Short answer: use AI to quiz you on your own biology notes, one question at a time, and answer from memory before it reveals anything. Studies of chatbots on real biology exams find they handle text questions well but do noticeably worse on questions built around images, figures and higher-level reasoning, which is exactly where biology exams test you hardest. So check every answer against your notes, and practise diagrams and data questions from your own material, not from the chatbot's description of them.
How good is AI at biology, really?
Better than you might fear on facts, worse than you might hope on everything else. Three studies give a fair picture.
At Hamilton College, Rai, Ngaw and Nannas (2025) gave the assessments from an introductory biology course to ChatGPT-4, ChatGPT-3.5, Google's Bard and Microsoft's Bing. All four passed, but their grades ranged from D− to C−, while the average student got a B. On assessments without image-based questions, the AI grades were a full letter grade higher. The title of the paper says the rest: passing grades, but poor performance at high cognitive complexity.
At graduate level, Stribling and colleagues (2024) ran GPT-4 through nine biomedical science exams. It scored at or above the student average on seven of them and gave fully correct short answers, but it did badly on questions built around figures. On one exam based on figures of simulated data, it scored about 42–45% while students averaged around 79%. Asked about a figure from a published paper it could not see, it sometimes described a figure that did not exist: in 25 repeated attempts, 10 produced detailed invented descriptions.
In a 110-question laboratory enzymology exam, Hasanefendic and colleagues (2025) found that students scored 85.5%, ChatGPT-4.0 scored 74.5% and ChatGPT-3.5 scored 52.7%, below the pass mark. Both chatbots did worst on the laboratory analysis questions.
Put together: a chatbot is a decent quiz partner for the factual layer of biology, an unreliable judge of anything visual, and it can be wrong in a fluent, confident voice. That shapes how you should use it.
What AI is good for in biology revision
- Turning your notes into questions. Biology has a lot of recall: organelles and their jobs, the stages of a process, definitions, equations. A chatbot given your notes can produce plenty of short questions on them.
- Asking you to explain, not just name. "Why does the rate level off?" is a better question than "what is the rate?". AI can generate the why-questions if you ask for them.
- Comparing similar ideas. Mitosis and meiosis, aerobic and anaerobic respiration, active transport and diffusion. Asking for questions that force you to tell the two apart is one of the most useful things it does.
- Re-asking what you got wrong. Paste yesterday's misses and ask for the same ideas in a different form.
What it should not do is the remembering for you. Reading an AI summary of a chapter feels productive but is just rereading in a new font. The benefit of studying comes from pulling the answer out of your own memory: retrieval strengthens long-term retention much more than restudying, especially over delays of days to weeks (Antony et al., 2017).
A biology revision routine with AI
- Pick one topic and give it your material. Paste or upload the notes, slides or textbook section you will be examined on. Say your level and exam board, because a chatbot left to itself often drifts into university-level detail you do not need.
- Ask for one question at a time. A list with answers underneath invites you to read ahead.
- Answer from memory first. Write or say your answer before you look at anything. If you have no idea, say so, then look it up in your notes.
- Check the answer against your notes, not the chat. Especially numbers, names and the exact wording of definitions, which mark schemes care about.
- Keep a list of misses and come back to it. Ask those again tomorrow and in a few days. Our spaced repetition schedule gives sensible gaps.
Prompts that make it test you
- "I'm revising [topic] for [exam and board]. Using only the notes below, ask me one question at a time. Wait for my answer, then tell me if I'm right and quote the line from my notes that shows it."
- "Ask me five questions that make me explain why, not just what, about this section."
- "Give me questions that test whether I can tell [process A] from [process B]."
- "Here are the steps of [process] from my notes. Shuffle them and ask me to put them back in order."
- "I got these wrong yesterday: [list]. Ask them again in a different way."
"Using only the notes below" and "quote the line" are the two phrases that matter. They keep the questions on your syllabus and make a wrong answer easy to catch. Our guide to checking AI study answers has more on catching errors.
Diagrams, micrographs and data: do these yourself
Biology exams lean on visuals: labelling a cell, reading a graph, identifying the stage of mitosis in a micrograph. The AQA A-level specification, for example, asks students to recognise the stages of the cell cycle and explain the appearance of cells in each stage, and one required practical has students identify mitotic stages in root tip squashes under a microscope. These are the questions the research above says chatbots handle worst.
So for the visual layer, practise from the real images: the diagrams in your textbook, your teacher's slides and the figures in past papers. Cover the labels and name each part, then check. Sketch a process from memory and compare it with the original. Our guide to image occlusion flashcards shows a quick way to turn diagrams into self-tests, and the required practicals guide covers the method questions.
This is also where Ace fits. It turns your own PDF, slides or Word notes into practice questions, and when your material has a labelled diagram, it can ask you to label it using a screenshot from your own file, never a generated picture. For processes it can ask you to put the steps in order, and under each answer it shows the sentence from your notes that backs it, so you are always checking against your course rather than the internet.
Frequently asked questions
Is ChatGPT good for biology revision?
It is good for quizzing you on facts and explanations from your own notes. Studies find it does much worse on image-based and higher-level questions, and it can be confidently wrong, so always check its answers against your material.
Can AI read biology diagrams accurately?
Not reliably. In the studies above, chatbots scored lower on figure-based questions than on text, and one model invented descriptions of a figure it could not see. Practise diagrams from your textbook and past papers instead.
Should I let AI write my biology notes?
A summary can help you see the shape of a new topic, but reading it is not revision. Spend the time answering questions from memory instead.
Is it cheating to use AI for revision?
Using it to test yourself is not cheating. Submitting AI-written work as your own in coursework is a different matter, and your school or exam board's rules apply. See our guide to using AI without cheating.
Sources
- Rai, M., Ngaw, M., & Nannas, N. (2025). Artificial intelligence performance in introductory biology: Passing grades but poor performance at high cognitive complexity. Education Sciences, 15(10), 1400. Summary: Hamilton College.
- Stribling, D., Xia, Y., Amer, M. K., Graim, K. S., Mulligan, C. J., & Renne, R. (2024). The model student: GPT-4 performance on graduate biomedical science exams. Scientific Reports, 14, 5670. PMC.
- Hasanefendic, B., Pasic, A., Duskan, S., Sehercehajic, E., & Jazic Durmisevic, A. (2025). ChatGPT answers the 110-question laboratory enzymology student exam: Pass or fail? Cureus, 17(4), e82168. PMC.
- Antony, J. W., Ferreira, C. S., Norman, K. A., & Wimber, M. (2017). Retrieval as a fast route to memory consolidation. Trends in Cognitive Sciences, 21(8), 573–576. PMC.
- AQA. AS and A-level Biology (7401/7402) specification, 3.2.2 All cells arise from other cells. aqa.org.uk.



