How to Measure Your Education Brand’s AI Visibility

Your education business can sell on two clocks: programs with an application deadline and an intake, and courses or apps students buy from a catalog every day.

Then someone asks whether ChatGPT recommends you. One person types one question, a screenshot goes round, and that screenshot stands in for both businesses.

Students applying to UK universities already put these questions to assistants. In UCAS's survey of students registered with it, 48% of those who use AI had used it to explore their higher-education options, and recommending universities and courses was the second thing they asked about.

Measuring it properly takes a spreadsheet and a chat assistant: write each program the way a student asks, save what both engines answer, and have an assistant sort the names and the sites for a person to check.

To see what one screenshot misses, we asked ChatGPT and Google AI Mode about a range of subjects in October, from data science and MBAs to trades and languages, several times each.

Every subject went in twice: once the way a student picks a school ("Best language schools"), once the way a learner picks a course or an app ("Best language-learning apps").

One limit before the steps: a citation shows which page an answer linked to, not which school or course the student chose, and nothing here measures enrollments.

You will leave with separate school and course lists, owners for both, and a brief for the recurring check. Near the end, we explain how we handle that work at Qvery.

This page measures what the answers name and cite. It doesn't check whether the facts they print about your programs are right; keeping one record per program and checking every copy is its own job.

Sort Every Program by What the Student Is Choosing, Then Write It Both Ways

Start from your program list, not your org chart. For each line, ask one question: is the student choosing a place to enroll, or a thing to use?

A place has an application, an intake and a start date. A thing has a sign-up page, a monthly price and a login.

A bootcamp can be a school to one applicant and a course to another. Write both questions even when one team sells the program.

So write every program both ways, in your students' words. These are pairs from our research:

On the median subject, the school question and the course question shared no cited site at all, on ChatGPT and on Google AI Mode alike; on each engine 54.17% of subjects shared none.

A subject's overlap here is the share of the sites cited for its course question that also turned up for its school question, across all its asks.

Languages show it plainest. ChatGPT answered the app question from Duolingo's, Babbel's, Pimsleur's and Busuu's own sites every time we asked; Google AI Mode cited a Reddit thread and Wirecutter every time.

The school question went to QS and GoAbroad on Google AI Mode. On ChatGPT, four asks out of five, it went to the schools of a different town each time.

Take the wording from your own inbox: the first line of an admissions email, the search box on your course catalog. An assistant can pull questions out of a term's inquiries for you, the way it would from sales calls and emails. Keep the subject inside the question.

Freeze the wording once it's written. A question reworded next month is a different question, and its answer can't be compared with this one.

Ask Each School Question From Where Your Applicants Are

ChatGPT narrowed 23.33% of its school answers to a place the question never named, and none of its course answers.

Asked "Top trade schools" from the US on five separate asks, it answered for Minnesota, then Jacksonville, Myrtle Beach, Leesburg in Virginia and Oakland. One answer for night-shift nurses opened, "Since you appear to be in Mississippi."

OpenAI’s help page says ChatGPT may use an approximate location from your IP address for local results. Ask school questions from the places you recruit.

ChatGPT also cited Google Maps listings in 10.0% of its school answers, against 0.83% of its course answers, so each campus's map listing is part of the school answer.

Use a colleague in the applicant’s city or a connection in their country.

Writing the city into the question is a different question; keep it as a second version if students type it that way. Ask the course questions from any country you sell into.

Use a session with no history. OpenAI says saved memories can shape the search ChatGPT runs, so use a temporary chat or no account; open Google AI Mode in a window signed out of Google.

Ask each question several times per engine on the same day. In our research, two asks shared 38.34% of cited sites on ChatGPT and 33.23% on Google AI Mode, on average; no question kept the same set throughout.

Save every answer whole: the text, and every link under it, each with an ID that says side, question, engine and ask (S4-CG-2 is school question four, ChatGPT, second ask).

Log the Names and the Pages Separately, Because School Answers Split Them

Google AI Mode cited no college's or school's own site in 72.65% of its school answers, and those answers still named schools.

Asked for the top colleges for business, one answer put the University of Pennsylvania's Wharton first and cited US News, Poets&Quants, QS, a college counselor's blog, YouTube and Instagram: no page of Wharton's.

ChatGPT works the other way on schools: it cited a school's own site in 93.33% of its school answers.

Keep two tabs. Names holds the options in answer order and whether their own sites were cited. Sites holds every link and its kind of site.

That reading is where the hours go, and it's the part an assistant does. Paste this prompt, then five saved answers under it, each starting with its ID line, then the answer, then SOURCES and one link per line:

You are filling two tables from saved AI answers about schools, courses and learning apps. Each answer below starts with a line "ID | engine | question", then the answer text, then "SOURCES:" with one link per line.

Use only what is written in the answers and their source lists. Never add a name, a site or a link from memory.

TABLE 1, NAMES. One row for each school, program, course, platform, app or tutoring service the answer puts forward as an option to choose. Tab-separated, in this order:
1. Answer ID
2. Order: 1 for the first option in the answer, 2 for the next, and so on
3. Name: exactly as the answer writes it
4. Own site cited: yes if one of the links in SOURCES is on that option's own website, otherwise no
Not a row: a site the answer only quotes as a source ("according to U.S. News"), a name that appears only in the SOURCES list, a kind of school or product with no name ("community colleges", "free apps"). Two different schools or providers in one line ("Preply / iTalki") are two rows; one school written two ways is one row.

TABLE 2, SITES. One row for every link in the SOURCES list, repeats included. Tab-separated, in this order:
1. Answer ID
2. Site: the address without "www." and without anything after the first slash (usnews.com)
3. Kind, by who owns the site, not by what the page says (a platform's own blog post is still provider), exactly one of:
   school = a college's, university's or other school's own site, its online or continuing-education pages included
   provider = the own site of a course platform, an app, a tutoring company, a bootcamp or a training company
   marketplace = a site that lists many providers' courses, tutors or apps (Coursera, Udemy, edX, LinkedIn Learning, italki, Preply, an app store)
   ranking = a ranking, a school directory, a best-of roundup, or a map or review listing of one place (a Google Maps or Yelp link), whoever publishes it
   public = a government body, a regulator, an accreditor or an exam body
   community = Reddit, a forum, YouTube or a social network
   news = a news or magazine article
   other = anything else
4. Named in the text: yes if the site belongs to a school or provider the answer puts forward as an option, otherwise no

After the two tables, one line per answer: the Answer ID, how many rows it has in Table 1, and how many links it has in SOURCES.

Table 1 goes into the Names tab and Table 2 into the Sites tab. Then a person reads every row beside its answer.

I checked the first prompt’s rows against the saved answers. My check changed one row in fifteen; none was invented. These mistakes shaped the printed prompt:

On the same answers, the revised prompt fixed map listings, the tutor marketplace, and the joined row. It still misclassified the coding platform’s posts and Poets&Quants.

The person's read stays in the routine, however good the prompt gets.

Telling a real change from one ask's variation takes the same questions on a fixed schedule, every day if you can, on both engines, with every source kept.

Coursera Sits Under the Course Answers and US News Under Google's School Answers

Once the Sites tab is full, count each kind of site per side: the share of answers that cited at least one of that kind.

From our research: the same subjects asked as a school question and as a course question on ChatGPT and Google AI Mode on 19 October 2026, several times each, mostly from the US with some UK, Canadian and Australian questions, each share pooled across the asks and taken of the answers that cited any source.

Google AI Mode leans on different kinds of site and cites fewer of them per answer, so read it against itself:

School answers cited rankings and directories such as US News and QS; course answers cited platform and app roundups. PCMag appeared under 14.55% of Google AI Mode’s course answers and none of its school answers.

The same split showed up in our degree and online-course questions, a different set asked the same week. And on course questions about credentials, ChatGPT leaned on the bodies that approve the program, which is worth a look if you teach to an exam.

Reddit is a Google AI Mode question here. It sat under 25.45% of Google AI Mode's course answers and none of ChatGPT's, which matches ChatGPT dropping Reddit citations in August.

On ChatGPT a Few Marketplaces Cover the Course Side, and the School Side Stays Long

Build each list the way the fintech tutorial builds its two: take the site under the most answers first, set those answers aside, take the site under the most of the rest, and stop at four fifths.

On ChatGPT, Coursera alone sat under 70.83% of the course answers. The school side's most-cited site, Southern New Hampshire University, sat under 18.33%. Reaching half of the school answers took four times as many sites as reaching half of the course answers, and four fifths took 3.33 times as many.

Google AI Mode spread its course answers across Reddit, Coursera, PCMag, and the providers, so its course list ran 1.67 times as long as its school list to half, and 1.25 times as long to four fifths.

So keep four lists, one per side on each engine, and never add them together.

Check whether the courses you already list on Coursera, edX, Udemy or LinkedIn Learning carry the names students search; whether to list on one you're not on is a business decision the citation share only informs. The school side is your own pages and the rankings that profile you, spread across many sites.

Your Admissions, Catalog and Compliance Leads Already Own These Sites

Every kind of site on a reading list is somebody's job already. The table names who, and what each checks first:

US News builds its rankings partly from data schools send in through its surveys, which makes the person who files the return part of how students hear about you. The roundups are earned the way any best-of list is, and getting onto the lists AI answers cite is its own playbook.

Re-Run the School List Around Intake Dates and the Course List Every Month

For UK undergraduate entry in 2027, UCAS lists 15 October 2026 for Oxford, Cambridge and most medicine courses, 13 January 2027 as the equal-consideration date for most courses, and 2 July 2027 for the opening of Clearing.

Read the school list before each deadline, a few weeks ahead, and again just after. A US college runs the same rule on its own deadlines.

The course side sells every day, so read it monthly, and again after any change to a marketplace listing, a price or a course name.

The asking and the reading run on two clocks of their own. Collect answers as often as you can, several asks a question each run, from your team or from a daily collection; read them on the dates above.

Comparing each run with the last, row by row, is the same job every time, which makes it a job for an AI agent. Here is the brief:

1. What It Reads

   The answers collected since the last read, the two question lists, each side's last reading list with its owners, and your own program and course pages. It doesn't ask the engines itself, and it never logs into a marketplace or a ranking site: where a row points at someone else's site, it hands the owner the address to open.

2. When It Runs

   The school list a few weeks before each intake deadline and the week after; the course list on the first working day of each month and after a listing or price change.

3. The Rule It Checks

   A row moves when its share of the side's answers changes by more than the gap between two asks of the same questions in your first run. Smaller changes are the engines' own variation and go unreported.

4. What It Tracks

   Two numbers per side and engine: the share of answers that name you, and the share that cite one of your own pages.

5. What It Produces

   One page per side: the rows that moved, each with its owner from the table above, and the answers behind them.

6. Who Reads It

   The admissions lead for the school side and the catalog lead for the course side. Each decides what to do, and the agent changes nothing outside the sheet.

7. How It's Counted

   The agent's work in hours: the time your first manual comparison took per side is the baseline; afterward, the time the owners spend reading its page. The visibility result is the two numbers above.

Set up the two lists and their owners first: without them, an agent can only report that the answers changed, never whose problem it is.

How Qvery Keeps Your Program Lists Current

For student discovery, we measure with the GEO Agent. It develops questions about your education business from the programs you sell, your industry, and competing providers, then runs them through Google AI Mode and ChatGPT daily for each country you select. It keeps the answers and citations for inspection.

You can examine sentiment alongside your visibility, see how much share of voice competing providers take, and break those measures down by country or topic. We use those findings to work on the pages, sources, and program facts behind the answers.

For the repeated comparison, we learn how admissions and catalog staff review their sheets, then create a custom agent in the systems they use. It runs on autopilot against the rule above and gives each owner the changed rows to approve.

In two weeks, we deliver a first working comparison and train your admissions and catalog leads to operate it; our weekly check then counts their hours saved and reporting errors avoided. Bring the sheets from your first run so we can work from the same baseline.

Put Both Re-Runs on the Calendar This Week, With a Name Beside Each

Schedule a school-side read before your next intake deadline, owned by admissions, and a course-side read on next month’s first working day, owned by your catalog lead.

Before those dates, test both questions for one bootcamp or certificate. Put the resulting lists in your sheets.