How We Classified 150+ University AI Policies Into a Single Framework

The Problem: Every Article Covers One School at a Time

If you search "can I use ChatGPT on my college essay," you will find hundreds of articles. Each one covers a single school. Yale says no. Georgia Tech says maybe. Harvard says nothing.

That is not useful when you are applying to eight, ten, or fifteen schools at once. You need to compare policies side by side, but there is no shared vocabulary for doing so. One journalist calls a policy "strict." Another calls the same policy "moderate." There is no underlying rubric, no common scale, and no way to see patterns across the landscape.

We set out to fix that. Over the past year, we reviewed admissions websites, application portals, and official communications for 150+ universities and classified each one using a three-dimensional framework we call L/D/E: Permission (L), Disclosure (D), and Enforcement (E).

This post explains why we chose three dimensions instead of one, what each dimension measures, and what the data reveals.

Why a Single Scale Breaks Down

The obvious approach is to line up schools on a single permissive-to-restrictive axis. We tried this. It failed immediately.

Consider Stanford University. At the institution level, Stanford has no explicit AI admissions policy. They encourage applicants to let their "genuine voice come through," but there is no formal ban, no required disclosure form, and enforcement amounts to standard manual review. Under our framework, that is L0/D1/E1 -- no policy, optional disclosure, soft review.

Now look at the Stanford Graduate School of Business. The GSB states: "It is improper to have another person or tool write your essays. Such behavior will result in denial." That is an outright prohibition with formal consequences. Under our framework, Stanford GSB is L4/D0/E3 -- AI prohibited, no disclosure mechanism (because they forbid it entirely), and formal verification.

Where does "Stanford" land on a single axis? You cannot answer that question. The institution and its business school occupy opposite ends of any permissive-restrictive spectrum. And Stanford is not unusual -- 48 of 166 schools (29%) in our database have at least one program whose AI policy differs from the institution-level classification.

A single score cannot represent both realities within the same institution. That is why we use three independent dimensions.

The Three Dimensions

Each dimension is scored independently. A school can be permissive on one axis and restrictive on another. Higher numbers indicate more restrictive postures.

Permission (L0 through L4): How Much AI Is Allowed?

The L dimension captures what kinds of AI assistance are permitted in application materials.

• L0 -- No Explicit Policy. The school has no admissions-specific statement on AI. A general honesty pledge may exist, but nothing addresses AI tools directly. This is the most common classification: 111 of 166 schools (67%) are L0.

• L1 -- Permissive / Integrative. AI-generated text may be included as long as the content is accurate and the applicant takes responsibility. Schools like Dartmouth, Emory, NYU, Rice, and Penn fall here. 13 schools (8%) are L1.

• L2 -- Line-Level Editing Allowed. AI may paraphrase, suggest edits, and provide style suggestions. Wholesale drafting is not allowed. Georgia Tech is the canonical example: "We believe there is a place for them in helping you generate ideas, but your ultimate submission should be your own." 26 schools (16%) are L2.

• L3 -- Brainstorming Only. AI is allowed for brainstorming, outlining, and basic mechanics like spelling and grammar. No AI rewriting or paraphrasing. Yale is a prominent L3 school: they allow grammar checks but consider AI-composed content "application fraud." 7 schools (4%) are L3.

• L4 -- Prohibited. No AI use of any kind for admissions writing. Georgetown states: "use of AI tools to complete any portion of the application is prohibited." Brown says AI is "not permitted under any circumstances in conjunction with application content." 9 schools (5%) are L4.

Disclosure (D0 through D3): Must You Reveal AI Use?

The D dimension captures whether and how applicants must report their use of AI tools.

• D0 -- No Disclosure Required. No mention of AI disclosure. 113 schools (68%) fall here.
• D1 -- Optional / Encouraged. The school suggests applicants may note how AI helped. Not required, but welcomed. 12 schools (7%) are D1.
• D2 -- Must Disclose AI Use. An explicit prompt or checkbox asks what AI did. 37 schools (22%) require disclosure.
• D3 -- Must Attest No AI Used. Applicants must certify "no AI beyond X" or "no AI at all," with stated consequences. Only 4 schools (2%) require a formal AI-specific attestation, including Georgetown and BYU.

Enforcement (E0 through E3): How Is It Checked?

The E dimension captures the mechanisms a school uses to verify compliance.

• E0 -- No Enforcement Stated. No enforcement mechanism is mentioned. 60 schools (36%) fall here.
• E1 -- Soft / Manual Review. Readers watch for voice consistency. The school may request clarification if something seems off. 62 schools (37%) rely on manual review.
• E2 -- Screening Tools. The school may use AI-detection software, forensic checks, or request timed writing samples. 35 schools (21%) use screening tools.
• E3 -- Formal Verification. Proctored or on-site writing is required. Offers may be contingent on producing an in-person sample that matches the application. 9 schools (5%) have formal verification, including Georgetown and Yale.

Surprising Findings

The most common combination across all 166 schools is L0/D0/E0 -- no policy, no disclosure, no enforcement. 50 schools (30%) have this triple-zero classification. These are schools that have said nothing specific about AI in admissions.

The next surprise: the dimensions really are independent. You might expect that schools banning AI (L4) would also require attestation (D3) and formal verification (E3). But Brown University is L4/D0/E1 -- a full ban with no disclosure requirement and only manual review for enforcement. Meanwhile, some L0 schools still have E2-level screening tools running in the background, even though they have never published a policy.

Schools with program-specific overrides account for 29% of our database (48 of 166). This means that for nearly a third of universities, the institution-level policy is not the whole story.

Georgetown: The Canonical Complex Case

Georgetown University is the most instructive example of why a single score fails. At the institution level, Georgetown is classified L4/D3/E3 -- the most restrictive combination possible. Undergraduate admissions explicitly prohibits AI and requires applicants to attest that no AI was used, with formal verification.

But Georgetown is not monolithic:

• Undergraduate Admissions: L4/D3/E3 -- AI prohibited, attestation required, formal verification.
• Law School: L3/D0/E1 -- brainstorming only allowed; the policy states "Only the applicant should write the actual essays"; no disclosure; soft review.
• Graduate Business Programs: L2/D2/E1 -- line-level editing permitted with the note that "AI may be used for support like brainstorming or grammar checking"; disclosure required; soft review.

A student applying to Georgetown Law and Georgetown's MBA program would encounter fundamentally different rules, even though they are at the same university. This is why we record program-specific overrides and why the three-dimensional framework exists.

How We Collect the Data

Our data comes exclusively from admissions-facing sources: official admissions websites, application portal text, admissions blog posts, and admissions FAQs. We do not use academic integrity pages written for enrolled students unless no admissions-specific guidance exists.

We follow a strict source hierarchy:

1. Admissions-authored AI policy statements -- explicit, purpose-built guidance
2. Application portal language -- text applicants encounter during submission
3. Admissions blog posts and FAQs -- official but less formal
4. General academic integrity policies -- used only as a last resort

Every classification in our database includes the specific source URLs, the exact quotes used to determine the coding, and the date each source was last accessed. You can verify any classification by visiting the linked sources on each school's detail page.

For the full rubric, including confidence levels and edge-case handling, see our methodology page.

Explore the Data

• Browse all 150+ schools -- search, filter, and sort the full database.
• Compare schools side by side -- select two or more schools and see their L/D/E codes together.
• Read the full methodology -- the complete rubric with examples for every level.

---

Related Reading

• Do Top 10 Colleges Check for AI? Official Policies
• AI Policies at T10-T20 Schools
• AI Policies at T21-T32 Schools
• Do Colleges Use AI Detectors? The Truth About Turnitin
• Most Colleges Have No AI Policy — What It Means