At 2:47 in the morning, six days before her economics exam, Ayesha sent me a photo of her kitchen table.
You couldn’t see the table. It was slides — printed slides, which tells you everything about the level of panic — plus two highlighters run dry and a notebook where she’d copied out the same chapter twice, in two colours. She’d been at it since dinner. Her mock test that afternoon had come back at 41%. The exam that actually counted was on Monday.
“Tell me what to download,” she wrote. “Anything. Everything.”
Here’s what I wish someone had told me in my first year: she didn’t have a knowledge problem. She had a method problem. Fourteen weeks of lectures had gone in one ear and onto slides she’d never once tested herself on. Everything she’d done that night — re-reading, highlighting, copying — felt like studying. It was studying the way pushing a car feels like exercise. Real effort. No distance covered.
So we didn’t download everything. We did less, in a different order, most of it free. This is her week, flops included — the flops are where the actual advice is hiding.
The first thing she tried made it worse
Full honesty: Ayesha’s first move, before she texted me, was to open ChatGPT and type “write my economics assignment.” Then she deleted it. Not out of pure virtue — out of timing. Two weeks earlier, a boy in her tutorial group, Hamza, had submitted a generated essay and got called in. Zero on the assignment, a formal warning, and the special humiliation of a lecturer reading two sentences aloud: “You didn’t write this. You don’t use the word ‘furthermore’.” Her university’s rule, like most these days, isn’t “no AI.” It’s: AI for revision, brainstorming, understanding your material — fine. Generated work with your name on it — misconduct. Check your own institution’s policy; they vary, and “everyone was doing it” has never once worked as a defence.
Ground rule settled, the scary way. Everything after that night was the legal stuff.
Her own lectures, talking back
The first real change was NotebookLM — the most underrated tool on this list, because it isn’t working from the internet’s version of economics. It’s working from hers. She uploaded her lecturer’s actual slides, all fourteen weeks, plus the course handout with the exam topics circled.
Then she could ask it things like “which topics connect to the elasticity chapter?” and get answers pointing back at specific slides. Slide 34. The ugly graph. That matters, because a summary of the internet tells you what’s generally important; a summary of your own course tells you what your lecturer taught — which is, conveniently, what the exam is on. Twice it smoothed over a detail her lecturer had put in bold. She caught both because the citation was right there. That’s the relationship you want: useful, supervised, never quite trusted.
The samosa breakthrough (yes, really)
Supply and demand had defeated her for a month. Curves shifting, curves sliding along themselves — she could recite the definitions and understand none of them. So she asked ChatGPT something gloriously stupid: explain it using the samosa stall outside the campus gate.
And it worked, in about ten minutes, because it finally had a place to live. Rainy day, fewer students walking past, the stall-owner’s nephew drops the price to move stock before it goes cold — demand, supply, equilibrium, all standing in a queue smelling of frying oil. The dumber and more personal the example, the harder it sticks.
The trick is what she did next — the difference between the tool teaching you and the tool entertaining you. She typed: “Now ask me questions. One at a time. Don’t tell me the answer until I’ve tried.” Ten questions. Four wrong, including one about her own samosa stall, which stung productively. An explanation you nod at is worth almost nothing. A question you get wrong in your kitchen at 1 a.m. — that, you remember. (She does this with Gemini too. Pick whichever’s already open.)
Tuesday worked. Thursday really didn’t.
Tuesday evening she turned her wrong answers into flashcards — Quizlet first, because it’s quick, and then Anki, which looks designed in 2006 and has the personality of a tax form, but whose spaced repetition is ruthless: each card returns right around the moment you’re about to forget it.
One card tormented her. “A tax on cigarettes raises the price. Why doesn’t quantity fall much?” — inelastic demand, the answer she’d written on the back in her own words. Wrong Monday night. Wrong Tuesday. Wrong Wednesday, twice, the second time with real confidence, which is the worst kind of wrong. She got it right Thursday while brushing her teeth and stood there, toothbrush in mouth, pointing at her phone.
That’s active recall doing its unglamorous work. Dragging the answer out of your own brain builds a memory re-reading never touches. It also produces a wrong-answers list, and by Friday that list was the most honest revision plan she’d ever owned. Not “Chapter 7, somehow.” These eleven things. Exactly these.
Thursday, though, she skipped the cards, decided she “basically knew it now,” and re-read her notes for two comfortable hours instead. Friday’s quiz with me: three of the eleven had evaporated. Including the cigarette tax. I wish I could say she was surprised.
The research detour, and one confident lie
Her coursework project needed sources, and for that she used Perplexity, for one simple reason: it shows where its answers came from. Actual links. Clickable. She built half her bibliography from pages she genuinely opened — which matters, because one afternoon Perplexity summarised a source as saying the opposite of what it said. Cheerfully. With the link right there, like it was daring her to check. She checked. Had she pasted that claim in unread, that’s her name on a wrong fact with a citation attached — worse than no citation at all. Treat every confident sentence these tools produce as “probably true, pending ten seconds of verification.” Especially the ones with footnotes. Especially those.
Grammarly entered at the very end, on a draft she’d written herself — words hers, argument hers, commas an ongoing disaster. It flagged the same comma splice ten times in two pages. Tenth time, she fixed it before the underline finished rendering and laughed out loud. What Grammarly didn’t do was write a single sentence of it — which is why, when her lecturer returned the work, the margin read: “Messy in places. But this sounds like you. Keep going.” It mattered more than the mark. And the mark — her experience, not a promise about yours — went from the low 40s on that mock to 68% on the real paper that Monday. Her tutor’s verdict: not a miracle. Just the first time her revision and the exam had been about the same subject.
Oh — and maths. Her stats questions went through Photomath and Khanmigo under one rule: never “what’s the answer,” always “why is this step wrong?” The steps are the learning. The answer is just the last line of it.
Questions students actually ask me
Be honest — will any of this get me in trouble?
Revision, summaries of your own notes, practice questions, explanations — fine at nearly every institution I know of. Submitting generated writing as yours is where the ground disappears, fast. Read your institution’s AI policy tonight, not after the email arrives. Hamza would endorse this message.
If I only download one thing, what is it?
A free chatbot — ChatGPT or Gemini — because explanations and self-quizzing fix the biggest hole in most studying. Then NotebookLM, because your own course material beats the internet for exam purposes. Ayesha’s whole setup cost her nothing but a bad Thursday.
Are flashcards actually better than re-reading, or is that just something tutors say?
Ask the cigarette-tax card. Re-reading is comfortable because your brain mistakes “I’ve seen this” for “I know this.” Ayesha had seen inelastic demand roughly forty times. She knew it on the fourth failed attempt to recall it, standing in a bathroom doorway. Comfortable isn’t the same as effective — usually it’s the opposite.
How do I know when it’s lying to me?
You often don’t — which is why you make it work from material you gave it (NotebookLM’s whole deal), prefer tools that show sources (Perplexity), and spot-check anything with a date, a number or a quote against your textbook. The danger isn’t that these tools are sometimes wrong. It’s that they’re wrong in the same confident voice they use for everything else.
Can it write my essay if I just edit it a bit afterwards?
It can produce words, certainly. But the first exam question on the same topic will hold the arrangement up to the light, and “I edited it” is a thin umbrella in that rain. Write the ugly draft yourself. Then let the tools do what they’re genuinely good at — explaining, quizzing, tidying. Different jobs.
Last thing
Ayesha still has the photo of that kitchen table. She shows it to first-years now, the way older students pass down scar tissue and call it wisdom.
Her exam didn’t go well because of an app. It went well because, for six days, her studying finally involved her brain having to produce answers instead of admire them — and because she stopped confusing familiar with known. The tools just removed the friction.
Start where she started. Not with everything — with your own lectures in NotebookLM, one honest quiz, and fifteen minutes of cards today. The secret was never the tools. Honestly? It was never really a secret.