Debated.
Start debating
All motions

Sources · AI help with coursework

This house would let students use AI to help write their coursework.

Fact bank v1 · 12 cards

  1. AIC-01Mostly true

    “Most students are already using AI for their schoolwork anyway.”

    Roughly right for US teenagers. In a Pew Research Center survey in late 2025, 54% of US teens said they had used AI chatbots for help with schoolwork. But only 1 in 10 said they do all or most of their schoolwork with chatbots' help.

    Most teens who use chatbots for school use them for some (21%) or a little (23%) of their work, and 45% hadn't used them this way at all. In the UK, Ofcom found in 2026 that 56% of 8 to 17 year olds had used AI, rising to 66% of 16 and 17 year olds, and about half of those who use it (51%) use it for schoolwork or homework. That works out at roughly 3 in 10 of all 8 to 17 year olds. Use is growing fast: in a 2024 Pew survey, 26% of US teens said they used ChatGPT for schoolwork, up from 13% in 2023. How many students use AI doesn't by itself settle whether it should be allowed in marked coursework.

    “More than half of teens say they have used chatbots to search for information (57%) or get help with schoolwork (54%).”
    Pew Research Center·How Teens Use and View AIIndependent research institute · 24 Feb 2026 ↗

    Further sources for this card

  2. AIC-02Mostly true

    “AI detectors wrongly accuse students who aren't native English speakers of cheating.”

    Backed by a well-known 2023 Stanford study. Seven widely used AI detectors labelled more than half of 91 essays by non-native English speakers as written by AI (61.3% on average), while getting essays by US eighth-graders almost all right.

    The essays were written for the TOEFL, an English test for non-native speakers, and came from a Chinese online forum. The researchers think detectors get fooled because simpler word choices look more like AI writing. When they used ChatGPT to make the same essays sound more like a native speaker, false alarms fell from 61.3% to 11.6%. They warned against using detectors to judge students' work. Not all research agrees: a larger 2024 study of essays from the GRE test, using the researchers' own detectors, found the opposite, with non-native speakers' writing looking less like AI writing than native speakers' did. Detectors have also changed since 2023: a 2025 test found the best commercial detector made almost no mistakes, but it did not test writing by non-native speakers or school essays (see AIC-11).

    “over half of the non-native English writing samples were misclassified as AI generated, while the accuracy for native samples remained near perfect”
    Patterns (open-access article on PubMed Central)·GPT detectors are biased against non-native English writers (Patterns)Government / parliament · 10 Jul 2023 ↗

    Further sources for this card

  3. AIC-03Mixed

    “Exam boards already let students use AI in their coursework, as long as they say they used it.”

    Only sometimes. Ofqual's 2026 guide for students in England says the rules vary by qualification, so students must check with their teacher first. Where AI is allowed it must be referenced, and all work must still be in the student's own words.

    Ofqual, the exams regulator in England, says using AI tools wrongly or not referencing them counts as malpractice, which can lead to being disqualified. Its resources for schools say using AI to generate coursework without proper disclosure is cheating. So the rules don't simply allow AI whenever you mention it, but they don't ban all use either. The Joint Council for Qualifications (JCQ) also gives schools and colleges detailed guidance on preventing and spotting AI misuse in assessments.

    “Check with your teacher before using AI tools as rules vary by qualification. If AI use is allowed it must be referenced correctly.”
    GOV.UK (Ofqual)·What you should know during your exams and assessments (Ofqual student guide to exams and assessments in 2026)Government / parliament · 15 Jan 2026 ↗

    Further sources for this card

  4. AIC-04True

    “A Harvard study found students learned twice as much from an AI tutor as from normal lessons.”

    Confirmed: in a 2025 study of 194 Harvard physics students, median learning gains with a specially built AI tutor were more than double those from the same lesson taught in class, and most students spent less time on it.

    The tutor was not plain ChatGPT. Its makers built it around good teaching methods and wrote step-by-step answers into its instructions so its explanations were accurate. The class lessons it beat were already well designed, with students working through problems together. The study covered two lessons on one university course and tested students soon afterwards. The researchers say AI should not be used where students are likely to use it as a crutch to skip thinking, and a tutor that teaches you is different from AI helping write marked coursework.

    “We find that students learn significantly more in less time when using the AI tutor, compared with the in-class active learning.”
    Scientific Reports (open-access article on PubMed Central)·AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting (Scientific Reports)Government / parliament · 3 Jun 2025 ↗
  5. AIC-05Unsupported

    “A big review of 51 studies proved that ChatGPT has a large positive effect on how well students learn.”

    That review can't be relied on. It was published in May 2025 but retracted in April 2026, after the journal's editor lost confidence in its analysis and conclusions.

    The review, by Jin Wang and Wenxiang Fan, pooled 51 studies from November 2022 to February 2025 and reported a large positive effect on learning performance. The editor withdrew it after concerns were raised about discrepancies in the analysis. Stronger single studies point both ways: a carefully built AI tutor helped university physics students learn more (see AIC-04), but plain ChatGPT-style help hurt high school maths students' exam results (see AIC-07). England's Department for Education says it has limited evidence on how AI use affects pupils' learning.

    “Taken together, the identified issues undermine the Editor’s confidence in the validity of the analysis and the conclusions drawn from it.”
    Humanities and Social Sciences Communications (Springer Nature)·Retraction Note: The effect of ChatGPT on students’ learning performance, learning perception, and higher-order thinking: insights from a meta-analysis (Humanities and Social Sciences Communications)Peer-reviewed research · 22 Apr 2026 ↗

    Further sources for this card

  6. AIC-06Mixed

    “Hardly anyone actually cheats with AI: only about 100 students got caught in GCSEs and A levels last year.”

    The number is right, but it only counts cheating that was caught and proven. In summer 2025 there were 100 proven cases of AI-related plagiarism in GCSEs, AS and A levels in England, 2.0% of all proven cases of students breaking exam rules. Ofqual says detected cases are relatively low, but teachers and school leaders are seriously worried about how big the problem could be.

    AI cases rose from 85 in summer 2024. In a test at a UK university, markers missed 94% of AI-written exam answers, so few cases caught doesn't prove few students cheat (see AIC-08). In the US, 59% of teens told Pew that students at their school use AI to cheat at least somewhat often. An Ofqual briefing for schools says 1,125 cases in 2025 ended with students losing a whole GCSE or A level, but that figure covers all kinds of rule-breaking, not just AI. Phones were the most common offence.

    “plagiarism involving the misuse of AI accounted for 75.0% of all student plagiarism cases (100 cases) and 2.0% of all student malpractice cases.”
    GOV.UK (Ofqual)·Malpractice in GCSE, AS and A level: summer 2025 exam seriesGovernment / parliament · 11 Dec 2025 ↗

    Further sources for this card

  7. AIC-07Mostly true

    “Studies show that students who use ChatGPT to do their work end up learning less.”

    Mostly right for unrestricted use. In a 2025 study of nearly 1,000 high school maths students, those given a plain ChatGPT-style tool scored 48% higher on practice problems, but 17% lower than students who never had it on a later exam without AI. A version with teacher-designed safeguards removed most of that harm.

    The study took place at a high school in Turkey in autumn 2023. Many students used the plain tool as a crutch, asking for answers and copying them. The safeguarded tutor gave teacher-designed hints instead of answers: its students did far better in practice (127% higher) but no better in the exam than students without AI. This was one subject at one school, measured in the short term, and the researchers note AI models have improved a lot since. A carefully built AI tutor helped university physics students learn more (see AIC-04), so how AI is used seems to matter a great deal.

    “on the exam, students in the GPT Base arm perform statistically significantly worse than students in the control arm by 17%”
    Proceedings of the National Academy of Sciences (open-access article on PubMed Central)·Generative AI without guardrails can harm learning: Evidence from high school mathematics (PNAS)Government / parliament · 25 Jun 2025 ↗

    Further sources for this card

  8. AIC-08True

    “In a real test at a UK university, markers failed to spot almost all the AI-written answers, and the AI got better grades than real students.”

    Confirmed: in a 2023 test at the University of Reading, researchers secretly handed in GPT-4 answers in five psychology modules. 94% were not spotted, and on average they got grades about half a grade boundary higher than real students' work.

    The answers were for unsupervised online exams, marked by staff who didn't know about the test, and the marking software had no AI detector. The AI answers were handed in unchanged, which the researchers call the most detectable way possible. In one third-year module the AI answers mostly did worse than real students. This was university work, not school coursework, where teachers may know how their own pupils write: Ofqual tells teachers that if work doesn't sound like the student, they should question it.

    “We found that 94% of our AI submissions were undetected.”
    PLOS ONE (open-access article on PubMed Central)·A real-world test of artificial intelligence infiltration of a university examinations system: A “Turing Test” case study (PLOS ONE)Government / parliament · 26 Jun 2024 ↗

    Further sources for this card

  9. AIC-09Mostly true

    “AI chatbots make up facts and even invent fake sources.”

    Mostly right. A 2023 study asked ChatGPT to write short literature reviews on 42 topics and found that 55% of the references from GPT-3.5, and 18% from GPT-4, were made up.

    Many real references had mistakes too: 43% of the real GPT-3.5 references and 24% of the real GPT-4 ones got key details wrong, such as the authors, title or date. The researchers said GPT-4 was a big improvement, but problems remained. These were 2023 versions, so today's tools may do differently. In a 2025 maths study, a plain ChatGPT-style tool often gave students wrong answers, which the researchers put down to hallucinations, meaning confident but false answers (see AIC-07). Either way, students need to check what AI tells them.

    “Within this set of documents, 55% of the GPT-3.5 citations but just 18% of the GPT-4 citations are fabricated.”
    Scientific Reports (open-access article on PubMed Central)·Fabrication and errors in the bibliographic citations generated by ChatGPT (Scientific Reports)Government / parliament · 7 Sept 2023 ↗

    Further sources for this card

  10. AIC-10Mostly true

    “Even teenagers think using AI to write your essays is wrong.”

    Mostly right. In a 2024 Pew Research Center survey, 42% of US teens said using ChatGPT to write essays is not acceptable, while only 18% said it is. Others were unsure.

    Pew says 15% to 21% of teens were unsure about each use it asked about. Teens were much more relaxed about other uses: 54% said it's fine to use ChatGPT to research new topics, and only 9% said it isn't. So many young people draw a line between AI helping you find things out and AI writing the work for you. Even so, fewer than half of all teens said essay-writing with AI is wrong. The survey was of US teens aged 13 to 17 in autumn 2024.

    “18% say it’s acceptable to use ChatGPT to write essays, and 42% say it’s not acceptable.”
    Pew Research Center·About a quarter of U.S. teens have used ChatGPT for schoolwork – double the share in 2023Independent research institute · 15 Jan 2025 ↗
  11. AIC-11Mixed

    “Schools can just use AI detectors to catch students who cheat with AI.”

    It depends on the tool. In a 2025 test, one paid detector, Pangram, made almost no mistakes, even on AI text disguised by a 'humanizer' tool. Two other paid detectors were easier to fool with it, and a free open-source one wrongly flagged a lot of human writing as AI.

    The 2025 test, by University of Chicago researchers, used news, blogs, reviews, novels and CVs, not school essays, and did not test writing by non-native English speakers. It is a working paper that has not been peer reviewed, and the authors expect detector performance to keep changing in an arms race with AI tools. They also say detectors can't yet tell how AI was used, for example writing a whole text or just fixing grammar. Earlier detectors did much worse: OpenAI withdrew its own detector because of its low accuracy, as it correctly spotted AI-written text only 26% of the time, and a 2023 Stanford study found seven detectors flagged more than half of essays by non-native English speakers as AI (see AIC-02).

    “Commercial detectors outperform open-source, with Pangram achieving near-zero FNR and FPR rates that remain robust across models, threshold rules, ultra-short passages”
    National Bureau of Economic Research·Artificial Writing and Automated Detection (NBER Working Paper 34223)Independent research institute · 1 Sept 2025 ↗

    Further sources for this card

  12. AIC-12Mostly false

    “Using ChatGPT rots your brain. An MIT study proved it.”

    The study doesn't show that. A small MIT study of 54 people found that those who wrote essays with ChatGPT had weaker brain connectivity while writing than people using no tools, and many couldn't quote what they had just written. That is a warning sign, not proof of lasting harm, and the study was shared as a preprint before peer review.

    The study ran four sessions over four months, with groups writing essays using ChatGPT, a search engine or no tools while their brain activity was recorded (EEG). The ChatGPT group showed up to 55% less brain connectivity than the no-tools group, and 83% of ChatGPT users couldn't quote from an essay they had just written. When ChatGPT users later wrote without it, they still showed weaker connectivity, while people who switched the other way showed more brain activity. The researchers call this 'cognitive debt' and suggest learners should do their own thinking before they use AI. Nature reported that scientists warned against reading too much into such a small experiment.

    “the LLM group exhibited up to 55% reduced connectivity compared with brain-only participants”
    British Journal of General Practice (article on PubMed Central)·Your brain on ChatGPT (British Journal of General Practice)Government / parliament · 25 Aug 2025 ↗

    Further sources for this card

The judge is an AI model. It can only rule on claims these cards cover. Everything else stays unverified, and it never rules on opinions. See every motion's cards.