Highlights
- The dataset: This report draws on over 1.08 million AI source citations, collected with prompts specific to the university sector. The underlying AI responses, from four AI models, were gathered in 13 countries throughout Q2 2026.
- Reddit climbed from #7 to #2 in the window (+31%): The only top domain to rise between the two halves.
- 83% of Google AI Overview's Swedish citations are local: Google AI Overview cites local .se domains 83% of the time on Swedish prompts, Grok just 26%.
- 22% of ChatGPT's citations are from institutional domains: Universities' own pages, the highest institutional share of any model.
- Times Higher Education alone makes up 13% of Microsoft Copilot's citations: That makes Microsoft Copilot the most concentrated of the four models.
- ~25% average overlap between any two models: Any two of the four share only about a quarter of their top-20 cited university domains.
Which sources should you target to get cited as a university?
Each model relies on its own distinctive top domains.
| Model | Top sources | Citation character |
|---|---|---|
| Microsoft Copilot | Times Higher Education (13%), edurank.org, unirank.org, usnews.com | Ranking & reference tables |
| Grok | topuniversities.com (9.6%), Times Higher Education (9.4%), usnews.com, scimagoir.com | Ranking aggregators & indices |
| ChatGPT | Reddit (10.3%), Times Higher Education, en.wikipedia.org, topuniversities.com | Community + reference |
| Google AI Overview | youtube.com, Reddit, topuniversities.com, instagram.com | Social + video |
Times Higher Education and QS Top Universities are the closest thing to common ground, they appear in every model's top tier, but their weight ranges from 13% of Microsoft Copilot's citations to under 3% of Google AI Overview's.
How have AI source rankings changed over time in the university sector?
We split the observation window at its midpoint (early May 2026) and compared the most-cited domains in each half. The ranking is far from static.
Times Higher Education held the top spot throughout, but beneath it the order reshuffled: community content (Reddit) rose while some niche ranking aggregators fell steeply. AI citation rankings are dynamic, not static, track them in the live AI visibility rankings.
What type of content do AI models cite for universities?
The models differ not just on which domains but on what type of content they trust. We classified every cited domain into six categories.
Microsoft Copilot is the ranking-and-reference engine: editorial ranking sites plus reference databases account for nearly 70% of its citations, and it almost never cites user-generated content (4%). ChatGPT is the most institutional, sending 22% of its citations to universities' own pages. Google AI Overview is the most commercial and social.
Do AI models cite local-language content for universities?
Outside English-speaking markets, models vary sharply in how often they cite local-country domains. Using country-code top-level domains as a proxy for local-language sourcing, the local-domain citation rate ranges from 52% (Italian) down to 33% (French). English prompts are excluded, English content overwhelmingly lives on generic domains.
Even in the most localized market, roughly half of cited sources are non-local, a sign that English-language and international ranking sites dominate university discovery far beyond the English-speaking world.
Which AI model relies most on local sources for universities?
The localization gap between models is large and consistent. Google AI Overview cites local-domain sources far more often than Grok in every language tested.
Google AI Overview and ChatGPT lean local; Grok consistently reaches for global, largely English ranking sites regardless of the prompt's language. For a university competing in Sweden, Germany, or Italy, local-language presence pays off on Google AI Overview and ChatGPT but does comparatively little for Grok visibility.
Should universities optimize for each AI model separately?
Yes, almost entirely. When the four models answer the same university questions, their top-cited domains barely converge. We compared the 20 most-cited domains for each model and measured how much they share.
ChatGPT and Grok are the most divergent pair, sharing just 14% of their top sources. A university visible on one model can be absent from three-quarters of the sources another model cites.
How many sources does each AI model cite per answer?
The models disagree even on how many sources to cite per answer.
Context
This analysis draws from Temso's AI visibility monitoring platform, which tracks how brands appear in AI model responses across ChatGPT, Microsoft Copilot, Grok, and Google AI Overview. The dataset covers university-related prompts ("best universities in [country/field]") across 13 countries and 7 languages over roughly sixteen weeks (March–June 2026).
It measures what AI models cite, not the underlying quality of any university. These findings reflect citation behavior for university-related prompts in the monitored markets, and may not generalize to all institutions, prompt types, or other industry verticals.
Methodology
How we measured this
We tracked four AI models responding to university prompts in local languages across 13 countries. Each response was parsed to extract source citations: the URLs, domains, and metadata referenced. Domain categories (editorial, commercial, UGC, reference, institutional) were assigned from a global domain registry. Cross-model overlap was measured using Jaccard similarity on each model pair's top-20 most-cited domains. Temporal analysis split the observation period at its midpoint and compared domain rankings in each half. Model coverage varied across the observation window, so temporal movements are reported as relative ranks within each half.
The analysis covers over 1.08 million cited sources from 107,281 AI responses. Sample sizes exceeded recommended thresholds for every reported finding. Localization is measured by a country-code-domain proxy (the source records carry no detected content language), so those figures are directional and exclude English-language prompts, which sit overwhelmingly on generic domains.
Frequently asked questions
Do AI models cite different sources for university recommendations?
Yes, dramatically so. Any two of the four models share only about 25% of their top-20 cited university domains, so roughly three in four sources differ. ChatGPT and Grok overlap the least, at 14%.
Which AI model cites the most sources per answer for universities?
Grok, by a wide margin, about 24 cited sources per answer, versus roughly 10.5 for Google AI Overview, about 8.7 for Microsoft Copilot, and 5.8 for ChatGPT.
What kinds of websites do AI models cite for universities?
It varies by model. Microsoft Copilot leans on editorial ranking sites and reference databases (about 69% combined); ChatGPT cites the most institutional sources (universities' own pages, 22%); Google AI Overview is the most commercial and social.
Do universities need local-language web content for AI visibility?
For some models. Google AI Overview and ChatGPT cite local-country domains frequently, Google AI Overview reaches 83% on Swedish-language prompts, so local-language presence helps there. Grok cites local domains far less, often 25–30%, favoring global English ranking sites.
Which AI model relies most on local sources for universities?
Google AI Overview. It cites local-country domains more than any other model in every language tested, 59–83% in German, Dutch, Swedish, and Italian, while Grok stays on global sources.
How fast do AI university source rankings change?
Quickly. Over roughly sixteen weeks, Reddit rose from the 7th to the 2nd most-cited domain while several ranking aggregators fell sharply; only Times Higher Education held its position.
Is one AI ranking enough to optimize for in higher education?
No. The models share only about a quarter of their top sources and weight them very differently, so visibility on one model does not transfer to the others. An AI-visibility strategy for universities has to be built model by model.
