foundations · Level 1
Use AI to study, research and write without losing your own thinking
Quiz yourself, check every claim and citation, write in your own voice, and stay inside your school, university or employer's rules.

Your next step
Start this course online
Register your start, then work through the lessons and exercises at your own pace. You can download the PDF separately. Where a scored exam is available, you will find it after the lessons.
Register or resume your course
Start with the essentials
The short answer
Use AI to support your thinking, not replace it. Attempt the work first, then ask for quizzes, explanations and feedback. Check every fact and reference against the real source, because AI can produce fluent but false statements and invented citations. Follow your school, university or employer's rules, say how you used AI, and never paste other people's personal data or confidential work into a tool.
What you will learn
- You will be able to explain, with evidence, why retrieval practice and spacing help learning, and what the evidence on AI does and does not show.
- You will be able to turn your notes into practice questions and use an assistant to quiz, explain and check your understanding.
- You will be able to check every claim and citation an AI gives you against the real source, and record the result.
- You will be able to write your own draft first, use AI for feedback, and log what came from where.
- You will be able to find the rules that apply to you, read what JCQ and the Department for Education say, and disclose AI use honestly.
- You will be able to keep other people's personal data and confidential work out of AI tools, and write your own AI-use rules for a course or job.
Who it is for
Students, adult learners and working professionals who study, research or write and want AI to help without doing their thinking for them or getting them into trouble. It is written for UK readers, but its sources mainly cover England and it does not cover Scottish qualifications: rules differ by nation, exam board, school, university and employer. No coding is needed.
Before you start
- Comfort with sending a chat assistant a prompt and following up, as taught in the Your first prompts workbook. A free chat assistant is useful for some steps, but every exercise can be done on paper.
Learn in your browser · Chapter 01 of 06
How you learn, and what that means for AI
Before deciding how to use AI, look at what helps people learn. Then ask of every use: does this do the mental work for me, or make me do more of it?
Learning is building a memory you can use later, without help. A 2013 review by Dunlosky and colleagues rated ten study techniques. Two received the top rating, 'high utility': practice testing, called retrieval practice here (trying to recall before you check), and distributed practice, or spacing (spreading study over days). Self-explanation (explaining in your own words) was rated 'moderate utility' because it had not been adequately evaluated in educational contexts. Rereading, highlighting and summarising were among the five rated 'low utility'.
A 2006 review by Cepeda and colleagues pooled 839 assessments of spacing from 317 experiments: the best gap grew as the time until the test grew.
In Roediger and Karpicke's 2006 experiments, university students studied short passages. After five minutes, restudying gave better recall (81% against 75%); after a week, a recall test did (56% against 42%). Students who read a passage about 14 times felt more confident, yet recalled less a week later than students who tested themselves (40% against 61%). Feeling fluent is not learning.
These were short memory tests, before today's AI tools: they show what to protect, not what AI does. The Department for Education (2025) says evidence on students' own use of AI is still emerging. Two field studies help:
Scroll sideways to see every column.
| Source | What it found | Limits |
|---|---|---|
| Bastani and colleagues, Proceedings of the National Academy of Sciences, 2025 (about 1,000 high school maths students, one school) | With a plain assistant, practice scores rose but a later exam without help was 17% below the textbook group's. A hint-giving version showed no significant drop | One school and subject; short term; run in autumn 2023 |
| Kestin and colleagues, Scientific Reports, 2025 (194 Harvard physics students) | A custom tutor with worked solutions behind it beat an active-learning lesson: more learning in less time | One course, autumn 2023; the authors do not expect it to win everywhere |
Try it yourself · Activity 01
5 minRetrieve first, then check
Test yourself without looking back.
- Cover the chapter. For three minutes, write from memory: the two habits rated highest, two rated low, the Roediger and Karpicke results at five minutes and at one week, and how unrestricted AI help can weaken learning.
- Check against the chapter and try again tomorrow.
How did recalling feel compared with rereading?
Worked answer
Retrieval practice and spacing were rated 'high utility'; rereading and highlighting 'low utility'. Restudying won at five minutes (81% against 75%); testing won at one week (56% against 42%). With unrestricted help, maths students scored 17% below the textbook group on the later exam without help; the hint version showed no significant drop. Anything missed is tomorrow's revision list.
Learn in your browser · Chapter 02 of 06
Study workflows: quiz, explain, check
Four workflows keep you doing the thinking. Each has a prompt you can copy.
Attempt first: write your answer before you see the assistant's. Use your own material: notes you wrote, or text you may use, with nothing private in it (see Chapter 6).
1. Quiz me.
I wrote these notes on [topic]. Ask me one question at a time, from recall to 'why' and 'how'. Wait for my answer, then say what I got right, what I missed and which line of my notes covers it. Do not reveal answers before I attempt them.
[paste your notes]2. Explain at different levels, then write your own version.
Explain [concept] for a ten-year-old, a first-year undergraduate and a specialist, saying what you left out each time.
Here is my own explanation. Where is it wrong or vague? Do not rewrite it.3. Check my understanding.
I will explain [concept] in my own words. Ask up to three probing questions. Do not correct me until I have answered.4. Notes into practice questions.
Turn the passage below into 10 questions: 4 recall, 3 'explain in your own words' and 3 'apply it'. Put the answers at the end, each with the sentence it comes from.
[paste your passage]Asking for the source sentence lets you catch invented answers. Check every answer against your notes: an assistant can mark you wrongly too. Then space it: retry the questions tomorrow, in a few days and in a week, adjusting to how long you need to remember.
Try it yourself · Activity 02
15 minPassage to self-test
Turn this passage into self-test questions. It is adapted, in our words, from guidance by JCQ, the Joint Council for Qualifications (April 2025): 'Chatbots write text from patterns in the text they were trained on. Replies are statistically likely to fit the request, so they can sound convincing yet be wrong, and some tools invent references. JCQ says such replies cannot be relied upon and must be independently verified and acknowledged.'
- Write five questions: two recall, two 'explain in your own words' and one 'apply it'.
- Write each answer and the sentence it comes from, then cover the answers.
- Answer them from memory tomorrow.
Which question was hardest to write?
Worked answer
One good set. Recall: what do chatbots base replies on? Patterns in their training text. Recall: what do some tools invent? References. Explain: why can a convincing reply be wrong? It is text likely to fit the request, not text checked against facts. Explain: why verify? Replies cannot be relied upon. Apply: a chatbot gives a reference; what next? Look it up, open the source and check it says what is claimed (Chapter 3).
Learn in your browser · Chapter 03 of 06
Research with AI: fluent, but sometimes false
An assistant can suggest search terms or explain a word. It is not a source: use it for leads, then find real sources yourself. Here is why, and a method to use every time.
These tools write text that is statistically likely to fit your request. JCQ says replies can seem convincing yet be wrong or biased, and that some tools produce fake references. The Department for Education calls false information presented as fact 'hallucination'.
In a 2023 study, Walters and Wilder asked two versions of a popular chat assistant for literature reviews on 42 topics and checked all 636 references: 55% (earlier version) and 18% (later) were fabricated, and 43% and 24% of the real ones had substantive errors. Tools have changed, so treat this as a reason to check, not a rate. The authors note that fabricated references tend to look legitimate at first glance, and that assistants often answer wrongly when asked to verify a citation: never ask an AI to check itself.
Check every claim and citation in five steps.
- Split the text into claims and citations.
- Find each source yourself by its exact title, in a library search or the publisher's site. If you cannot find it, treat it as unverified (it may be invented, or just missing from your index): ask a librarian and do not cite it until you have seen it.
- Match authors, year, title, journal, volume and pages. A real paper with wrong details is still a wrong citation.
- Open the source and find the passage that supports the claim.
- Give a verdict (supported, partly supported, contradicted or not found), rewrite the claim to say what the source says, and cite what you read.
Keep a verification log: one row per claim, with source found, details match, where you looked and verdict.
Try it yourself · Activity 03
20 minVerify a citation end to end
Written for this workbook, not by an assistant. Treat both citations as unchecked: 'A review of 184 experiments found that the best gap between study sessions grew as the delay before the test grew (Cepeda, Pashler, Vul, Wixted and Rohrer, 2006, 'Distributed practice in verbal recall tasks: a review and quantitative synthesis', Psychological Bulletin, 132(3), 354–380). AI-generated quizzes make spacing even more effective (Marchetti and Ellsworth, 2019, 'AI-generated quizzes and long-term retention in undergraduates', Journal of Adaptive Study Practice, 11(2), 45–61).'
- Search for each exact title. For the first, compare every detail with the record and check the abstract against the claim.
- Give each citation a verdict and rewrite the paragraph so that it is true.
Which detail would you have missed if you had only glanced at the reference?
Worked answer
Citation 1 is real and every detail matches the record: Cepeda, Pashler, Vul, Wixted and Rohrer, 2006, Psychological Bulletin, 132(3), 354–380 (DOI 10.1037/0033-2909.132.3.354). The abstract supports the finding about the best gap, but reports 839 assessments in 317 experiments in 184 articles, so '184 experiments' is wrong. Verdict: partly supported. Corrected: 'A review of 317 experiments found that the best gap between study sessions grew as the delay before the test grew.' Citation 2: on 26 September 2026 searches of two public scholarly indexes (Crossref and Europe PMC) and the web found no such work or journal. Verdict: not found, so unverified; delete the sentence. (It was invented for this exercise.)
Learn in your browser · Chapter 04 of 06
Writing: your draft first, then feedback
Writing is where AI use is most tempting and most risky. One rule keeps the work yours: your first draft comes before the assistant.
Drafting is where you find out what you think. If an assistant writes the first version, you inherit its structure, claims and gaps. The maths experiment in Chapter 1 found students used unrestricted help as a 'crutch', but its authors note that writing lacks the objective marking criteria that maths has, and that other settings need more study. Treat draft-first as sound practice, not a proven result.
- Write a rough draft from your own notes, with no AI.
- Ask for feedback, not a rewrite.
- Decide each change yourself: accept, reject or adapt.
- Do the rewriting yourself, in your own words.
- Log what came from where, and keep dated drafts: with the log, they are your best evidence if the work is ever questioned.
For step 2, try these prompts.
Here is my draft. Do not rewrite it. State the main point of each paragraph, then tell me where the order is unclear.
Mark sentences that are hard to follow and say why. Offer no replacement wording.
What is the strongest objection a sceptical reader would raise? Supply no facts or sources.Ask before using even light help such as grammar checking: Jisc's guidance for the education sector notes that students need clear advice on when feedback on writing should be declared.
Keep your voice. JCQ tells teachers that generic content, little specific knowledge or a style unlike your earlier work may raise questions. The aim is not to beat a detector but to make the ideas, examples and judgement yours: add what only you know.
Scroll sideways to see every column.
| Where | What the assistant suggested | What I did | Why |
|---|---|---|---|
| Paragraph 2 | Move the cost argument before the benefits | Moved it and rewrote both sentences myself | It matched my plan better |
Try it yourself · Activity 04
20 minRewrite a paragraph in your own voice
Invented for this workbook, not real assistant output: 'Part-time study offers numerous benefits for modern learners. It provides flexibility, allowing students to balance education with work and family commitments. It also fosters time management skills that employers seek. In today's fast-paced world, it is an invaluable pathway to growth.'
- Underline phrases that could appear in any essay on any topic. List three specifics from your own life or reading, or invent an example learner.
- Rewrite from your specifics, in your words, without looking at the original, and fill in a change log.
- Read it aloud: could you explain every sentence to your tutor?
Which change made the paragraph most clearly yours?
Worked answer
Example learner: a hospital worker on night shifts. Rewrite: 'I study part time because I work three night shifts a week on a ward. I study on Mondays and Thursdays and read on the bus. It is slower than full-time study, but I use what I learn at work the next week.' Change log: 'numerous benefits', 'fast-paced world' and 'invaluable pathway': cut as filler. 'Flexibility, work and family': idea kept, replaced by my real shifts (mine). 'Skills that employers seek': cut, I could not source it. From the assistant: only the order of ideas.
Learn in your browser · Chapter 05 of 06
Integrity: rules, disclosure and detectors
Whoever assesses you sets the rules, and the rules differ. Find them first, follow them, and be open about what you did.
Policy first. Read the assignment brief, course handbook or academic integrity policy, the exam board's rules or your employer's policy, and ask in writing if unsure. 'Not banned' is not 'allowed': permission for feedback may not cover drafting.
Scroll sideways to see every column.
| Source and date | Covers | What it says |
|---|---|---|
| JCQ, AI Use in Assessments (30 April 2025) | Coursework and non-exam assessment for JCQ awarding bodies (GCSE, A level, vocational) | Work must be your own. If you use AI, name the tool and date, keep your questions and its replies as a screenshot and explain how you used it. AI content earns no marks. Misuse is malpractice: sanctions run from a warning to disqualification. Students must not be able to use AI tools in supervised exams |
| Department for Education (updated 12 August 2025) | Schools and colleges in England only | Output can be inaccurate, biased or false; it recommends that personal data is not used in these tools; each school or college chooses its approach |
| Russell Group (3 July 2023); Jisc (January 2026) | Universities and colleges | Appropriate use differs by subject. Students should know when AI use is inappropriate and when to acknowledge it. You are accountable for the accuracy of what you use. Universities aim to let students ask about their use without fear of penalisation |
Disclose. Use the template below, naming the tool and version, not just 'AI'. JCQ's student guide adds: do not sign a declaration that the work is your own until every use is acknowledged.
AI use statement
I used [tool name and version] on [date] to [purpose].
I wrote the first draft myself. It suggested [what]; I [accepted / changed / rejected] this by [what you did].
I checked every fact and reference against the original source and kept my prompts and its replies.
The arguments, examples and final wording are my own.Detectors are not proof. JCQ says detection tools vary in accuracy and belong within a holistic judgement. Jisc says no tool can conclusively prove AI authorship and that they produce false positives. In a 2023 test of 14 tools, none scored above 80% accuracy and the authors judged them unsuitable as evidence of misconduct (Weber-Wulff and colleagues). In another 2023 study, seven detectors wrongly flagged an average of 61.3% of 91 essays by non-native English writers as AI-generated (Liang and colleagues). A score can still start a conversation, so be ready to talk through your work.
If you are wrongly accused (guidance, not legal advice; your own written procedure governs):
- Stay calm, tell the truth and delete nothing: JCQ treats knowingly giving misleading information as malpractice.
- Ask in writing for the allegation, the evidence and the procedure. For JCQ coursework, concerns raised before you sign the declaration of authentication follow your school or college's own policy. After you sign, they go to the awarding body; if there is enough evidence, JCQ's procedure says you must be told the allegation, made aware of the evidence and given time to respond in writing.
- Gather dated drafts, notes, version history, your log and saved chats, and be ready to talk through the work in your own words.
- Get support from your students' union, a parent or carer, or an adviser; JCQ allows a candidate to be accompanied at an interview.
- Know the appeal route. JCQ appeals normally must be made within 14 days of the outcome, through your school or college unless you are a private candidate. For higher education in England and Wales, the Office of the Independent Adjudicator's framework (October 2018) says the provider carries the burden of proof and should allow an appeal.
Try it yourself · Activity 05
10 minDraft a disclosure paragraph
You used an assistant to get feedback on your draft's structure, to make ten practice questions from your notes and to check the grammar of one paragraph. Write a disclosure of 80 to 150 words.
- Use the template, with placeholders for the tool name and version, and never guess a date. Say what you accepted, changed or rejected, and what you checked.
- Check it against JCQ's requirements in the table above: tool name, date, copies kept, brief explanation.
Which use might your course not allow, and who would you ask?
Worked answer
Example: 'AI use statement. I used [tool name and version] on 12/09/2026 for three purposes: feedback on the structure of my draft (I moved my counter-argument earlier and rewrote the paragraphs myself); ten practice questions from my notes, for revision only; and pointing out grammar errors in paragraph 4, which I corrected myself. I wrote the first draft myself, checked every fact and reference against the original sources and kept my prompts and its replies as screenshots. The arguments, examples and final wording are my own.' It gives tool, date, purposes, actions, checks and copies kept.
Learn in your browser · Chapter 06 of 06
Privacy, confidentiality and your own rules
A chat assistant runs on someone else's computers. What you type leaves your control, so decide in advance what never goes in.
The Department for Education says many free tools use what you enter to train their models and that entered data should not identify anyone. The National Cyber Security Centre (March 2023) recommends keeping sensitive information, and anything that would cause problems if made public, out of public tools.
Scroll sideways to see every column.
| Do not paste | Why |
|---|---|
| Other people's personal data: names with details, health information, messages, interview transcripts | The Information Commissioner's Office (ICO) explains that the UK GDPR defines personal data as any information relating to an identifiable person. A code in place of a name is still personal data if it can be linked back |
| Confidential work: employer or client documents, unpublished research, anything you may not share | Jisc advises against putting work that is not your own, or personal information, into free tools, and says personal information must never go into a system your institution has not contracted with |
| Passwords, keys, ID numbers, bank details | No study or writing task needs them |
The stranger test: would I be comfortable if a stranger read this? If not, leave it out. Where your university or employer provides an approved tool, use that for work data and check its settings.
Try it yourself · Activity 06
15 minWrite your personal AI-use rules
Choose one course or job and write a one-page set of rules for it, based on its written policy.
- Find the written policy or brief and note its three key points; if there is none, ask in writing.
- List uses that are fine, need permission first, or are ruled out. Add your data rule and the approved tool.
- Add your habits, then date the rules and set a review date.
Which rule will be hardest to keep when you are short of time?
Worked answer
Example for an invented module, 'Introduction to Sociology'. Policy (invented wording): AI may be used for feedback if declared, not for drafting. Fine: quizzing me on my notes, explaining concepts, feedback on my draft's structure. Ask first: grammar checking, translation. Never: submitting text an assistant wrote, or using a tool in a timed assessment. Data rule: no names or details of any classmate, interviewee or colleague; only the university-approved tool for placement material. Habits: write first, verify every fact and reference, keep a log, add an AI use statement. Written 26/09/2026; review each term.
Keep learning
The complete workbook
This workbook shows how to use an AI assistant to study, research and write without handing over your own thinking. You will use learning research to choose study workflows, learn a repeatable method to verify every claim and citation, and draft, get feedback and keep your own voice. You will also read what UK exam bodies and the government say about AI use, disclose your use honestly and keep other people's data out of AI tools.
- 01How you learn, and what that means for AIRead here · 1 exercise
Before deciding how to use AI, look at what helps people learn. Then ask of every use: does this do the mental work for me, or make me do more of it?
- 02Study workflows: quiz, explain, checkRead here · 1 exercise
Four workflows keep you doing the thinking. Each has a prompt you can copy.
- 03Research with AI: fluent, but sometimes falseRead here · 1 exercise
An assistant can suggest search terms or explain a word. It is not a source: use it for leads, then find real sources yourself. Here is why, and a method to use every time.
- 04Writing: your draft first, then feedbackRead here · 1 exercise
Writing is where AI use is most tempting and most risky. One rule keeps the work yours: your first draft comes before the assistant.
- 05Integrity: rules, disclosure and detectorsRead here · 1 exercise
Whoever assesses you sets the rules, and the rules differ. Find them first, follow them, and be open about what you did.
- 06Privacy, confidentiality and your own rulesRead here · 1 exercise
A chat assistant runs on someone else's computers. What you type leaves your control, so decide in advance what never goes in.
Also inside: a 9-point checklist, a glossary of 12 terms and 10 questions and answers to test yourself. 6 hands-on exercises, each with a worked answer at the back where the workbook gives one.
The PDF is free. We ask for your name and email before downloading. Read the privacy notice at the download form. Free to download and use for personal learning, study groups and inside your own team. Please do not resell the workbooks or republish them as your own. Link people to trust-agent.ai instead.
Get your free PDF workbook
Enter your name and email, then download the complete workbook. This form also works without JavaScript.
Test yourself
Questions and answers
Is using AI to study cheating?
It depends on the rules and the use. Quizzing yourself on your own notes is a different matter from submitting text an assistant wrote as your own. For JCQ-regulated coursework, AI use must be acknowledged and you cannot earn marks for content the AI produced. Check your policy before you start.
How can I study with AI without it doing my thinking?
Attempt first, then ask. Have it quiz you, explain at different levels and probe your understanding, and check its answers against your notes. The evidence is mixed: one field experiment found unrestricted help harmed later exam scores, while a version built to give hints did not.
Why do AI tools invent references?
JCQ explains that chatbots produce replies that are statistically likely to fit a request, so they cannot be relied upon: they can sound convincing yet be wrong, and some produce fake references. A 2023 study found 55% and 18% of references fabricated in two versions. Tools change, so always check.
How do I check a citation properly?
Find the source yourself by its exact title. Match authors, year, journal and pages. Open it and find the passage that supports the claim. Give a verdict, rewrite the claim to match what the source says, and cite what you read, not the AI.
Can I ask the AI whether its own references are real?
No. Walters and Wilder note that earlier studies found assistants often answer wrongly when asked whether a citation is correct. Verify against a library search, a scholarly index or the publisher's page instead.
Do I have to say that I used AI?
Follow your policy. JCQ requires the tool's name and the date, a saved copy of your questions and its replies, and a brief explanation of how you used it. Universities and employers vary. If unsure, ask, and disclose.
Can AI detectors prove that I used AI?
Not on their own. Jisc says no tool can conclusively prove it and that they produce false positives. A 2023 test of 14 tools found none above 80% accuracy. JCQ says scores belong within a wider judgement. A score can still start a conversation.
What should I do if I am wrongly accused?
Stay honest and delete nothing. Ask in writing for the allegation, the evidence and the procedure. Gather dated drafts and notes, write your own account, get support from your students' union or an adviser, and use the appeal route within its deadline. Your written procedure governs; this is general guidance, not legal advice.
What can I safely paste into an AI tool?
Your own notes with nothing private in them, text you are allowed to use, or made-up examples. Keep out other people's personal data, confidential work and passwords, and anything that would be a problem if public. Use tools your institution or employer approves.
Does this workbook tell me which AI tool to use?
No. It does not rank or recommend tools. Use whichever your school, university or employer approves, read its privacy terms, and see The free AI toolkit workbook for how to compare tools and check their terms.
Write as you go
My notes
Use this scratchpad for your own notes, working or answers as you read. Notes are saved in this browser on this device. Downloading makes a local copy; it does not send your notes to Mickai.
Consolidate your learning
Review your work
Work through the exercises and questions, then compare your reasoning with the worked answers. A scored exam and verifiable certificate are not yet available for this course.
Review the questionsLearn the language
Key terms
- Retrieval practice
- Trying to recall something from memory before you check it, for example with a quiz or a blank-page summary. The 2013 review calls it practice testing.
- Spaced practice
- Spreading study over several sessions instead of one. Also called distributed practice.
- Self-explanation
- Explaining, in your own words, how or why something works while you study it.
- Hallucination
- When an AI presents something false or invented as though it were true, such as a reference to a work that does not exist. The Department for Education uses the word for nonsensical, inaccurate or false information presented as fact.
- DOI
- Digital object identifier: a permanent code for a published paper or book. Searching for it takes you to the official record.
- JCQ
- The Joint Council for Qualifications, the joint body of UK awarding organisations such as AQA, OCR, Pearson and WJEC. Its regulations cover their exams, coursework and non-exam assessments.
6 of the workbook's 12 terms. The complete glossary is in the workbook.
Follow the evidence
Sources and checks
Facts last checked: .
These workbooks use AI assistance. See how the workbooks are made.
- AI Use in Assessments: Your role in protecting the integrity of qualifications (revision two, 30 April 2025)Joint Council for Qualifications (JCQ)
- AI and Assessments: a quick guide for students (effective from September 2026)Joint Council for Qualifications (JCQ)
- Suspected Malpractice: Policies and Procedures (effective September 2026)Joint Council for Qualifications (JCQ)
- Generative artificial intelligence (AI) in education (last updated 12 August 2025)Department for Education, GOV.UK
- Principles on the use of generative AI tools in education (3 July 2023; full text in the PDF linked from this page)Russell Group
- A Generative AI Primer (version 1.7, January 2026)Jisc National Centre for AI
- Good Practice Framework: Disciplinary procedures (October 2018)Office of the Independent Adjudicator for Higher Education (OIA)
- ChatGPT and large language models: what's the risk? (14 March 2023)National Cyber Security Centre (NCSC)
- What is personal data?Information Commissioner's Office (ICO)
- Improving Students' Learning With Effective Learning Techniques: Promising Directions From Cognitive and Educational Psychology (Dunlosky, Rawson, Marsh, Nathan and Willingham, 2013)Psychological Science in the Public Interest, 14(1), 4–58 (abstract read via PubMed)
- Distributed practice in verbal recall tasks: A review and quantitative synthesis (Cepeda, Pashler, Vul, Wixted and Rohrer, 2006)Psychological Bulletin, 132(3), 354–380 (abstract read via PubMed)
- Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention (Roediger and Karpicke, 2006)Psychological Science, 17(3), 249–255 (author-hosted copy, Purdue University)
- Generative AI without guardrails can harm learning: Evidence from high school mathematics (Bastani and colleagues, 2025)Proceedings of the National Academy of Sciences, 122(26), e2422633122 (open-access copy, PubMed Central)
- AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting (Kestin and colleagues, 2025)Scientific Reports, 15, 17458 (Nature Portfolio)
- Fabrication and errors in the bibliographic citations generated by ChatGPT (Walters and Wilder, 2023)Scientific Reports, 13, 14045 (Nature Portfolio)
- Testing of detection tools for AI-generated text (Weber-Wulff and colleagues, 2023)International Journal for Educational Integrity, 19, article 26 (Springer)
- GPT detectors are biased against non-native English writers (Liang and colleagues, 2023)Patterns, 4(7), 100779 (Cell Press; open-access copy, PubMed Central)
Created by Mickarle Wagstaff-Irons - Micky Irons with the Mickai team. Published by Mickai LTD. Last updated 26 September 2026.
NextKeep going
Where to go next
Recommended for you
The free AI toolkit: how to choose tools and stay safe
How free AI tools work, how to choose and test them, how to check their current terms and privacy, how to spot fake apps and scams, and how to build your own starter stack.
Recommended for you
Which AI model should I use right now? A living, refreshed comparison
A refreshed, six-factor framework for choosing an AI model: reasoning, effort, context length, multimodal support, latency and data handling, tested on your own work.