Halo SEO

AI Visibility & Generative Search Readiness

A prospective student no longer types a college name into Google and reads ten blue links. They ask ChatGPT or Perplexity, or read Google’s AI Overview, and they act on whatever the answer says. This report measures what those engines say about 124 NIRF-ranked colleges: whether each college is mentioned, how accurately, and with what tone. The answer is uneven, and most colleges have no idea where they stand.

Cohort 124 NIRF colleges

Data April 2026

Engines ChatGPT × 3 + Google AI Overviews

Verified 22 May 2026

69%

Colleges ChatGPT never mentions from memory alone.

120/120

Colleges Google AI Overviews reach at least once.

1/120

Colleges an AI engine gave an accurate NIRF rank.

0/120

Colleges AI engines described with negative sentiment.

What feeds this report

The cohort is 124 NIRF-ranked autonomous and privately-managed Indian colleges, drawn from the NIRF India Rankings College, Engineering and Management categories. You can look up any institution in the full NIRF college list. Each was measured by a 22-stage research pipeline. This report draws on the AI visibility stage, which puts the same ten standardised prospective-student prompts (general information, reputation, programs, admissions, campus, fees, faculty, alumni, student reviews and peer comparison) to four AI answer engines and records how each one responds.

We chose the four engines to test different things. ChatGPT knowledge-only (gpt-4o-mini, no web access) tests what the model recalls from training. ChatGPT web-grounded (gpt-4o-search-preview) tests the same model when allowed to search. Real ChatGPT, scraped live from the consumer product on gpt-5, tests what a student actually sees today. Google AI Overviews tests the answer block now sitting above Google’s organic results. Sentiment on every mention is scored with VADER. The denominator is 120 colleges with a complete AI visibility return; the master enrichment file carries all 124. Coverage is stated on every chart, and every figure is recomputed from the April 2026 dataset. Nothing is estimated.

One distinction matters throughout. AI visibility, the subject of this report, is the outcome: whether an answer engine surfaces and recommends a college. AI-readiness, the subject of the companion technology and performance report, is the cause: whether the college website carries the schema and structured signals an AI crawler can read. This report measures the first.

124

NIRF-ranked colleges in the cohort

120

Colleges with a complete AI visibility return

4

AI answer engines probed per college

4,800

AI responses scored (120 colleges × 10 prompts × 4 engines)

Nine things the data makes plain

Each finding below is computed from the April 2026 AI visibility dataset. Every chart is interactive. Hover any bar, segment or dot for the underlying figure, and each chart animates as you scroll it into view.

Section 01 · The Visibility Gap

69% of colleges are invisible to ChatGPT’s memory

Ask ChatGPT about a college with web access switched off, and it answers only from what it learned in training. On that test, 83 of 120 colleges (69%) are never mentioned across ten standard prospective-student questions. Only 5 colleges are named in all ten; another 27 are named in five to nine. For most NIRF colleges, the model has no stored knowledge to draw on at all.

The visibility score, a 0–100 measure blending coverage, mention volume and sentiment, runs from 37 to 100 around a median of 54. That median is not a healthy middle: a college scores 54 by appearing in Google AI Overviews and nowhere else. The distribution is bimodal. A large cluster of low-visibility colleges rides entirely on web grounding, while a smaller group of well-known names is one ChatGPT has actually memorised. The same blind spot in the website signals behind this gap is mapped in why most education institutions are not ready for AI search.

83

Colleges ChatGPT never names from memory alone

5

Colleges ChatGPT names in all 10 prompts from memory

37–100

Full AI visibility score range across 120 colleges

54

Median AI visibility score (a web-only baseline)

AI visibility score: how the cohort distributes

120 colleges grouped by AI visibility score band · score blends coverage, mention volume and sentiment, 0–100.

69%

Of the 120 colleges, 83 are never mentioned by ChatGPT when it answers from training knowledge alone. For two-thirds of the NIRF cohort, the model has no memory of the institution. It can only describe the college if a live web search puts the page in front of it.

Section 02 · Four Engines

Four engines, two different worlds

We measured four AI engines. Three of them are ChatGPT, run in three different ways, and the difference between those three carries the central finding of this section, so it is worth reading slowly.

The first is ChatGPT answering from trained memory with browsing switched off: it replies only from what it absorbed during training, with no access to the live web. The second is ChatGPT’s web-search model, the version that looks things up before it answers. The third is the real ChatGPT product, the one a student actually opens at chatgpt.com, which decides on its own when to browse the web. The fourth engine, Google AI Overviews, is the answer block now sitting above Google’s organic results.

The reason all three ChatGPT modes were measured is the spread between them. ChatGPT’s trained memory barely knows this cohort: it reaches only 37 of 120 colleges, a 31% reach. The real ChatGPT product reaches 118 of 120, a 98% reach. The same model, asked the same questions, returns an almost-empty answer in one mode and a detailed one in another, because the real product browses the live web in real time and the trained memory does not. A college’s AI visibility today therefore rests almost entirely on its live website, not on being “known” by the model. That is visibility the college does not own. One product change at OpenAI, and it can be lost.

ChatGPT
From trained memory, browsing off
37 /120
Colleges mentioned at least once · 31% reach
✗ Low reach
ChatGPT
The web-search model
37 /120
Colleges mentioned at least once · 31% reach
✗ Low reach
ChatGPT
The real product at chatgpt.com
118 /120
Colleges mentioned at least once · 98% reach
✓ Full reach
AI Overview
Google answer block
120 /120
Colleges mentioned at least once · 100% reach
✓ Full reach

Mention depth: average prompts that name the college, by engine

120 colleges · weighted average from the same 10-prompt set per engine. Higher means the engine names the college on more of the 10 questions.

Section 03 · By Question

Where ChatGPT’s memory holds, and where it fails

Even for colleges ChatGPT does recall, its memory is patchy by topic. We fixed the engine at knowledge-only mode and counted, per question, how many of the 120 colleges it names. The model is most willing to talk about programs, financial aid, student reviews and campus infrastructure, naming around 32 to 34 colleges for each. It is far weaker on general information (8 colleges) and peer comparison (15). Those are the open-ended questions where the model has no recruiter list or fee table to anchor a confident answer.

The pattern matters because the weak questions are the ones a prospective student asks first: “tell me about this college”, “how does it compare to others I’m considering”. When ChatGPT cannot answer those from memory, the student gets a generic deflection or, worse, a confident answer assembled from a college with a similar name. The fix is the same structured-content work that drives the wider GEO and AI search readiness audit.

ChatGPT knowledge-only: colleges named, by question type

120 colleges · ChatGPT answering from training memory with no web access · bar shows how many of the 120 colleges it names for each of the 10 questions.

Section 04 · Sentiment

Sentiment runs uniformly warm, and that is the problem

Every one of the 120 colleges carries a positive overall sentiment label. Not one is described negatively. On the surface that reads as good news; in practice it reads as noise. When an answer engine is positive about every institution in a category, the tone carries no information, and a student cannot use it to choose. The mean VADER compound score is 0.929, and 62 colleges sit above 0.95, the band of glowing, almost promotional language.

The spread underneath the label is where the signal sits. 39 colleges land in the measured 0.80–0.90 band, and one slips below 0.80. The doughnut below splits the cohort by tone. The colleges in the warmer bands are not better. As Section 07 shows, they are simply less known, so the engine has nothing concrete to temper its praise with.

120/120

Colleges with a positive sentiment label, none negative

0.929

Mean VADER compound sentiment across the cohort

62/120

Colleges scoring above 0.95, with glowing language

Sentiment tone: how warm the AI description runs

120 colleges · all labelled positive · split by VADER compound score band of the average mention.

Section 05 · Accuracy

The fact gap: warm words, no numbers

A warm description is not a useful one. A student deciding where to apply needs specifics: the NIRF rank, the average placement package, the fee, who recruits on campus. We checked, across all 120 colleges, how often an AI engine volunteered each of those facts. The result is stark. An engine gave a specific NIRF rank for 1 college, an average placement figure for 0, a fee figure for 0, and a recruiter list for 2. It is more willing with softer facts, naming notable alumni for 12 colleges and top programs for 29, but the hard numbers a decision turns on are almost entirely absent.

The engine is not being cautious here. It simply has nothing to cite. Where a college’s own site publishes placement and ranking data in a structured, readable form, an answer engine repeats it. Where the site does not, and almost none do, the engine fills the space with general praise instead. It is the same gap behind the question of whether an institution is ready for AI search, and the structured-content work that closes it is the core of SEO for universities and colleges.

Which facts the AI volunteers, and which it skips

120 colleges · bar shows how many colleges an AI engine gave a specific fact for, out of 120, across all four engines and ten prompts.

1/120

Colleges given an accurate NIRF rank by an AI engine

0/120

Colleges given a placement or fee figure

29/120

Colleges with at least one program named

Section 06 · Category Divide

The category divide: famous arts colleges win the memory test

Split by NIRF category, AI visibility does not follow the pattern of the technical report, where Management institutes led. Here the 89 general Colleges average 65.9 on the AI visibility score, ahead of the 8 Engineering colleges at 64.2 and the 23 Management institutes at 52.8. The reason is reputation, not website quality. The cohort’s Colleges include long-established Delhi and Chennai arts names that ChatGPT has genuinely memorised: they average 2.9 knowledge-only mentions of 10, while Engineering and Management colleges average effectively zero. On the live-web engines, real ChatGPT and AI Overviews, the categories are near-identical, all reaching 9.6 to 10 prompts of 10. The divide is a memory gap.

65.9

College (n=89), mean AI visibility score

64.2

Engineering (n=8), mean AI visibility score

52.8

Management (n=23), mean AI visibility score

College vs Engineering vs Management: AI visibility compared

Group means by NIRF category over the 120-college AI visibility return · College n=89, Engineering n=8, Management n=23.

Section 07 · Visibility vs Tone

The better-known a college, the cooler the tone

A reader might expect the most-visible colleges to also be the most warmly described. The data shows the opposite. We plotted every college’s ChatGPT memory mentions against its sentiment score, and the result is a clear downward slope, with Pearson r = −0.79. The 20 most-visible colleges average 0.858 on sentiment; the 83 least-visible average 0.956.

The mechanism is straightforward. When ChatGPT actually knows a college, it knows the texture: the competition, the workload, the infrastructure complaints. A balanced description like that scores lower on a sentiment scale. When the model knows nothing, the only safe answer is bland praise, which scores high. An AI engine volunteered a critical concern for all 20 of the most-visible colleges, against 28% of the cohort overall, most often infrastructure or administrative issues. High visibility is not a free win. It brings scrutiny, and the only way to shape that scrutiny is to publish the accurate picture first.

−0.79

Pearson correlation: ChatGPT memory vs sentiment

0.858

Mean sentiment of the 20 most-visible colleges

20/20

Most-visible colleges with an AI-flagged concern

Memory vs warmth: one dot per college

120 colleges · x = ChatGPT knowledge-only mentions (of 10), y = average mention sentiment · Pearson r = −0.79.

Section 08 · Description Depth

For most colleges, the AI has an empty profile

Behind every AI answer is a structured profile the engine assembles, covering the established year, rank, programs, recruiters, alumni and fees. We scored how complete that profile is for each college. The result mirrors the fact gap. For 83 of 120 colleges the AI built a 0% profile: it mentioned the college but could attach not one specific structured fact to it. Only 2 colleges reach a 100% profile, and the cohort median is 0%.

An empty profile is what produces a fluent, friendly, and almost contentless answer. The college is named, praised, and then the student is none the wiser about whether to apply. The 32 colleges with a half-complete or better profile are, predictably, the same well-known names from Section 07, and even they leave most fields blank. A complete, structured profile is something a college can supply directly. The path to it runs through AI marketing for education institutions.

83/120

Colleges with a 0% AI structured profile

2/120

Colleges with a fully complete AI profile

19.7%

Mean structured-fact completeness across the cohort

AI structured-fact completeness: where the cohort sits

120 colleges grouped by how complete a structured fact profile the AI could assemble, 0–100%.

Section 09 · Leaderboard

The leaderboard

Ranked by AI visibility score, the 0–100 measure blending engine coverage, mention volume and sentiment, the 120-college return runs from 100 at the top to 37 at the bottom. Hindu College, Lady Shri Ram College, Miranda House and Shri Ram College of Commerce share the top, each scoring a perfect 100. The top ten are colleges ChatGPT has genuinely memorised: every one is named in 9 or 10 of the 10 knowledge-only prompts.

The bottom ten we do not name. The scores are real; the colleges are masked. A score in the high 30s or 40s means the college lives almost entirely in Google’s AI Overview and has no presence in ChatGPT’s memory at all. If you want to know whether your institution sits in that group, request your scorecard and we will tell you privately.

RankCollegeVisibilityChatGPT memorySentiment
1Hindu College, Delhi10010 / 100.883
2Lady Shri Ram College for Women, New Delhi10010 / 100.899
3Miranda House, Delhi10010 / 100.880
4Shri Ram College of Commerce, Delhi10010 / 100.872
5Atma Ram Sanatan Dharm College, New Delhi959 / 100.858
6Kirori Mal College, Delhi959 / 100.864
7Loyola College, Chennai959 / 100.879
8Madras Christian College, Chennai959 / 100.847
9PSGR Krishnammal College for Women, Coimbatore959 / 100.830
10Sri Venkateswara College, Delhi959 / 100.850
RankCollegeVisibilityChatGPT memorySentiment
111540 / 100.974
112540 / 100.986
113540 / 100.833
114510 / 100.920
115510 / 100.982
116510 / 100.935
117470 / 100.974
118470 / 100.893
119440 / 100.966
120370 / 100.977

We don’t name the bottom 10. The scores are real; the colleges are withheld on purpose. To find out where your college ranks, in the bottom group or anywhere else, request your scorecard and we will tell you privately.

Ranked across the 120-college AI visibility return by AI visibility score, the 0–100 measure blending engine coverage, mention volume and sentiment, computed from the April 2026 dataset and verified 22 May 2026. Several colleges tie on score; ties are ordered alphabetically.

Every measure, side by side

A scorecard of every headline figure in this report, grouped by theme. Each row carries the cohort value, an inline bar reading it against its own scale, a status pill, and the denominator it is computed against. Coverage is the 120-college AI visibility return unless stated otherwise.

MetricCohort valueReadingStatusCoverage
Engine reach
ChatGPT knowledge-only reach37 / 120
Failing120 colleges
ChatGPT web-grounded reach37 / 120
Failing120 colleges
Real ChatGPT (consumer) reach118 / 120
Healthy120 colleges
Google AI Overviews reach120 / 120
Healthy120 colleges
AI visibility score
AI visibility score (median)54 / 100
Weak120 colleges
AI visibility score (range)37–100
Neutral120 colleges
Colleges scoring 85+ (well-known)20 / 120
Weak120 colleges
Colleges at the web-only baseline (50–54)79 / 120
Failing120 colleges
ChatGPT memory
Colleges invisible to ChatGPT memory83 / 120
Failing120 colleges
Colleges ChatGPT recalls in all 10 prompts5 / 120
Failing120 colleges
Avg ChatGPT memory mentions (of 10)2.1 / 10
Failing120 colleges
Sentiment
Colleges with a positive sentiment label120 / 120
Neutral120 colleges
Mean VADER compound sentiment0.929
Healthy120 colleges
Memory × sentiment correlation (Pearson r)−0.79
Weak120 colleges
Most-visible colleges with an AI-flagged concern20 / 20
Failing20 most-visible
Fact accuracy & depth
Colleges given an accurate NIRF rank1 / 120
Failing120 colleges
Colleges given a placement or fee figure0 / 120
Failing120 colleges
Colleges with a 0% AI structured profile83 / 120
Failing120 colleges
Mean structured-fact completeness19.7%
Failing120 colleges

Status pills read each metric against its good direction: Healthy meets the mark, Weak is borderline, Failing misses it; Neutral marks a structural or descriptive figure with no good or bad direction. The inline bar shows the value against its own scale. AI visibility figures are computed from the April 2026 AI visibility dataset (120-college return) and verified 22 May 2026.

Four moves that compound

Each of these is a defined content or technical change, not a strategy. Done in order, they give every AI engine the same accurate facts to repeat. They sit alongside the wider Thrivemattic research programme on how Indian institutions market themselves.

01

Publish the facts the AI keeps skipping

An AI engine volunteered a placement figure for 0 colleges and an accurate NIRF rank for one. It cannot cite what the site does not state plainly. Put the rank, average and highest package, fee, recruiter list and entrance exams in clean, structured text on the admissions and placement pages, so every engine has the same numbers to repeat.

02

Treat ChatGPT’s blind spot as the priority

Real ChatGPT and Google AI Overviews already reach nearly every college through live web search. The gap is ChatGPT’s memory, where 69% of the cohort is invisible. Consistent, well-structured authority content is what eventually moves a college from web-only visibility into the model’s recalled knowledge.

03

Get ahead of the scrutiny that comes with visibility

Every one of the 20 most-visible colleges had a critical concern volunteered by an AI engine, usually infrastructure or administration. Visibility brings scrutiny. Publish the honest, current picture of campus facilities and student support first, so the engine cites your account rather than an old forum thread.

04

Check what four engines actually say about you

The four AI engines disagree sharply. A college can be invisible in one ChatGPT mode and described in detail in another. Run the same ten prospective-student prompts through each engine, record what comes back, and fix the wrong or missing facts at the source. This is the AI-search work covered by SEO for universities and colleges.

Request your institution’s AI visibility scorecard

The scorecard covers the AI visibility dimensions in this report and shows exactly where your institution sits against the cohort, including, privately, whether you are in the bottom ten.

It includes:

  • How each of the four AI engines mentions your college
  • Your ChatGPT memory reach vs the 31% cohort baseline
  • The facts AI gets right, gets wrong, or skips entirely
  • Your sentiment tone and any AI-flagged concerns
  • Your AI visibility rank within the 124-college cohort

    Replies come from [email protected] within five working days. Tier preference is interest only, not a commitment.

    Free for any of the 124 institutions in this study.

    Common questions about this report

    Questions we hear from college principals, directors, and marketing teams about the AI visibility findings.

    Does ChatGPT know about NIRF-ranked colleges?

    Mostly not, from memory alone. When ChatGPT answers from its own training knowledge with no web access, 83 of 120 colleges (69%) are never mentioned across ten standard prospective-student prompts. Only 5 colleges are named in all ten. The moment the same question triggers a web search, coverage jumps: real ChatGPT scraped live mentions 118 of 120 colleges and Google AI Overviews mention all 120. AI visibility for a college is now decided by web grounding, not by what the model remembers.

    What is the difference between AI visibility and AI-readiness for a college website?

    AI-readiness measures whether a college website carries the structured signals (JSON-LD schema, an llms.txt file, clean headings) that AI crawlers can read. AI visibility measures the outcome: whether AI answer engines actually surface and recommend the college when a student asks. They are related but distinct. This report measures AI visibility; the companion technology and performance report in the same study measures AI-readiness.

    How accurately do AI answer engines describe NIRF colleges?

    They describe colleges warmly but vaguely. Across 120 colleges, an AI engine volunteered an accurate NIRF rank for just 1, a placement figure for 0, and a fee figure for 0. It named notable alumni for 12 and listed top programs for 29. Where the college’s own site publishes nothing structured, the engine fills the gap with general praise rather than the specific facts a student needs to decide.

    Do AI answer engines say anything negative about NIRF colleges?

    Not in label, but yes in tone. Every one of the 120 colleges carries a positive overall sentiment label. The deeper signal is that sentiment cools as visibility rises: the 20 most-visible colleges average 0.858 on the VADER scale against 0.956 for the 83 least-visible, and an AI engine volunteered a critical concern, most often about infrastructure or administration, for all 20 of the most-visible colleges. More AI attention means a more measured, more critical description.

    Which AI engine should a college worry about most?

    The one a prospective student is most likely to open. Google AI Overviews and real ChatGPT already reach nearly every college through live web search, so the immediate priority is making sure the facts they surface are accurate. ChatGPT’s knowledge-only mode is the long game: 69% of the cohort is invisible there, and only consistent, well-structured authority content moves a college into the model’s recalled memory.

    How can I see my college’s AI visibility scorecard?

    Request your scorecard using the form on this page. We send a per-college evaluation covering how each AI engine mentions your college, the accuracy of the facts it volunteers, your sentiment tone, your AI visibility score against the cohort, and the prioritised fixes. It tells you, privately, whether you sit in the bottom ten. Free for any college in the 124-college study.

    See exactly what AI says about your college

    Request a per-college scorecard: how each of the four AI engines describes your college, the facts it gets right or skips, your sentiment tone and your AI visibility rank. Free for any college in the 124-college study.