Methodology Overview

To understand how AI search platforms create responses and recommend colleges, we built a corpus of 200,000 student prompts across 100+ programs (undergraduate and graduate) and run them against live AI in all 50 states — covering ChatGPT today, with Gemini, Claude, Perplexity, and Copilot next.

Version 1.0Updated June 2026 · Refreshed every month

The Prompts Dataset

Our prompt database has been curated, deduplicated and organized from four complementary signals: our own State of College Search data showing the actual questions students use during college discovery, Google’s AI Overview keyword dataset, Reddit and Ahrefs.

  • Tier 1 · Core36%
  • Tier 2 · Factors46%
  • Tier 3 · Niche18%
High-Volume Core
A high-volume core query representing the most common way students search for a program. Captures natural phrasing variations of the fundamental “best programs for X” intent.“What are the best universities for a bachelor’s in computer science?”
Decision Factors
A decision-factor query that adds a specific modifier (affordability, career outcomes, accreditation, etc.) to the core program search.“What are the most affordable computer science programs?”
Niche Intents
A program-specific niche query that reflects long-tail, field-specific student intents.“Which CS programs are best for getting into companies like Google or Amazon?”

How We Collect The Data

We extract what major AI Search platforms show to students when they search for college-related prompts across our three tiers. Because AI responses can change from day to day and vary between users, we analyze the semantics behind AI’s responses and average results to ensure consistency.

200,000
Prompts
100+
Program areas
50
States
  1. 1

    Data Collected

    Student questions, keywords and prompts are collected, analyzed, deduped and organized into topics and program areas.

  2. 2

    Tested live on AI

    All prompts are run manually into respective AI Search platforms to assess the response format, quality and consistency — including the response being relevant to respective programs and listing higher ed institutions.

  3. 3

    Prompts set across editions

    Prompts are locked across editions to ensure data consistency.

What We Measure?

What schools does AI recommend, what does it talk about those schools, what programs, how often, what influences AI answers and what is the sentiment.

  • AI Index Score

    Our proprietary 0–100 metric that combines how often a school is recommended in AI answers with how prominently it appears.

  • Citations

    How often a domain is cited as a source in AI answers. Includes both total citation counts and the percentage of prompts where the domain appears.

  • Programs

    What academic programs AI recommends when it recommends a school.

  • Mentions

    When and how often schools appeared in AI answers.

How We Rank Results

We measure results in specific metrics (mentions, citations, sources, topics, sentiment, programs, academic levels) and our proprietary AI Index Score (AIS) which contributes in national and state level rankings.

AI Index Score

0–100

The higher an institution’s AI Index Score (AIS), the stronger its presence is on AI search engines.

Each graph uses specific metrics such as mentions, citations, sources, topics, sentiment, programs, academic levels. We also report citations, sources, queries and ads when returned by AI responses.

  • National Rank

    Schools ranked by their AI Index Score across all states and platforms.

  • State Rank

    Schools ranked by their AI Index Score within their home state.

Managing variability.

AI answers differ between sessions and users. We run every prompt multiple times, analyse the semantics of each response rather than matching exact strings, and average across runs so a single unusual answer cannot move an institution’s score. Prompts stay locked across editions, so month-to-month movement reflects a change in AI behaviour rather than a change in our questions.

Made with Modulify