Letting AI write your school assignment or job application vs writing it yourself
Last reviewed 2026-05-28
Evidence quality 4.0/5
Eight-dimension review score against the
quality rubric
. Each dimension scored 1–5.
D1 Source verification
5/5
D2 Source authority & independence
4/5
D3 Regret-rate accuracy
2/5
D4 Source comparability
3/5
D5 Gilovich pattern
4/5
D6 Prose quality
4/5
D7 Caveat completeness
5/5
D8 Sample quality
5/5
Average4.0/5
Proxy data — no direct regret survey exists for this decision. Rates are derived from satisfaction scores and access-barrier data rather than questions that directly asked about regret. See caveats below.
Action regret
Using AI to write the assignment
60%
~60% of student AI users agreed AI for schoolwork harms critical thinking (Dec 2025 proxy); 11% of ChatGPT-using job seekers were denied a job after the employer discovered it
US students ages 12–29 who use AI for schoolwork; separately, US adult job seekers who used ChatGPT for resumes/cover letters
December 2025 (RAND); February 2023 (ResumeBuilder)
Inaction regret
Writing it yourself
18%
~18% (proxy; estimated from share of non-AI students who report concern about falling behind classmates who use AI; no direct survey)
US students ages 12–29 who do not use AI for schoolwork
2024–2026
% who regret this choice
Using AI to write the assignmentWriting it yourself
60%18%
action dominates — Action dominates — most regret acting.
Related decisions
Semantically similar decisions — same territory, different trade-offs.
About 60% of US student AI users agreed that using AI for schoolwork harms critical thinking, according to the RAND American Youth Panel’s December 2025 wave (n=1,214 youth ages 12–29). The headline figure across all students is higher — 67%, up from 54% earlier in 2025 — but the gap between non-users (78% concerned) and actual users (60%) is the more informative number: even people who use the tool report substantial concern about what it is doing to them. Use kept rising anyway, from 48% to 62% of students in seven months, which is the cleanest revealed-preference signal that observed concern coexists with low behavioral regret. On the job-application side, a ResumeBuilder survey of 1,000 US job seekers found that 78% got an interview after submitting ChatGPT-written application materials and 59% were hired, but 11% were denied a job once the interviewer realised AI had written the application — the cleanest concrete-consequence number on this side of the ledger.
The evidence base for actual harm is stronger than the evidence base for actual regret. A field RCT by Hamsa Bastani and colleagues at Wharton, conducted with about 1,000 Turkish high-school math students, found that students given unrestricted ChatGPT-4 access during practice solved 48% more problems correctly during sessions but then scored 17% lower than the control group on a subsequent no-AI exam. The crucial follow-up finding: students did not recognise that they had learned less. If the harm is invisible to the person doing it, self-report regret will systematically understate true regret on a skill-development axis — students who lost skill cannot regret a loss they cannot see. Surveys of college applicants (foundry10, 2024) found about a third had used AI for admissions essays, with 6% having AI write the final draft; 72% of student ChatGPT users in the Intelligent.com survey agreed that using ChatGPT to write college essays counts as cheating, indicating moral discomfort that runs alongside continued use.
The inaction side is genuinely thin. No published survey directly asks non-AI-using students whether they regret writing it themselves. The 18% rate used here is constructed conservatively from the largest available undergraduate study — the SERU Consortium survey of 95,000+ undergraduates across 20 research universities, published in Science in May 2026 — which found that roughly a third of undergraduates do not use AI and that non-use skews lower-income, female, and racially underrepresented, with the researchers warning that these students “may fall behind in college and eventually the workplace.” That framing suggests retrospective regret is possible among non-users whose non-use is access-constrained rather than principled. The RAND finding that 78% of non-users believe AI for schoolwork is harmful pulls the figure down: most students who don’t use AI think they are right not to, which is the opposite of a regret signal. The residual share is small.
The 2023-to-2025 norm shift in this domain has been faster than any other decision in the dataset. ResumeBuilder’s hiring-manager survey in early 2023 found only 18% of hiring managers could identify a ChatGPT cover letter; by 2025, ResumeBuilder’s follow-up and TopResume’s industry data show that roughly a third of hiring managers can spot AI-generated copy in under twenty seconds, 19.6% of recruiters would reject an AI application outright, and 52% accept proofreading or drafting support. Schools have shifted from blanket bans to acceptable-use policies. The figures cited here were taken in one short snapshot of a moving target; the directional finding (action generates more reported concern and concrete adverse outcomes than inaction) is robust, but the magnitudes will move again, plausibly within the next twelve months.
Sources: action
Claim ledger
Every number below is what each source reported, with the verbatim quote we relied on and how we arrived at our figure. Click any link to verify directly.
[1]RAND Corporation / Schwartz & Diliberti — More Students Use AI for Homework, and More Believe It Harms Critical Thinking: Selected Findings from the American Youth Panel
Reference source
67% of students agreed AI for schoolwork harms critical thinking in Dec 2025 (up from 54% earlier in 2025); among actual AI users the share was 60%, vs 78% among non-users; 62% of students used AI for homework, up from 48% in May 2025
Excerpt
“"67 percent of students said using AI for schoolwork harmed critical thinking… up from 54 percent earlier in the year. Non-AI users: 78 percent expressed concern vs. 60 percent among actual users. The percentage of middle school, high school, and college students using AI for homework rose from 48 percent to 62 percent. Female students: 75 percent believed AI harmed critical thinking; male students: 59 percent."
”
Source data from
2025-12-01
Accessed
2026-05-28
Calculation
RAND American Youth Panel survey of 1,214 youth ages 12–29, fielded December 2025, results published March 2026. This is the headline action-side proxy. The 60% figure (concern among actual AI users) is preferred over the 67% headline because non-users' moralism about AI inflates the population-level number; the users-only subgroup is the relevant comparator for "would a user, today, endorse a statement that mirrors their own behavior is harming them?" — the closest available proxy to retrospective regret on the action side. Female-skewed concern (75% vs 59%) suggests the concern signal is partly identity-driven and partly genuine, so 60% is treated as an upper bound on personal regret. No survey has directly asked "do you regret having used AI for your schoolwork?" — this is the closest substitute. The 14% increase in usage even as concern rose by 13 percentage points (54→67) is a behavioral revealed-preference signal that regret has not yet translated into behavior change at scale.
[2]ResumeBuilder.com (Pollfish panel) — 3 in 4 Job Seekers Who Used ChatGPT to Write Their Resume Got an Interview
Reference source
46% of job seekers used ChatGPT for resumes/cover letters; 78% got an interview; 59% were hired; 11% were denied a job when the interviewer discovered ChatGPT use; 88% said they were 'somewhat' or 'highly' likely to continue using it
Excerpt
“"46% of job seekers are using ChatGPT to write their resumes and/or cover letters. 78% got an interview when using application materials written by ChatGPT. 59% were hired after applying to a job using materials written by ChatGPT. 11% were denied a job when the interviewer discovered they used ChatGPT. 88% of the total sample say they are 'somewhat' (41%) or 'highly' (47%) likely to continue using ChatGPT to write their job application materials in the future."
”
Source data from
2023-02-07
Accessed
2026-05-28
Calculation
ResumeBuilder.com online survey of 1,000 US job seekers, fielded February 2023 via Pollfish. Reputable-reference (industry panel) not peer-reviewed. The 11% denied-a-job figure is the strongest available concrete-consequence proxy on the action side for the job-application cohort: it is a hard adverse outcome rather than an attitude. The 88% continuation rate is a strong revealed- preference signal against high regret — most users would not report planning to repeat a behavior they regretted — but is consistent with regret being concentrated in the ~11% who paid a concrete cost. Not used as the headline rate because (a) sample is job seekers only, not the schoolwork cohort, and (b) the field date precedes mature employer detection norms. Used to anchor the lower bound of plausible action regret.
[3]Bastani, Bastani, Sungu, Ge, Kabakcı, Mariman — Wharton / SSRN working paper — Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics
Primary study
Students with unrestricted ChatGPT-4 access during practice performed 17% worse on a subsequent no-AI exam than the control group; the AI-assisted group solved 48% more practice problems correctly but failed to retain the skill
Excerpt
“"When access is subsequently taken away, students actually perform worse than those who never had access (17% reduction for GPT Base). Students with ChatGPT solved 48 percent more of the practice problems correctly, but they ultimately scored 17 percent worse on a test. Students said they did not think that ChatGPT caused them to learn less even though they had."
”
Source data from
2024-07-15
Accessed
2026-05-28
Calculation
Field RCT with ~1,000 Turkish high-school math students across three grade levels, four 90-minute practice sessions, posted to SSRN July 2024 by Wharton researchers. Used as the mechanism evidence underpinning the action-side regret framing: the harm the RAND respondents fear (skill erosion) is empirically real, even though students themselves did not perceive it. This creates the gap that makes "regret" particularly hard to measure — students who lost skill did not realize they had lost it, so self-report regret rates probably understate true regret. Not a US sample; treat the 17% effect size as directional. The paper is a working paper, not yet peer-reviewed at last revision.
Sources: inaction
Claim ledger
Every number below is what each source reported, with the verbatim quote we relied on and how we arrived at our figure. Click any link to verify directly.
[1]UC Berkeley News (reporting on SERU Consortium study published in Science) — The Largest Study of AI Use by Undergrads Is In, Revealing Disparities in Access — and in Cheating
Peer-reviewed
Two-thirds of undergraduates used GenAI as of spring 2024; low-income, racially underrepresented, and female students used AI less, and researchers warned these students 'may fall behind in college and eventually the workplace'
Excerpt
“"Low-income, racially underrepresented and female students using AI less, and these students may fall behind in college and eventually the workplace because of unequal access to or practice using AI. Students from wealthier families can access advanced AI tools… students who don't have resources may only… use free AI tools."
”
Source data from
2024-04-01
Accessed
2026-05-28
Calculation
SERU Consortium survey of 95,000+ undergraduates at 20 research universities, fielded spring 2024, published in Science May 2026. Used as inaction-side opportunity-cost proxy: the published finding frames non-use as a future disadvantage, which is the mechanism that would generate retrospective regret among non- users. The 18% inaction-regret estimate is constructed conservatively from the roughly one-third of undergrads who did not use AI (per this study and Berkeley News reporting), scaled by the share whose non-use was driven by access barriers rather than principled choice — roughly half of non-users, yielding ~17% as the share who would plausibly report retrospective regret about not using AI. No direct survey asks non-users "do you regret not having used AI?" — the question would assume the decision was deliberate, which for many non-users it was not. This is the data-sparsest side of the entry and the rate has wide uncertainty.
[2]RAND Corporation / Schwartz & Diliberti — More Students Use AI for Homework, and More Believe It Harms Critical Thinking: Selected Findings from the American Youth Panel
Reference source
78% of non-AI-using students believed AI for schoolwork harms critical thinking (Dec 2025); 38% of students were not using AI for homework as of Dec 2025
Excerpt
“"Non-AI users: 78 percent expressed concern vs. 60 percent among actual users. The percentage of middle school, high school, and college students using AI for homework rose from 48 percent to 62 percent."
”
Source data from
2025-12-01
Accessed
2026-05-28
Calculation
Same RAND panel. The 78% concern rate among non-users is direct evidence that abstention from AI is mostly principled, not regretful: the modal non-user thinks AI use is harmful and is therefore unlikely to regret their non-use. This pulls the inaction-regret rate down sharply. The 18% headline estimate reflects the residual share of non-users whose non-use is access-constrained rather than principled (the SERU demographic gap finding), centered conservatively. The revealed-preference signal — total AI usage rising from 48% to 62% in seven months — is consistent with some inaction-side regret driving conversion, but cannot be cleanly decomposed from peer effects, capability improvements, or institutional normalization.
Caveats
Both sides are proxies — no published survey directly asks "do you regret having used AI for your assignment?" or "do you regret having written it yourself?" The action-side anchor (60%) is the share of actual student AI users who agreed AI for schoolwork harms critical thinking, taken from the RAND American Youth Panel December 2025 wave. It is a concern measure, not a regret measure: a student can believe their behavior is harmful and still not regret it (e.g., if the time saved outweighs the perceived harm in their own utility function). The continuation-intent data (88% of ChatGPT-using job seekers said they would continue) is direct evidence that observed concern coexists with low behavioral regret for many users. The 11% denied-a-job rate (ResumeBuilder, 2023) is the cleanest concrete- consequence number on the action side, but applies only to the job- seeker cohort, and the survey predates mature employer detection norms — 2025-era detection and tolerance are different (the field has shifted toward acceptance of AI as a drafting tool while penalizing fully AI-generated submissions, per TopResume and ResumeBuilder hiring-manager surveys). The Bastani et al. RCT shows real skill erosion (17% lower exam scores after unguarded ChatGPT practice) but is from Turkey, not the US, and students in the study did not recognize the skill loss — meaning self-report regret almost certainly understates true regret on a skill-development axis. The inaction- side 18% is the data-sparsest figure in the pair: it is constructed from the SERU finding that ~33% of undergrads do not use AI and that non-use correlates with low-income, female, and racially under- represented status (suggesting access-constrained rather than principled non-use for some), scaled down to reflect the RAND finding that 78% of non-users believe AI is harmful (meaning most non-use is principled and unlikely to be regretted). The two cohorts are not fully comparable: action-side anchors are mostly K-12 and undergrad students; the job-seeker corroborating data is adult applicants. Aggregating "school assignment" and "job application" as one decision papers over a meaningful distinction in incentives, detection norms, and stakes. The 2023→2025 norm shift is rapid: behavior, perception, and employer/instructor tolerance have all moved more than once per year, so this entry has a shorter shelf life than most in the dataset. The regret_delta of 0.42 is directionally robust — action generates substantially more reported concern and concrete adverse outcomes than inaction across every available proxy — but the magnitude should not be over-interpreted given the proxy nature of both sides.
