The AI Recommendation Index: Higher Education
Which colleges ChatGPT, Gemini, Claude, and Perplexity recommend to students, and how accurate they are about cost, admissions, and outcomes. Q4 2026 edition.
By Jarred Smith · Data collected October 1, 2026 · Published October 2026 · Next update January 2027
Students are already asking AI where to go to college. The Chronicle of Higher Education reported in April that about a third of high schoolers who use AI in their search found new schools to consider that way, and about a fifth dropped a school because of what a chatbot told them. So I asked ChatGPT, Gemini, Claude, and Perplexity the questions a student, a parent, or a working adult would ask, worded the same way every time, with web search on, and wrote down what came back.
That came to 687 usable answers. The questions cover four segments (undergrad programs near Oklahoma City, online degrees, graduate and MBA programs, and community college and workforce training), the same local questions in five cities, four kinds of buyers, ten schools' published facts checked against College Scorecard, and a set of scripted conversations where the buyer narrows down to one school.
The short version: the four engines rarely agree on which schools to name, they get acceptance rates and accreditation right but quote a different price for the same school, and they learn most of what they know about colleges from Niche, US News, and other directory sites rather than federal data.
AI Recommendation Index · Higher Ed, Q4 2026
of the schools recommended for a question were named by only one of the four engines
of answers cited Niche. College Scorecard, the federal source, was cited in 7%
first net-price figures landed within 10% of College Scorecard (acceptance rates and accreditors: every one matched)
conversations where AI named Fort Hays State or OCCC on its own, against 8 of 12 for SNHU and Google
Source: AI Recommendation Index: Higher Education, Q4 2026, Jarred Smith. 687 AI answers from ChatGPT, Claude, Gemini, and Perplexity with web search on, October 1, 2026.
Key findings
- The four engines rarely agree on which schools to recommendFor the same open-ended question, two engines' lists overlapped 23% on average, and 72% of the schools recommended for a question were named by just one engine. The same engine asked the same question again overlapped with itself 42% of the time.
- Western Governors University and University of the People lead the online degree answersWestern Governors University showed up in 50% of open-ended online degree answers and University of the People in 44%, ahead of Southern New Hampshire University at 30%. University of the People is tuition-free and spends a fraction of what the big online universities spend on advertising. Two small schools, Fort Hays State and Newlane University, each showed up in more than a fifth of answers.
- Location pushes the national online brands outAsked for the best online bachelor's degree in business with no location given, 0% of recommended schools were in Oklahoma. When the student said they lived in Oklahoma, 68% to 85% were, and WGU and SNHU mostly dropped off the list. To separate the two, I asked the same question with only "I live in Oklahoma" in front of it. Location alone moved the list to 38% Oklahoma schools, so where the student lives explains a good part of the shift and the buyer's situation and budget account for the rest.
- Acceptance rates and accreditation come back right, prices don'tEvery acceptance rate and accreditor matched College Scorecard. The first net price quoted was within 10% of Scorecard in only 20 of 39 answers. For the University of Oklahoma, the four engines' first figures were $16,952, $18,574, $22,312, and $40,145, against Scorecard's $15,300.
- Directory sites supply more of the answer than federal dataNiche was cited in 33% of answers and US News in 19%. College Scorecard appeared in 7% and NCES in 3%. ChatGPT pointed 80% of its citations at schools' own .edu pages, while Claude, Gemini, and Perplexity leaned on directories.
- Smaller schools get endorsed once someone asks about themIn scripted conversations, AI named Fort Hays State or Oklahoma City Community College on its own in 1 of 12 conversations, compared with 8 of 12 for SNHU and Google Career Certificates. When the buyer asked about them directly, the smaller schools got a clear yes in 10 of 12 conversations and the leaders in 8 of 12.
AI college rankings by segment
Pick a segment to see the schools the engines recommended most often, then switch engines to see how each one differs. These shares come from open-ended questions only, the kind where no school is named in the question (for example, "What are the cheapest legitimate online bachelor's degree programs?"). Each question was asked three times per engine.
Which colleges does AI recommend?
Two things stand out. Each segment has a long tail: the engines named 84 different online schools across 46 answers, and the top three schools account for only 23% of all online recommendations. Each engine also has favorites the others ignore. Perplexity named University of Phoenix in 6 of its 11 online degree answers, and the other three engines never named it.
The same question in five cities
I asked four local questions in Oklahoma City, Atlanta, Phoenix, Chicago, and Boston, changing nothing but the city. The engines localize well: every city got its own schools, and national online brands didn't appear in any local answer outside Phoenix, where Arizona State and Grand Canyon are the local schools.
| Question | Oklahoma City | Atlanta | Phoenix | Chicago | Boston |
|---|---|---|---|---|---|
| Best colleges for nursing | Oklahoma City Community College, University of Central Oklahoma, Oklahoma State University-Oklahoma City (6.2) | Emory University, Georgia State University, Clayton State University (5.5) | Arizona State University, Grand Canyon University, Maricopa Community Colleges (6.7) | University of Illinois Chicago, Rush University, Loyola University Chicago (7.3) | Boston College, Northeastern University, Simmons University (6.5) |
| Most financial aid | Oklahoma City University, Oklahoma Baptist University, Cameron University (2.0) | Emory University, Oglethorpe University, Morehouse College (5.3) | Grand Canyon University, University of Phoenix, Phoenix College (1.0) | University of Chicago, University of Illinois Chicago, City Colleges of Chicago (4.0) | Harvard University, MIT, Wellesley College (5.8) |
| Part-time MBA | Oklahoma City University, University of Oklahoma, University of Central Oklahoma (4.2) | Georgia Tech Scheller, Emory Goizueta, University of Georgia Terry (3.0) | Arizona State W. P. Carey, Arizona Eller (2.3) | Northwestern Kellogg, Loyola Quinlan, Chicago Booth (3.5) | Boston College Carroll, Boston University Questrom, Northeastern D'Amore-McKim (6.0) |
| Community colleges for welding | Metro Technology Centers, Francis Tuttle Technology Center, Oklahoma City Community College (4.5) | Atlanta Technical College, Chattahoochee Technical College, Georgia Piedmont Technical College (4.0) | Mesa Community College, GateWay Community College (2.5) | Richard J. Daley College, Elgin Community College, College of DuPage (7.3) | Bunker Hill Community College, North Shore Community College (2.8) |
Top schools by number of answers naming them; the figure in parentheses is the average number of schools recommended per answer. Oklahoma City has 11 to 12 answers per question, the other cities 3 to 4.
The financial aid question is the one that surprised me. In Boston, Atlanta, and Chicago, "which colleges give the most financial aid" got answered with wealthy, highly selective schools (Harvard, MIT, Emory, the University of Chicago), the ones with big endowments and need-based aid that most applicants won't be admitted to. In Oklahoma City and Phoenix, the engines named one or two schools per answer.
How the buyer changes the answer
A second set of questions opened with the buyer's situation, home state (Oklahoma), and budget, and then asked the same thing as the baseline. The table shows the share of recommended schools that are Oklahoma institutions.
| Question | No buyer | Location only | First-gen senior | Parent | Working adult | Transfer student |
|---|---|---|---|---|---|---|
| Best accredited online bachelor's in business | 0% | 38% | 81% | 68% | 85% | 78% |
| Best online MBA under $30,000 | 5% | 46% | 67% | 29% | 77% | 17% |
| Is UCO a good school? | 83% | 100% | 60% | 100% | 100% | 100% |
Persona answers: 4 per persona per question (one per engine). Baseline: up to 8 answers. Location only: up to 8 answers.
The online MBA question shows that buyer type matters even when everyone lives in the same state. Every persona said Oklahoma, yet the working adult got Oklahoma programs 77% of the time and the transfer student 17%.
How accurate AI is about colleges
I asked each engine four factual questions about ten schools (net price after aid, acceptance rate, graduation rate, and accreditor) and graded the answers against the College Scorecard, the U.S. Department of Education's public data, pulled October 1, 2026.
| Fact | Matches Scorecard | Right figure appears later in the answer | Off | No answer |
|---|---|---|---|---|
| Accreditor | 37 | – | 0 | 3 |
| Acceptance rate (or open admission) | 37 | – | 0 | 3 |
| Graduation rate, within 5 points | 30 | 8 | 1 | 1 |
| Net price after aid, within 10% | 20 | 13 | 6 | 1 |
The first net price each engine quoted
| School | Scorecard | ChatGPT | Claude | Gemini | Perplexity |
|---|---|---|---|---|---|
| University of Oklahoma | $15,300 | $16,952 | $22,312 | $40,145 | $18,574 |
| Georgia State University | $15,931 | $15,931 | $21,607 | $16,854 | $16,854 |
| Southern New Hampshire University | $36,708 | $33,163 | $29,878 | $33,163 | $36,309 |
| Oklahoma State University | $17,447 | $20,000 | $15,425 | $19,561 | $17,447 |
| Oklahoma City Community College | $4,739 | $4,700 | $5,905 | $5,905 | $4,739 |
| Arizona State University | $14,967 | $17,779 | $17,779 | $17,779 | $17,779 |
| University of Central Oklahoma | $18,309 | $18,000 | $18,660 | $18,660 | $18,660 |
| Fort Hays State University | $12,569 | $13,774 | $13,186 | $13,186 | $13,186 |
Scorecard's net price trails by a year or more, and some engines quoted newer figures straight from a school's own site (ChatGPT's $16,952 for Oklahoma is OU's published 2024–25 number), so some of these mismatches come down to timing. For a student, the bigger issue is the spread. Gemini led its answer about cost after aid with OU's $40,145 sticker price, and its net-price answers contained anywhere from 7 to 29 dollar figures each.
Where AI engines get their college information
The engines cited 6,940 web pages across 546 answers. Directory and ranking sites made up 52% of those citations, schools' own sites 40%, and government data 3%.
Most-cited sources in AI college answers
Share of answers citing each domain at least once
6,940 citations from 546 search-on answers. Source: AI Recommendation Index: Higher Education, Q4 2026.
| Engine | Citations | Per answer | To school .edu sites | To government sources |
|---|---|---|---|---|
| ChatGPT | 778 | 5.6 | 80% | 8% |
| Claude | 1,709 | 12.4 | 26% | 1% |
| Gemini | 1,772 | 13.1 | 38% | 1% |
| Perplexity | 2,681 | 19.9 | 38% | 4% |
This matches what the main index found for brands: ChatGPT reads your own site, and the other three read what everyone else says about you. For a college, that splits the work between your .edu pages and your profiles on Niche, US News, College Board's BigFuture, and the long tail of college-comparison sites.
Memory versus live search
With web search turned off, engines lean on big-name flagship programs they learned about in training. With search on, they shift toward online-only, career-focused, and low-cost regional schools that rank well on directory sites. Only 22% of the schools named from memory (18 of 82) also appeared in the search-on answers to the same questions.
| Question | From memory (search off) | With live search |
|---|---|---|
| Best accredited online bachelor's in business | Arizona State, Univ. of Florida, Penn State World Campus, Oregon State, UMass Amherst | Penn State World Campus, WGU, University of Phoenix, Maryville, National University |
| Best online MBA under $30,000 | Univ. of Illinois Gies iMBA, UW MBA Consortium, Univ. of Florida, Penn State World Campus | Georgia Southwestern, UT Permian Basin, Texas A&M International, Texas A&M-Corpus Christi |
Search-off round: ChatGPT, Claude, and Gemini, one answer each (Perplexity always searches). Directional.
What happens in a conversation
Each pair of schools went through the same scripted three-turn conversation: an open question, then "where does [school] fit among those options?", then "given my budget and constraints, should I choose [school]?". The first buyer was a 34-year-old working adult with some college credits and a budget under $15,000 a year. The second was a 29-year-old retail manager in Oklahoma City who wanted an IT job within a year and could spend under $5,000.
| School | Role | Named on its own in turn 1 | Verdict at turn 3 |
|---|---|---|---|
| Fort Hays State University | Challenger | 0 of 6 | 5 yes, 1 conditional |
| Southern New Hampshire University | Leader | 4 of 6 | 2 yes, 3 conditional, 1 no |
| Oklahoma City Community College | Challenger | 1 of 6 | 5 yes, 1 conditional |
| Google Career Certificates | Leader | 4 of 6 | 6 yes |
ChatGPT, Claude, and Perplexity, two conversations each. Gemini conversations were rejected by the provider's rate limits too often to include.
That's the same pattern as the AI Verdict Study. The visibility problem for a school like Fort Hays State sits in the first answer. Once a student asks about it, the engines like what they find.
What this means for colleges
Four things stand out to me, and they line up with what I see when I audit brands in other categories.
First, a single check tells a college very little. With engines agreeing on 23% of their recommendations, a school needs to test the questions its students ask, on all four engines, more than once.
Second, the competition is local. The national online giants vanish from answers once a student mentions where they live, and the schools that replace them are in-state options at that student's budget.
Third, the price is the weak spot. Engines quote acceptance rates and accreditation correctly, then give students four different costs. A plain, current net price on your own site, matching what you report to the Department of Education, is the easiest fix on this list.
Fourth, the sites that describe you matter as much as your own. Your Niche, US News, and BigFuture profiles, and the comparison sites that copy them, shape three of the four engines' answers. If those pages are stale, so is what AI tells a seventeen-year-old about you.
Want to see where your school shows up?
The free AI Visibility Quick Check takes about 20 minutes. The AI Visibility Audit runs this kind of testing on your programs and your competitors.
I wrote more about why this happens in Explainable.
Frequently asked questions
Which online colleges do ChatGPT and other AI engines recommend?
Across ChatGPT, Claude, Gemini, and Perplexity, the online schools named most often in open-ended questions were Western Governors University (50% of answers), University of the People (44%), and Southern New Hampshire University (30%). ChatGPT named the most schools per answer, about 6.5.
Do AI engines agree on which colleges to recommend?
Not often. For the same question, two engines' lists of recommended schools overlapped 23% on average, and 72% of the schools recommended for a question came from only one of the four engines.
Is AI accurate about college costs?
Less than it is about other facts. Every acceptance rate and accreditor in the test matched federal data, but the first net price an engine quoted was within 10% of College Scorecard in 20 of 39 answers, and four engines gave four different prices for the University of Oklahoma.
Where do AI engines get college information?
Mostly from college directories and ranking sites. Niche appeared in 33% of answers, US News and research.com in 19% each. Federal sources were rare: College Scorecard in 7% of answers and NCES in 3%. ChatGPT was the exception, with 80% of its citations pointing to schools' own .edu pages.
Does a student's location change AI college recommendations?
Yes, a lot. Asked for the best online bachelor's degree with no location, none of the recommended schools were in Oklahoma. When the student said they lived in Oklahoma, most of the list became Oklahoma schools and the big national online brands mostly dropped out.
How can a college show up more often in AI answers?
Start with the sites the engines read: keep your Niche, US News, and College Board profiles accurate and complete, publish net price and outcomes plainly on your own site for ChatGPT, and test local questions in your own metro, since that's where the national online brands drop out of the answers.
Method and limits
Questions
Four segments with eight questions each, worded the way a student or parent would ask: discovery questions ("What are the best...?"), comparisons ("SNHU vs WGU"), questions about a specific school ("Is the University of Central Oklahoma a good school?"), and requests for alternatives. Rankings on this page use only the open-ended questions where no school is named. Local questions name Oklahoma City; the city comparison repeats four of them for Atlanta, Phoenix, Chicago, and Boston.
Engines and settings
ChatGPT (gpt-5 with web search), Claude (claude-haiku-4-5 with web search), Gemini (gemini-2.5-flash with Google Search grounding), and Perplexity (sonar-pro), called through the API with no system prompt, no memory, and no location settings. API models can differ from the consumer apps.
Runs
All data was collected on October 1, 2026 with Citingly. Open-ended leaderboard questions were asked three times per engine. City, buyer, and location-only questions were asked once or twice per engine, so treat individual percentages in those sections as directional. 16 calls failed after five attempts because of provider rate limits and are left out.
Measurement
Schools were extracted from each answer by a separate model and counted once per answer when the answer recommended them. Spelling variants were merged (for example, "WGU" and "Western Governors University"), and degree types, certifications, financial aid programs, tools, and publications that appeared as names were removed. A hand check of 20 random extractions found every one correctly attributed; five of the 20 were credentials or tools rather than schools, which is why that filter exists. Fact answers were graded against College Scorecard: rates within 5 percentage points, net price within 10%.
Conversations
Scripted three-turn conversations with fixed wording, judged by an automated classifier reading the final answer (yes, conditional, no). No human labeling was done for this edition.
Limits
This is a one-day snapshot. Results come from API calls and don't reflect personalization or chat history in consumer apps. Persona prompts combine buyer type and location, so the location-only control is there to separate the two. Mention rate measures presence in answers, not enrollment outcomes.
Download the data and cite the index
The full rankings by segment and engine are free to download and use with attribution and a link to this page (CC BY 4.0).
Smith, J. (2026). The AI Recommendation Index: Higher Education, Q4 2026. jarredsmith.com. https://www.jarredsmith.com/research/ai-recommendation-index-higher-education
Download the data (CSV) · Get the next edition first
Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services, and he runs AI visibility audits and workshops for marketing teams.