Contract Cheating After Generated Text: What Changed
Contract cheating is the form of academic misconduct that used to leave a paper trail. A student arranges for a third party to complete assessed work and submits it as their own. For most of its history the third party was a person: a ghostwriter, an essay mill, a tutor who went further than tutoring, a relative with time on their hands. That fact shaped how markers were trained to spot it, what evidence an integrity panel expected, and the laws two countries eventually passed against the businesses that sold it.
Free text generators did not invent a new kind of misconduct so much as remove the human from the old one. The third party is now a machine. It charges nothing, keeps no records, answers within seconds, and never emails back to negotiate a deadline. Each feature that had made contract cheating detectable, prosecutable and comparatively rare has gone, and the machinery built to deal with it is pointed at a problem that has changed shape.
The short answer. Contract cheating meant outsourcing assessed work to a paid third party, and it was caught through mismatch and money: prose that did not sound like the student, and records proving a transaction. Generated text removes the price, the correspondence and the seam, so a case can no longer be built from the product. It must be built from the process.
What contract cheating is
The term was coined by Robert Clarke and Thomas Lancaster in a 2006 paper presented at the Second International Plagiarism Conference in Newcastle. They had been watching freelance auction sites on which students posted programming assignments as jobs and contractors bid to complete them, and they chose the word contract deliberately: the defining feature was a bargain struck with someone outside the assessment, whether or not money changed hands. Later definitions widened it to cover unpaid help from friends and family and impersonation in online examinations.
The definition separates contract cheating from plagiarism. Plagiarism reuses existing text, and the plagiarism service most universities license catches it by matching a submission against a corpus. A contract-cheated essay is original and matches nothing, which is why essay mills advertised plagiarism-free work, and why two decades of investment in text matching never touched them.
How common contract cheating was before generated text
The most careful estimate comes from Philip Newton's 2018 systematic review in Frontiers in Education, which pooled self-report surveys of more than 54,000 students across 65 studies and found that 15.7 percent of students in samples collected from 2014 onward admitted to some form of contract cheating, against a historical average of 3.5 percent, with the trend rising. Newton was clear that self-report understates the true figure, since respondents are confessing to conduct that can end a degree.
Two further findings bear on what follows. Bretag and colleagues, surveying Australian universities in 2019, found the strongest predictors were circumstantial: dissatisfaction with teaching, plentiful opportunity, and speaking a language other than English at home. Lancaster and Cotarlan, in 2021, documented a sharp rise in requests to a homework-help site once the pandemic moved assessment out of the examination hall. Contract cheating tracked opportunity, and opportunity meant work done out of sight.
The essay mill law: England and Australia
Australia legislated first. The Tertiary Education Quality and Standards Agency Amendment (Prohibiting Academic Cheating Services) Act 2020 made it an offence to provide or advertise academic cheating services to students at Australian providers, with penalties of up to two years' imprisonment, and let the regulator seek the blocking of cheating websites. England followed. Part 4 of the Skills and Post-16 Education Act 2022 created offences of providing, arranging and advertising cheating services to students at post-16 institutions in England.
Both statutes share a theory of the problem: contract cheating is a market, markets have suppliers, and suppliers can be prosecuted. The theory was right about the world in which it was written, and that world ended within months. A general-purpose language model falls outside either Act, since it is not marketed for cheating and completes assignments only in the sense that a word processor does. The market was undercut by a free good, which leaves the statutes intact but largely idle.
What a free generator changes
Four properties of contract cheating shifted at once. The first is cost: a mill priced its work by length and deadline, and the price kept the practice, in Newton's samples, to a minority. A generator is free. The second is time. Mills needed days; a generator needs under a minute, so the decision to cheat can be made at eleven at night without planning, which our review of the homework question treats as the central fact about unsupervised work.
The third is the record. A purchase left a bank entry, an order confirmation, an email thread, and sometimes a writer's file with the student's brief attached; in the blackmail cases that surfaced from 2018 onward, mills themselves contacted universities to expose students who had refused to pay. A generator leaves nothing outside the student's own device. The fourth is style. The seam between the student's voice and the contractor's was the main clue a trained marker used; Dawson and Sutherland-Smith showed in 2018 that markers alerted to purchased work could identify a meaningful share of it, at the cost of false positives. A generator can be handed the student's earlier work and told to imitate it, and the seam disappears.
Mills could not compete with a free substitute, and the survivors pivoted to selling rewriting and tutoring. The detection consequence is the subject of a sibling article on what the research shows about detector accuracy. The residual signal in generated prose is fluency, and Liang and colleagues found in 2023 that detectors reading fluency as a signal flagged the majority of essays by non-native English speakers while passing native speakers' work. The instruments sold to find a seam fall hardest on the students Bretag had identified as most exposed.
From purchase records to process evidence
| Dimension | Essay mill contract cheating | Generated-text misuse |
|---|---|---|
| Cost to the student | Paid per assignment, by length and deadline | Free |
| Time to obtain | Days | Under a minute |
| Evidence left behind | Payment record, correspondence, mill account, writer's file; occasionally the mill's own disclosure | Nothing outside the student's device; no transaction, no third party |
| Stylistic signal | A seam between student and contractor, readable by trained markers | Weak; the model can be told to imitate the student |
| Legal status | Provider commits an offence in Australia (2020) and England (2022); student faces institutional penalty | Tools are lawful; only the student's use is in question, under institutional rules |
| What settles a case | The transaction, plus whether the student can discuss the work | The process record, plus whether the student can discuss the work |
In every row but the last, the evidence that used to settle a contract cheating case sat outside the student, in a record someone else kept. A panel could ask "did you buy this" and prove an answer either way. In the last row, the half of the old test that survives is the half that was always about the student's mind: can they explain what they wrote, why they chose that structure, what they rejected? That half now has to carry the whole case, and it can only do so if the assignment was designed to leave a record of the work as it was made.
This is the argument the site made in Proof of Work, applied to one category of misconduct. The record has three familiar forms. The draft trail, which in a tracked document becomes version history as evidence of authorship. The conversation, a sampled three-minute oral examination, which is the test panels always applied at the end of a hearing, moved to the front. And disclosure, a short account of what assistance was used, which turns a suspicion into a discussion.
None of this is new pedagogy. Collins, Brown and Holum argued in 1991 that instruction improves when expert thinking is made visible to the apprentice; the same visibility, run in the other direction, lets a teacher see that the thinking was the student's. Contract cheating flourished where that visibility had been traded for efficiency. The generator has made the old weakness impossible to ignore.
Frequently asked questions
What is the definition of contract cheating?
Contract cheating is the submission of assessed work that a third party completed on the student's behalf, presented as the student's own. The term comes from Clarke and Lancaster's 2006 study of freelance sites where assignments were posted as jobs. Payment is common but not required; substantive help from a friend, a relative or a machine falls within most institutional definitions.
Is contract cheating illegal?
For the provider, in some jurisdictions. Australia's 2020 amendment to the TEQSA Act and England's Skills and Post-16 Education Act 2022 both criminalise providing or advertising cheating services to students. Students are not prosecuted under either law; they face institutional penalties under academic misconduct rules, which range from a zero mark to expulsion.
Is using a language model to write an essay contract cheating?
Under most institutional definitions, yes, if the machine produced the substance of work the student was required to complete personally and the student presented it as their own. The provider laws do not apply, because a general-purpose tool is not a cheating service. The assignment's stated rules decide: where assistance is permitted and disclosed, the same act is not misconduct.
How is contract cheating detected?
Purchased work was detected through mismatch with the student's known writing, references that did not fit the course, and occasionally the mill's own records. Generated work removes those signals, and automated detectors are unreliable and biased against non-native writers. The dependable route is process evidence: drafts, version history, a short conversation about the work, and a checkable disclosure statement.
Where this leaves a school
Contract cheating was a market problem with a market's evidence, and schools learned to prosecute it with receipts. A free generator has ended the market without ending the conduct, and the receipts have gone with it. A school that keeps treating this as a detection problem will keep losing cases it cannot prove and accusing students it cannot fairly convict. What remains is to design the assignment so the work leaves a record of its own making, and to let that record, with a short conversation, carry the judgment.
The Academic Integrity hub gathers the rest of this lane. For a practical way to assemble the record across a term, begin with the process portfolio.