" and "can I retake subjects later?". Answer questions like these directly on the programme page. Collect them from student advisors, chat logs and emails to the admissions office.
Test what AI assistants say about you
Set aside an hour each month during the application period. Ask ChatGPT, Gemini, Copilot and Google's AI Mode the same questions Sara asked. Note which sources they cite, which numbers they give and where they are wrong. When an error keeps coming back, trace it: is it an old page, a PDF, a news story or a gap in your own information?
Also run a simple technical check that key pages can be read without JavaScript, and that robots.txt does not block search engines you want to be found in.
[QA: add own example from testing what AI assistants said about a study programme]
The student advisor is not going away
Many applicants will still want to talk to a person before they decide. But by the time they do, they often have a shortlist from a chat. It is better if that shortlist rests on correct figures from your own page than on a news story from last year.
I am responsible for the education segment at Frontkom, where we work on websites and programme page structures for universities, university colleges and vocational colleges.
Sources
- Barn og medier 2026, Medietilsynet, August 2026
- Søkertall for universitet og høgskoler i Samordna opptak 2026, HK-dir, 2026
- Student Generative AI Survey 2026, HEPI, 12 March 2026
- Google lanserer noe nytt i Norge i dag, ITavisen, 8 October 2025
- AI Features and Your Website, Google Search Central, updated 10 December 2025
- EducationalOccupationalProgram, Schema.org
Transparency note: This article is based on the author's own analysis and experience. AI has been used as an editorial aid for language and structure.
