How Harvard Admits Opened and Closed Their Main Essays

Nov 8

The Ivy Institute | College Admissions Research | November 2026

Abstract

Fifty of 88 distinct essays (56.8%) opened in a situated moment or setting. Forty-four (50.0%) closed with an interpretation, 31 (35.2%) projected forward and 13 (14.8%) ended chiefly in a scene, image or exchange. Opening and closing language also showed a measurable connection: 63 essays (71.6%) reused at least one eligible word across their first and last 60 words, compared with 23.5% of pairs made by matching different essays. A shared word could mark a meaningful return or simply name a continuing subject. Close readings distinguish the two.

Introduction

An essay can revisit the same object while changing what it means. A collection of toy cars can establish the size of a childhood world on the first page and measure the limits of ownership on the last. An essay can also move forward without returning: a student who once received help as a newcomer can finish by imagining the children he now helps doing the same for someone else. Both forms give a reader a sense that something has happened between the opening and the ending. Only one requires the original object to reappear.

Paley described a rhetorical difficulty of the application essay: a student must disclose something personal to an audience whose evaluative criteria remain partly hidden [3]. Beck and Godley later documented recurring ways of constructing the self across exemplary essays selected by colleges [4]. Here, recurring patterns are measured at three levels: how the writer enters the account, what the ending does, and how often the language at the two ends overlaps.

Publication history and source boundaries

The print anthology 50 Successful Harvard Application Essays began in 1999. Subsequent editions appeared in 2005, 2010, 2014, 2017 and 2024. The separately packaged, sponsored online series can be traced through nine consecutive undergraduate editions, from 2018 through 2026. Its 2018 edition is independently identified in the Crimson’s own sponsorship materials [1–2]. The two formats overlap in content. For example, the 2026 journalism essay published under the heading George begins with the same newsroom assignment and opening passage as the sixth-edition publisher excerpt labeled Neb F. Publication year therefore cannot stand in for application year or establish a writer’s identity.

The unit of analysis was a published essay body. Ten placements were recovered for each online edition. Same-year archival captures supplied 2018–2024, and the publisher’s interactive editions supplied 2025–2026. Repeated words inside a student’s essay were retained. Display pull quotes and reviewers’ summaries were removed. The earlier instance of each of two cross-edition repeat texts was retained, yielding 88 distinct essays.

Methods

Openings. The coder examined the opening movement, including an immediately following line that supplied context for initial dialogue or an image. A situated opening places the reader in an enacted event, a conversation with a particular speaker or a composed physical setting. An imagined scene can qualify when it is presented concretely. A generalized opening starts with a habit, identity claim, definition, question or summary. Quotation alone does not establish a scene: a question that a writer is repeatedly asked can introduce a general predicament.

Endings. The last meaningful movement was assigned one dominant function. An interpretive ending states what an experience means. A forward-looking ending makes an aim, pledge or projected action its payoff. A scene, image or dialogue ending leaves the reader chiefly with an action, a visually rendered moment or an exchange. A passage can combine all three. The code follows the function that supplies its final emphasis. The opening and ending codes were assigned in one AI-assisted analysis, without independent intercoder testing.

Lexical recurrence. The first 60 and last 60 tokens were compared after tokenization. An eligible form contained at least four characters and was outside the fixed English stopword list. A match required the same normalized form in both windows. For a descriptive baseline, each opening was also paired with all 87 endings belonging to other essays. This yields 7,656 directed mismatched pairs. These pairs share the genre and general vocabulary of the collection. They share observations with one another, so they cannot be treated as independent trials. As a further control, the 60-token comparison was restricted to mismatched essays assigned the same primary topic in the companion content analysis. The calculation across all topics was repeated with windows of 40 and 80 tokens. No essay was short enough for its own windows to overlap.

Table 1. How openings and endings combine
Ending	Situated opening	Generalized opening	Total
Interpretation	24	20	44
Forward look	17	14	31
Scene or image	9	4	13
Total	50	38	88

Results

A situated opening followed by an interpretive ending was the largest single combination (24/88, 27.3%). Generalized openings also led to interpretation in 20 cases. Among the 50 situated openings, 17 finished with a forward look and nine closed in a scene or image. An opening scene thus left several endings available. No single combination accounted for even three in ten essays.

Figure 1. Shared words across openings and endings

Percentage of matched windows containing at least one shared eligible form

Same essay · 40 words54.5%

Different essays · 4010.3%

Same essay · 60 words71.6%

Different essays · 6023.5%

Same essay · 80 words83.0%

Different essays · 8037.1%

Baseline openings and endings come from different essays. Both kinds of match become more common when the windows grow.

At 60 tokens, the same-essay match rate exceeded the different-essay rate by 48.1 percentage points (63/88 versus 1,798/7,656). Restricting the mismatched pairs to the same primary topic produced a rate of 24.3% (283/1,166). A shared broad topic explained little of the observed gap, although two essays about family can still describe entirely different people and events. The mean number of shared eligible forms was 1.53 for a true pair versus 0.27 across all mismatches. Increasing the window length raised the baseline as well as the same-essay rate. At all three lengths a writer’s own ending was more likely to share vocabulary with the opening; the size of this difference depends on how much text is inspected.

A second, simpler measure found an explicit form of learn, realize or understand in 72 essays (81.8%). This fixed word list includes selected inflections and is printed in the supplement. A word in this group may appear in dialogue, refer to another person or express a failure to understand. It is evidence of a common vocabulary of learning, not a direct measurement of a change in the narrator.

Three ways a pattern can work

Repetition can enact an idea. The 2026 mowing essay returns to the same sequence of physical movements several times. Early repetitions describe a chore the writer resents. By the end, rehearsing the sequence helps the writer regain perspective without a lawn mower. A repeated phrase is especially easy for the lexical measure to detect. Its changed function emerges only from reading what happens between repetitions.

An image can change its referent. The 2024 zoo essay begins with a child watching an animal. Later the writer is subjected to classmates’ scrutiny. On returning to the zoo image, the narrator can refuse the role of an exhibit. The connection depends on a reversal in who looks and who is looked at. Repeated subject vocabulary alone cannot distinguish this reversal from continued description of a visit.

A decision can complete a scene. In the 2020 account of a bus journey to teach, the writer has reason to turn back after a co-teacher becomes ill. The bus moves toward the school while the writer weighs that decision. A short exchange at the destination resolves the question. The ending works through an answer given under pressure. A general statement about commitment would perform a different rhetorical function. In the 2020 newcomer essay, by contrast, the ending imagines a future chain of assistance. Both essays concern helping others, yet their resolutions differ.

Interpretation and limits

The same-essay baseline comparison demonstrates continuity of language over the span of a text. It cannot establish that 71.6% of writers consciously planned a circular ending. Topic continuity, a repeated character and an intentional refrain can all produce the same positive result. Conversely, a writer who returns to an image through synonyms may receive a negative result. Even the same-topic baseline varies in author, event and specific subject matter. It cannot isolate the effect of deliberate structure.

The three ending categories compress mixed cases. For example, a future ambition expressed through a concrete image may deserve a different code from a second reader. If both repeat placements are restored, interpretive endings account for 45 of 90 placements, the same 50.0% observed after deduplication. The evidence supports a modest conclusion about this series: opening and closing material frequently connects, and the connection can be accomplished through reflection, projection or action. The archive contains no reader scores from which to infer which arrangement Harvard rewarded.

Sources and disclosures

Staff of The Harvard Crimson. 50 Successful Harvard Application Essays. St. Martin’s Griffin. Edition records: 1999, 2005, 2010, 2014, 2017, 2024.
The Crimson Brand Studio and Crimson Business. Undergraduate successful-essay features, edition-specific sources: 2018 · 2019 · 2020 · 2021 · 2022 · 2023 · 2024 · 2025 · 2026. 2026 Crimson host · later 2022 edition page · Crimson sponsorship prospectus.
Paley, K. S. (1996). The college application essay: A rhetorical paradox. Assessing Writing, 3, 85–105. Publication record.
Beck, S. W., & Godley, A. J. (2023). “What makes you, you”: The discursive construction of the self in US college application essays. American Journal of Education, 129, 539–564. Publication record.

Study disclosure. Prepared for The Ivy Institute, which contributed professional reviews to the Crimson series, including in 2024 and 2025. Sponsor reviews, applicant profiles and advertising were excluded from the evidence. The study used publicly displayed texts.

Data availability. The accompanying research supplement includes the source register, edition and duplicate identifiers, essay-level codes, full word-frequency table, feature definitions and reproduction scripts. Source links above lead to the original publications or their archived pages. All counts describe the specified editions as reconstructed on November 8, 2026.