How AI Agents Can Help You Study in 2026: A Revision System Built on What Actually Works (Anki, Quizlet, NotebookLM, Jobbit)

Most revision feels productive and is not. Rereading the chapter, highlighting the textbook until it glows, copying notes out neatly: when Dunlosky and colleagues reviewed ten study techniques for Psychological Science in the Public Interest in 2013, all three of those landed in the low-utility group. The two techniques that came out on top, practice testing and spreading your study over time, are the ones students do least, because they are effortful and they need someone to write the questions, mark the answers and keep the schedule. That is exactly the work an AI agent is good at. Studying with an AI agent is not about getting essays written for you. It is about having a tireless examiner who quizzes you, marks you, explains what you got wrong and tells you when to come back to it.

This guide is about how to study with AI, the system rather than the tool list. It shows you how to set up an agent with your own notes and syllabus in twenty minutes, how to turn any topic into practice tests and a spaced timetable, how to mark past papers against a real mark scheme, how to use AI for essays without breaking JCQ or university rules, and which apps (Anki, Quizlet, NotebookLM, ChatGPT, Claude, Jobbit) do which part of the job. If you want a ranked list of tools instead, read the best AI tools for students; this post is about how to study with them.

What the research says works, and what an agent can do about it

The 2013 review by John Dunlosky, Katherine Rawson, Elizabeth Marsh, Mitchell Nathan and Daniel Willingham is still the most cited ranking of study techniques, and its findings have held up in later meta-analyses. Here are all ten, with the rating the authors gave each one and what an AI agent adds:

The pattern is simple. The high-value techniques are the ones that make you retrieve information from memory and come back to it later. They are also the ones that need a second person. An agent is that second person, available at 11pm the night before a mock.

Set up your study agent in twenty minutes

An agent is only as good as what it knows about your course. Give it the right material once and every later session gets better.

1. Create a project for the subject. In Jobbit, a project holds shared files, instructions and notes for every chat inside it; see Jobbit Projects. Make one per subject, not one for everything.
2. Upload the syllabus or specification. For GCSE and A-level, the exam board specification is a public PDF; for university, the module handbook and reading list. This is what stops the agent quizzing you on things that will never be examined.
3. Add your notes and the textbook chapters you are allowed to use. Photos of handwritten notes work; the agent reads them. Lecture slides and recordings can go in too.
4. Add past papers and mark schemes. These are the single most valuable files. Exam boards publish several years of both for free.
5. Write the instructions. Three lines are enough: "I am revising for [exam] on [date]. Quiz me from my notes and the specification, one question at a time, mark strictly against the mark scheme, and never write an answer for me unless I ask for a model answer after attempting it."

That last instruction matters. An agent that writes your answers for you is a faster way to fail; an agent that makes you write them is a tutor.

Active recall: make the agent quiz you

Practice testing works because the act of retrieving something from memory strengthens it, and it works even when you get the answer wrong, as long as you find out the right answer afterwards. The agent's job is to keep asking. Prompts that work:

• "Quiz me on [topic] with 10 short-answer questions, one at a time. Wait for my answer before showing the next question. Mark each answer out of the marks available and explain any gap."
• "Turn my notes on [topic] into flashcards: question on the front, a two-sentence answer on the back. Give me a CSV I can import into Anki."
• "Ask me the questions I got wrong yesterday, in a different order and with different wording."
• "Give me a 6-mark exam-style question on [topic], then mark my answer like an examiner would."

Two rules make this work. Answer out loud or in writing before you look, every time; the moment you read the answer first it stops being a test. And keep the list of what you got wrong; ask the agent to hold it in the project notes so tomorrow's session starts with it.

Spaced repetition without the admin

Distributed practice is the second high-utility technique, and the reason apps like Anki exist. The evidence says a topic reviewed on day 1, day 3, day 7 and day 21 sticks far better than four hours on day 1. The problem has always been the scheduling. Ask the agent:

• "Here is my exam timetable and the list of topics in the specification. Build a revision timetable from today to the exam that covers every topic at least three times with growing gaps, in 45-minute sessions, with Sundays off. Give it to me as a table."
• "Each morning, tell me which topics are due for review today based on the timetable and what I got wrong last time."
• "I missed two days. Rebuild the plan without cramming everything into this week."

If you already use Anki or Quizlet for flashcards, keep them: they schedule reviews well and work offline. Use the agent to generate the cards from your notes and to run the sessions the apps cannot, such as long-answer questions and essay plans.

Past papers and mark schemes: an examiner in your pocket

Past papers are the closest thing to the real exam, and most students waste them by doing the paper, glancing at the mark scheme and moving on. With the mark scheme in the project, the agent can mark properly:

1. Do a question under timed conditions, on paper if that is how the exam works.
2. Type or photograph your answer and ask: "Mark this against the mark scheme for question 4 of the June 2024 paper. Show which marking points I hit, which I missed, and what a full-mark answer includes that mine does not."
3. Ask for the pattern after five or six questions: "Across the questions I have done this week, what am I consistently losing marks on?"

This is where an agent beats a static app. It reads your actual answer, it knows the actual marking points, and it can explain the difference in plain English. For GCSE and A-level, the exam board mark schemes are free to download; for university, ask your lecturer for past exam questions and marking criteria, which many departments publish.

Understanding, not just memorising

Elaborative interrogation and self-explanation, the two moderate-utility techniques, are about asking why and explaining how. They are the fastest way to find out that you do not really understand something you can recite. Try:

• "I am going to explain [concept] to you as if you were a 14-year-old. Stop me at the first sentence that is wrong or vague." This is the Feynman technique with a listener who does not get bored.
• "Ask me why [fact] is true, then keep asking why until I cannot answer, and show me where the gap is in my notes."
• "Give me three worked examples of [type of problem] with increasing difficulty, then a fourth for me to solve alone."
• "Take these two topics from different parts of the course and set me a question that needs both." That is interleaving, and it is what the hardest exam questions do.

For essay subjects, ask the agent to argue the other side: "Give me the strongest objection to the argument in my essay plan." You will meet that objection in the exam anyway.

Essays and coursework without breaking the rules

The rules are clearer than most students think, and the penalty for getting them wrong is severe. For GCSE, A-level and other regulated qualifications in England, Wales and Northern Ireland, the Joint Council for Qualifications' guidance on AI use in assessments (revision 2, published April 2025 and current for 2025 to 2026) says that you may use AI as a source, but any AI-generated content you use must be referenced, you must keep a copy of it, and you must submit a short explanation of how you used it with your work. Passing off AI output as your own is malpractice. Universities set their own policies, and they vary by module, so check the assessment brief before you start.

Within those rules, an agent is a legitimate and powerful help:

• Planning. Ask for a structure, counterarguments and the questions your essay needs to answer, then write it yourself.
• Research. Ask the agent to find and summarise sources with citations, then read the sources. Never cite a paper you have not opened; agents can still produce plausible references that do not exist. The deep research guide shows how to get real sources with links.
• Feedback. Paste your draft and ask for the three weakest paragraphs and why. That is editing help, not ghost-writing, and it is what a good tutor would do.
• Presentations. For the seminar or the viva, the agent can build the slides from your own essay; see how to make a presentation with AI.

Keep a folder with every AI output you used and a one-line note of what you did with it. It takes seconds and it is exactly what JCQ asks for.

A four-week exam plan

Here is the system in calendar form, for one subject with an exam in four weeks. Scale it up by running one project per subject.

• Week 1: map and test. Upload everything, ask for a topic list from the specification, take a 30-question diagnostic quiz across the whole course, and let the agent rank your topics from weakest to strongest. Build the timetable from that ranking, not from the order of the textbook.
• Week 2: retrieve and explain. Two 45-minute sessions a day: one quiz on a weak topic, one self-explanation session on a topic you think you know. Log every wrong answer in the project notes.
• Week 3: past papers. One timed paper or half-paper a day, marked against the mark scheme by the agent, followed by a ten-minute review of the consistent losses. Interleave: pick questions from different topics.
• Week 4: spaced review and sleep. Shorter sessions, only the topics and question types that are still failing, plus a final pass on the whole specification through rapid-fire questions. Stop new material three days out; the evidence on sleep and memory is stronger than any last-night cram.

Mistakes that make AI worse for studying

• Letting it answer first. If you read the answer before you try, you have done rereading, the technique the research rates lowest.
• Trusting citations without opening them. Always click through.
• Quizzing on the wrong material. Without the specification in the project, the agent will happily test you on topics your exam board dropped years ago.
• Passive use. Asking "explain photosynthesis" and reading the reply feels like study. Asking "quiz me on photosynthesis" is study.
• One giant chat. Long chats get slow and the agent loses track of what you got wrong; use a project with notes and start a fresh chat each session.
• Ignoring the rules. Referencing AI use takes a minute; an academic misconduct hearing takes a term.

Which tool for which job

The practical combination for most students is Anki for cards, the exam board's own past papers for practice, and one agent that holds your notes and does the quizzing, marking, explaining and scheduling. Jobbit pricing explains what the free plan covers; for most revision it is enough.

Frequently asked questions

Can an AI agent really help me study, or is it just cheating?

It depends entirely on what you ask it to do. Asking it to write your essay is cheating and, under JCQ and university rules, malpractice. Asking it to quiz you, mark your practice answers against a mark scheme, build a spaced revision timetable and explain what you got wrong is tutoring, and it uses the two techniques research rates most effective: practice testing and distributed practice.

What is the best way to use AI for revision?

Put your specification, notes, past papers and mark schemes into one project, then make the agent test you rather than teach you: one question at a time, you answer before you see the answer, it marks strictly, and it keeps a list of what you got wrong for the next session.

How do I make flashcards with AI?

Ask the agent to turn a set of notes into question-and-answer cards and to give them to you as a CSV. Import that into Anki, which schedules reviews with spaced repetition. Check the cards before you import; a wrong card is a wrong memory.

Can I use AI for GCSE or A-level coursework?

Yes, within the JCQ guidance: AI use must be referenced, you must keep a copy of the AI content you used, and you must submit a short explanation of how you used it. Your teacher must be satisfied the work is your own. Check with your school for how they want it recorded.

Does AI help with university essays?

For planning, finding sources, structuring an argument and getting feedback on a draft, yes. For writing the essay, no: most universities treat unacknowledged AI text as misconduct, and the essay is where the learning happens. Always open every source the agent cites.

Is Anki or an AI agent better for studying?

They do different jobs. Anki is the best tool for daily flashcard reviews on a spaced schedule. An agent is better for everything a flashcard cannot do: long-answer questions, marking against a mark scheme, explaining why you were wrong, planning the timetable and building the cards in the first place. Use both.

How do I start?

Pick the subject with the nearest exam, gather the specification, your notes and two past papers with mark schemes, and open jobbit.uk. Make a project, upload the files, and type: "Quiz me on this course, one question at a time, and mark my answers." The first session tells you where you stand.