Last updated July 2026.
According to a fall 2025 EAB survey of more than 5,000 high school students, 46% of prospective students used AI in their college search, up from 26% in spring 2025. In less than one academic year, AI went from a minority tool to the majority channel for early-stage college research.
The same survey found that 18% of those students removed a college from consideration based on information surfaced through AI-generated search results. That figure deserves to sit in front of every enrollment marketing leader: nearly one in five students made an elimination decision based on what an AI engine told them, before any admissions professional had a chance to make a case.
(Note: both figures come from EAB, an education technology and research company. This is a single vendor’s self-published survey, not independent academic research.)
This is the stealth applicant effect in numbers. It is not a website traffic problem. It is not a rankings problem. It is an AI brand visibility problem, and it requires a different kind of audit.
Why traditional enrollment metrics miss this entirely
Your enrollment team tracks website sessions, campus visit registrations, RFI form fills, and FAFSA completions. All of those metrics measure applicants who made it into your funnel. None of them measure the students who consulted an AI engine, received an answer that did not include your institution, and moved on.
The prospective student who asks ChatGPT “best MBA programs in the Midwest for working professionals” and receives a response naming five universities, none of which is yours, is invisible to you. They did not bounce from your homepage. They never reached your homepage. Their session never happened.
Traditional share of voice metrics do not capture this either. You might rank well in Google organic search for those same keywords. The student who went to an AI engine first never saw that ranking.
The AI visibility problem is a pre-funnel problem. The audit has to match.
Step 1. Define your program-comparison prompt set
The starting point is the specific questions your prospective students type into AI engines. These are not the same as your SEO keyword list. AI prompts tend to be longer, more comparative, and more intent-rich.
For each program you want to protect, build a set of four to six prompt variants that reflect real student language. Here is a working template:
| Query intent | Example prompt | Notes |
|---|---|---|
| Program comparison, general | ’Best [program] programs for [student profile]‘ | Widest competitive exposure |
| Head-to-head comparison | ’Compare [your institution] vs [peer 1] vs [peer 2] for [program]‘ | Tests direct brand presence |
| Outcome-focused | ’Which schools have the best [outcome] rate for [program]?’ | Pulls in rankings and accreditation signals |
| Cost and aid comparison | ’Cheapest accredited [program] with [feature]‘ | Exposes if you appear in value discussions |
| Location-specific | ’Best [program] in [city/region] for [student profile]‘ | Tests local competitive set |
| Accelerated or format-specific | ’Online [program] programs with [specific feature]‘ | Format-driven queries are fast-growing |
Build this table for each of your three to five most strategically important programs. That gives you 18 to 30 prompts per program as your core tracking set.
Step 2. Run your baseline audit across AI engines
Run each prompt on the four engines that matter most for student research: ChatGPT, Perplexity, Google AI Overviews, and Gemini. AI outputs are probabilistic, so run each prompt at least five times per engine before recording a result. A single run is noise.
For each response, record:
- Is your institution named?
- Where in the response does it appear (first mention, secondary list, footnote, or absent)?
- Which peer institutions appear?
- Is the information about your institution accurate?
- Is the sentiment neutral, positive, or negative?
The output of this step is a baseline share-of-voice table. Here is a simplified example:
| Program-comparison prompt | ChatGPT | Perplexity | AI Overviews | Gemini | Your presence |
|---|---|---|---|---|---|
| Best MBA for working professionals, Midwest | MIT, Northwestern, Indiana | Notre Dame, Michigan, Ohio State | Indiana, Notre Dame, Michigan | Michigan, Northwestern, Ohio State | Absent on all 4 |
| Online MBA accredited programs, flexible schedule | Penn State, Arizona State, Indiana | ASU, Penn State, Florida | Penn State, ASU, Indiana | Penn State, Indiana, UNC | Absent on all 4 |
| Compare [your institution] vs [peer 1] [program] | Peer 1 named, you omitted | Head-to-head answer given, omits you | Peer 1 discussed, you in footnote | Full comparison, balanced | Weak: present on 1 of 4 |
A table like this, built across your full prompt set, makes the stealth applicant effect visible for the first time. You can see exactly which programs have the largest coverage gap, which engines are most likely to omit you, and which peers are consistently named instead.
Your methodology for this audit should be documented so results are comparable quarter over quarter.
Step 3. Diagnose why you are absent
Being absent from an AI answer has a small number of root causes. Diagnosing which one applies to each gap saves you from wasting effort on fixes that do not match the problem.
Cause 1: Thin source material. AI engines pull from third-party sources (rankings sites, media coverage, review platforms, accreditation bodies) far more than from your own website. If those sources do not describe your program in detail, the model has little to draw on.
Cause 2: Inaccurate or outdated information in the retrieval layer. The AI engine may have ingested old information about your program. A program that changed its format, cost, or requirements in the past two years may be described inaccurately, causing the model to deprioritize it or contradict your current positioning.
Cause 3: Competitor content dominates the retrieval sources. Your peer institutions may have published detailed, structured, comparison-ready content that the AI engine has indexed heavily. Your program pages, by contrast, may be written for human readers rather than structured for AI retrieval.
Cause 4: Technical crawlability. AI crawlers are distinct from Googlebot. If your site uses heavy JavaScript rendering, CDN security rules that block non-browser user agents, or robots.txt configurations that restrict AI crawlers, your content may not be reachable by the retrieval layer at all.
Run a crawlability check against the major AI crawler user agents as part of your audit. Fixing a crawl block is the fastest lever in the playbook.
Step 4. Choose your measurement stack
Manual audits answer the baseline question. Ongoing share-of-voice tracking requires tooling.
Temso is the easiest all-in-one starting point for higher-ed marketing teams. At $89/mo, it tracks share of voice, brand mentions, citations, and sentiment across eight AI engines, including ChatGPT, Perplexity, Google AI Overviews, Gemini, Google AI Mode, Grok, Microsoft Copilot, and Meta AI. Its built-in workflow converts visibility gaps into a prioritized action queue and executes content and citation fixes inside the same subscription. For an enrollment marketing team without a dedicated AEO analyst, the guided setup and integrated action plan are the key differentiator.
Peec AI is a strong choice for institutions running multi-program or multi-campus audits. It offers daily prompt-level tracking across nine-plus engines, unlimited user seats on every plan, and an Actions feature that converts gaps into a prioritized execution queue. Starting at 85 euros per month, it fits teams that need to track a larger prompt set across multiple program families and want to share access across enrollment, marketing, and communications staff without per-seat fees.
Otterly.AI is the most accessible entry point for teams with a tight budget. Its Lite plan starts at $29/mo and covers prompt-level citation tracking across six platforms, with a structured GEO Audit Engine across 20-plus on-page factors. For institutions just starting to build an AI visibility practice, Otterly.AI’s G2 High Performer rating and structured audit guidance make it a credible starting point.
Semrush offers complementary capability for teams already in the platform: AI Overviews tracking at the keyword level, content optimization guidance, and competitive benchmarking. It does not replace dedicated AI visibility tooling for multi-engine share-of-voice measurement, but it is a useful cross-check on keyword-level AI Overview presence for programs with a strong SEO foundation.
The full ranked comparison of tools is at /rankings/ai-visibility-tools.
Step 5. Prioritize your fix queue
Once you have a baseline and a diagnosis, the fix queue follows a clear priority order.
Fix 1: Earn citations on the sources AI engines trust most. For higher education, those sources are rankings publications (US News, QS, Times Higher Education, Princeton Review), independent program review sites, regional and national news coverage, and professional accreditation bodies. A citation on a source the AI engine already treats as authoritative carries more weight than a page update on your own website.
Fix 2: Rewrite program pages for AI retrieval. Program pages written for human readers tend to bury the most important information in paragraph form. AI retrieval rewards structured, factual, directly comparative content. For each program, create a dedicated page that answers: total cost, format, length, accreditation status, outcome data, application deadlines, and distinguishing features versus named peers. Use headers, short paragraphs, and fact tables. Put the most important information in the first 30% of the page.
Fix 3: Correct inaccurate information in the retrieval layer. If an AI engine is describing your tuition incorrectly, stating an outdated accreditation status, or misrepresenting your program format, the fix starts with the third-party sources the engine is drawing from. Update those sources directly. Corrections to your own website matter less than corrections to the authoritative sources the model has already indexed.
Fix 4: Fix crawlability. Audit your robots.txt for AI crawler restrictions. Test your program pages against the major AI crawler user agents. Verify that JavaScript-heavy pages render fully for non-browser agents. A crawl block is the fastest fix in the stack and the one most often missed.
Step 6. Track drift, not snapshots
A one-time audit tells you where you stand. Ongoing drift measurement tells you whether your fixes are working.
Set up weekly tracking on your core prompt set. Each week, record your share of voice across the four primary engines for each program family. The signal you are looking for is a sustained upward trend over eight to twelve weeks. Single-week spikes are noise. A consistent directional shift is signal.
Track these metrics side by side:
- AI share of voice: percentage of prompts in which your institution is named, versus the competitive set
- Citation rate: how often your program pages are used as sources in AI-generated answers
- Accuracy rate: percentage of responses in which factual claims about your institution are correct
- Sentiment: whether descriptions of your institution are neutral, positive, or negative
Week-over-week drift is the headline metric. Absolute share-of-voice numbers tell you where you stand in the competitive set. Direction tells you whether you are winning or losing ground.
The audit at a glance
| Step | Action | Output |
|---|---|---|
| 1 | Build program-comparison prompt set | 18 to 30 prompts per priority program |
| 2 | Run baseline audit, 5 runs per prompt per engine | Share-of-voice baseline table |
| 3 | Diagnose root cause of gaps | Source gaps, accuracy errors, crawl blocks |
| 4 | Select measurement stack | Ongoing weekly tracking in place |
| 5 | Execute fix queue (citations, content, crawl) | Prioritized action plan by program |
| 6 | Track week-over-week drift | Trend data over 8 to 12 weeks |
What to do this week
Start with one program. Pick the program most important to next cycle’s enrollment goals. Build its prompt set (six prompts across the query intents above). Run each prompt five times on ChatGPT and Perplexity. Record which institutions appear and whether yours does.
That audit takes two to three hours and produces the data you need to make the case internally that the stealth applicant effect is real and measurable at your institution.
Once you have the baseline, set up automated tracking in Temso or Peec AI so you stop doing this manually each week. Then move to the fix queue.
The students are already asking AI engines which programs to consider. The question is whether your institution appears in the answer.
For the full ranked list of AI visibility tools, including pricing and engine coverage, see /rankings/ai-visibility-tools. For definitions of share of voice, citation rate, and sentiment, see the /glossary.