There are two stories about cheating on the TOEFL, and they are usually told separately.
The first is about an industry. In China and beyond, a network of agencies sells guaranteed test scores, sitting exams for students, remote-controlling their laptops, and even dressing up on camera to impersonate them. The second is about enforcement. Every year, thousands of test-takers open an email from the ETS Office of Testing Integrity telling them their scores have been cancelled, with no refund, limited appeal, and evidence they are not allowed to see.
These two stories are the same story. One is the reason the other exists. And caught in the middle are real people: some who cheated and gambled their futures on it, and some who insist they did nothing wrong and cannot prove it. This is a look at how we got here, how the detection actually works, and what it means if you are preparing for the test honestly.
A problem as old as the test
Large-scale cheating on ETS exams is not new, and the history is worth knowing because it explains why the response today is so aggressive.
The landmark case came in 2002. ETS discovered that on the computer-based GRE, test-takers in parts of Asia were memorizing questions and posting them, sometimes with answers, on websites. Because the computer-based test reused a pool of questions, this worked. According to reporting at the time, toward the end of a testing cycle, national average verbal scores were climbing by as much as 100 points in China and 50 points in South Korea and Taiwan, on a 200 to 800 scale. ETS's investigation reached across dozens of countries. The company's response was drastic: it pulled the computer-based test in the affected regions and reverted to paper exams with single-use questions.
That episode set the pattern that still governs ETS's thinking. Reused content is a vulnerability, so the organization rotates and retires enormous numbers of questions every year. And when a region shows statistically implausible results, ETS is willing to take sweeping action rather than adjudicate one test-taker at a time. Both of those instincts matter enormously to what is happening now.
The score-guarantee industry
Search Chinese e-commerce platforms for TOEFL or GRE help and you will find sellers offering something specific: a guaranteed score. The practice has a name, baofen, which translates roughly to "guaranteed score," and by multiple accounts it is a multimillion-dollar business.
The services are more sophisticated than a cheat sheet. According to reporting by Rest of World and others, agencies take the exam for their clients, either by sitting beside them or by remotely controlling their computers during at-home tests. To beat the spoken sections, some impersonators appear on camera dressed to look like the student. Payment often runs through escrow on platforms like Taobao, and prices reportedly range from around 2,000 to 6,000 US dollars per exam. One agency told reporters it handled 100 to 200 clients a month.
The move to at-home testing during the pandemic poured fuel on this. A remote proctor watching through a single webcam simply cannot see what a proctor in a test center can. ETS reported a roughly 200 percent increase in TOEFL and GRE score cancellations in the second year of at-home testing compared with the year before. That number cuts two ways. It shows ETS catching far more fraud. It also hints at how much more there was to catch.
How ETS detects it now
Modern detection runs on two layers: people and software.
The human layer is proctoring. In the Home Edition, live proctors and automated monitoring watch the room and the screen, and a proctor can stop a session during check-in or mid-test if something looks wrong. Many of the disputes you read about start here, with a proctor flagging a water bottle, a phone, a reflection, or a setup that does not look right.
The software layer is where it has changed most. Detecting copied or templated writing is a mature technology; text-similarity and plagiarism detection have existed for years, which is why memorized essay templates have been getting Writing scores cancelled for a long time. The newer development is that ETS has extended the same idea to Speaking. In a 2026 research paper, ETS described a system called the Automated Speech Similarity Detector, or AutoSSD. It transcribes every spoken answer into text, then compares each response two ways: against other test-takers' responses, and against the prompt itself. Answers that are too similar to each other suggest a shared template. Answers that mostly echo the question suggest the speaker is padding rather than answering. Flagged pairs are sent to a human expert who makes the final decision. ETS reported using this system on a large test administration between late 2022 and early 2023.
You can hear this logic in ETS's own cancellation language. Test-takers who have been flagged report receiving an email stating that their scores "could not be reported because your Writing response did not reflect the assigned task, or it contained ideas, language and/or examples found in published sources or other test takers' responses." That single sentence is the whole detection philosophy: answer the actual task, in your own words, or the score does not count.
There is also a third, quieter mechanism that has nothing to do with plagiarism. ETS can cancel scores for statistical reasons, when a performance looks internally inconsistent or improbable enough to question its validity. This is the route that most surprises people, because it does not require anyone to prove you copied anything.
The delayed reckoning
The part that makes ETS enforcement feel so frightening is timing. A cancellation does not always arrive the week of the test. Test-takers report scores being pulled months, and in some accounts more than a year, after the exam, long after the score was used for an application.
When that happens, the consequences described in these accounts are severe. Reported outcomes include voided scores, multi-year bans from ETS tests, and in the most serious cases, universities revoking admission or degrees and, for some international students, the collapse of their immigration status. These are largely self-reported stories circulating in test-prep communities, so they should be read as accounts rather than verified case files. But they are numerous, consistent, and sobering, and the human cost in them is real. Some of the people describing these situations are in genuine distress. If a piece of this describes you or someone you know, please treat it as the serious matter it is and reach out to someone you trust for support.
The lesson buried in the delay is the one the guarantee sellers never mention. A "guaranteed score" is not guaranteed at all. It is a deferred risk that can detonate at the worst possible moment, after you have built a life on top of it.
The fairness problem, honestly
It would be easy to end there, with a clean warning against cheating. But the honest version of this story has a harder second half, because the same system that catches cheats also catches people who say, credibly, that they did nothing wrong.
Three things make this genuinely difficult.
First, opacity. When ETS questions a score, the evidence is often described as proprietary and is not shared with the test-taker. You can be told you cheated without being shown how.
Second, appeal. The options are narrow and depend on where you tested. ETS's process, when a score is questioned, can include voluntary cancellation, a free retest, a voucher, or arbitration that ETS pays for. But by ETS's own terms, the arbitration option is available only to people who test in the United States and its territories. Test outside the US, and a cancellation is effectively final. Experienced TOEFL teachers put it bluntly to worried students: there is, in practice, no appeal. And a warning worth repeating from those same teachers, do not try to force the issue by reversing your test payment through your bank, because that can get you barred from future ETS tests.
Third, the false-positive risk of judging non-native English by machine. This is subtle and important. Separate from ETS's own similarity system, researchers at Stanford found that general AI-writing detectors falsely flagged a large majority of TOEFL essays, essays written by non-native speakers, as machine-generated, while flagging almost none of the essays by US students. That study was about a different class of tool, not AutoSSD specifically, so it is not evidence against ETS's system. But it illustrates the underlying worry: automated systems that judge whether English "looks authentic" can be systematically unfair to the very people who take the TOEFL. When you combine template-heavy prep, which makes thousands of honest answers look alike, with opaque evidence and no real appeal, you get a system in which some innocent people will look guilty and have no way to prove otherwise.
None of this means the enforcement is a scam, and it does not mean any particular person who was flagged is innocent. There is a real, large, well-documented cheating industry, and ETS has a legitimate reason to fight it hard. It does mean the current system asks honest test-takers to carry the cost of that fight, and gives them very little recourse when it goes wrong.
What this means if you are preparing honestly
You cannot control ETS's evidence, its thresholds, or its appeals policy. You can control whether your answers could ever be mistaken for the thing it is hunting. That is the practical takeaway, and it happens to line up exactly with what earns a high score anyway.
Never use a score service. The guarantee is the trap. Even setting aside the ethics, the risk does not expire when the score posts; it follows you into your degree and your visa.
Never memorize a template. Not because structure is bad, but because a fixed set of memorized sentences pasted onto any prompt is precisely the pattern the detectors look for, and the ones taught to thousands of students are the easiest of all to match. Learn a flexible shape for your thinking, and change the words every time to fit the exact question.
Answer the specific prompt, in your own words, with your own specific examples. A real detail from your life, a name, a place, a moment, cannot match anyone else's template, and it also happens to be what a strong answer is made of. On both Writing and Speaking, the safest response and the highest-scoring response are the same response: a genuine one.
The realistic picture if a score is questioned
This table describes patterns reported by test-takers and test-prep professionals, not formal legal advice. ETS's official policies govern, and they can change, so check the current terms directly.
FAQ
Can ETS cancel my TOEFL score after I have already been admitted?Yes. Cancellations can arrive months or, in some reported cases, more than a year after the test, which can affect an application or enrollment that already relied on the score. This is one reason the risk of any shortcut is so long-lived.
Is there an appeal if my score is cancelled?It is very limited. ETS's arbitration option is available only for tests taken in the United States and its territories. Outside the US, a cancellation is generally final, though you can submit information for ETS to consider.
Can I be flagged for cheating even if I did not cheat?It is possible, and it is the core fairness concern. Memorized templates used by many test-takers can make an honest answer look canned, and the evidence behind a decision is often not disclosed. The best protection is to answer in your own words with your own examples.
Is a "guaranteed score" service safe if the agency seems professional?No. These services are the exact target of ETS's detection and enforcement, and the consequences, including cancelled scores long after the fact, revoked admissions, and bans, fall on the test-taker, not the agency.
Does memorizing a template count as cheating?Reciting fixed, pre-written sentences that do not genuinely respond to the prompt is what similarity detection is built to catch, and ETS's own cancellation notices reference answers that contain "ideas, language and/or examples found in published sources or other test takers' responses." A flexible structure that you adapt to each question is a different thing and is fine.
Should I cancel my payment with my bank if I feel cheated by the process?Test-prep professionals warn against it, because reversing payment has been reported to result in a ban from future ETS tests, which usually makes the situation worse.
Why does ETS use automated detection at all?Because the cheating industry it is fighting is large, organized, and well funded. The difficulty is that automated systems judging non-native English can also misfire, which is why the lack of transparency and appeal is such a live issue.
The takeaway
Cheating on the TOEFL is not a fringe problem. It is an industry, and ETS's response has hardened into a system that is powerful, opaque, and unforgiving, one that catches real fraud and, at times, sweeps up people who insist they are innocent and cannot prove it. You do not get to set the rules of that system. What you can do is make sure you never resemble the thing it hunts. Do your own work, answer the real question in your own words, and keep your score, and your future, entirely yours.
This article covers test security and score cancellation, topics that carry real financial and personal stress for the people affected. If you are dealing with the aftermath of a cancelled score and struggling, you do not have to handle it alone; reach out to someone you trust, and if you are in distress, seek appropriate support in your country.
Sources
- Chinese students use remote access software to cheat on US college entry exams — Rest of World (2022)
- Sophisticated test scams from China invade U.S. college admissions — Hechinger Report
- ETS Says GRE Scores From China, South Korea, and Taiwan Are Suspect — Chronicle of Higher Education (2002)
- AutoSSD: Automated Detection of Similar Speech Responses — ETS Research (RM-26-02)
- Why and How ETS Questions GRE Test Scores (options and arbitration) — ETS
- TOEFL Scores Canceled: Why and What to Do — Test Resources (Michael Goodine)
- GPT detectors are biased against non-native English writers — Stanford (Liang et al., 2023)