Entrance Exams with AI: A Plan, Practice, and Mistake Analysis

Entrance exams have one unpleasant trait: they don't test what you know, they test what you can show in ninety minutes on a paper you've never seen before. That's why a sharp kid sometimes does worse than a classmate who knows less — the classmate just practiced the format, watched the clock, and didn't let the first confusing question rattle them.

This guide is about building your prep as a system instead of panic. There's one core idea: don't keep re-studying what you already know — train on official tests, then break down your mistakes by root cause and drill exactly what's tripping you up. AI is an excellent sparring partner for this: it explains things step by step, generates similar problems, sorts your mistakes into categories, and builds you a plan. What it never does is replace the exam provider — the questions you train on have to be official, not invented in a chat window.

This guide applies to entrance exams (like the Czech unified exams, SAT/ACT, 11+ or your country's equivalent) as much as it does to college admissions tests (aptitude tests and subject tests) — the phases are the same, only the subject matter differs. You don't have to read it in one sitting; come back to whichever phase you're in. Parents should read Phase 6.

A typical scenario

Adéla is in ninth grade and applying to a selective high school. Five months remain until the exam. So far, her routine looks like this: she opens her workbook in the evening, flips through it for a while, does a few problems she's already comfortable with, and closes it feeling like she “did something.” Her dad asks how it's going, she says “fine,” and they both pretend that's true.

The change wasn't dramatic. Adéla sat one complete official test from past years under real time conditions, at a desk, with no phone. She scored 28 out of 50 — and, more importantly, it turned out that of the twenty-two points she lost, only about a third were genuine gaps in knowledge. The rest were calculation slips, two questions she never finished reading because of time, and one where she missed the word “not” in the instructions. Four different problems, each with a different fix.

From that point on she had a system: every other Saturday, one full practice test under real conditions, immediately followed by a mistake analysis sorted by cause, and from that a plan for the next two weeks. She had every wrong answer explained step by step, had similar problems generated for her weak spots — and had every one of those double-checked, because a few of them turned out to be badly written. In three months she went from 28 to 41 points, and most of that gain came from fixing carelessness and time management, not from learning new material.

Marek is dealing with the same thing four years later, on a general aptitude test for college admissions. His problem isn't knowledge — it's pace: on his first practice test he spent twelve minutes on one analytical question and had five minutes left for the last eight questions. The rest of this guide is the process both of them used.

Phase 1: get the facts straight and build a plan backward from the date

Prep that doesn't start with the actual date and format of the exam is just random studying. Before you open your first workbook, you need to know exactly what's being tested, when, how long it takes, and how it's scored.

Start with real facts, straight from the source

This is the one part of your prep where AI has nothing useful to say. Dates, test content, time limits, scoring rules, and even how many applications you're allowed to submit change from year to year — and a model reciting them from memory could easily be describing information that's three years out of date. Get your information from the primary source:

• A national entrance exam: official information from the exam provider (in the Czech Republic, Cermat) and the specific school's website. A school can add its own component on top of the standard exam — a talent test, an interview, points for grades — and you'll only find that on the school's own admissions criteria for that particular year.
• College admissions: the specific faculty and program's website. Requirements vary even between programs within the same faculty. For aptitude tests, also check the test provider's website.

One practical detail worth confirming right away: whether the test deducts points for wrong answers. That decides your entire end-of-test guessing strategy, and it's information you'll find in the test instructions, not on a forum.

Once you have the official text in front of you, it's worth turning it into a clear reference card — but strictly from what's actually written in it:

I'm pasting in the exact official information about the entrance exam,
which I downloaded from the [exam provider / faculty] website:

[paste text]

Turn this into a clear reference card:
1. What's being tested (subjects, test sections, question types)
2. How many minutes per section
3. Scoring: total points, whether points are deducted for wrong
   answers, whether there's a passing threshold
4. Deadlines and what I need to submit by when
5. What I'm allowed to bring (calculator, formula sheets, drawing tools)

Strict rule: use ONLY information from the pasted text. Anything not
in there, mark as "NOT IN SOURCE — needs to be looked up" and don't
fill it in from your own knowledge, even if you think you know it.

This returns a one-page overview you can print and hang over your desk. The last paragraph of the prompt is the important part — without it, the model fills gaps with a plausible-sounding guess. And if anything important shows up as “NOT IN SOURCE,” that's your list of questions for the school.

Baseline diagnostic: one test, no illusions

Before you plan anything, you need a starting point — and you can't estimate that, you need one complete test from past years, taken under real time conditions. Without it, you'll spend five months practicing what you enjoy instead of what you're missing.

Grade the test against the official answer key and record the result by section: how many points from geometry, how many from word problems, how many from spelling, how many from reading comprehension. A summary number like “28 out of 50” tells you nothing — a breakdown by topic is what a plan needs.

A calendar built backward from the date

Build the plan from the end. First the date, then the mandatory milestones, then a buffer, and only then does the actual material fill the remaining space.

I'm preparing for [a national entrance exam for a four-year program /
a college admissions test for program X]. The date is [date], today
is [date].

My baseline from the practice test:
[list points by topic, e.g. geometry 4/12, word problems 6/10,
spelling 8/10, reading comprehension 5/12]

I can realistically give [4] hours a week to prep, except for
[a period when I have zero time, e.g. ski trip Feb 10–17].

Build me a week-by-week plan from today to the exam date:
- what topic I practice each week (specific, not "math")
- how much time to give each topic based on how many points
  I'm losing on it
- when I sit a full practice test (schedule these regularly)
- what should be done by the end of each week so I can tell
  I kept up

Leave the last two weeks before the date for review and calm —
don't schedule any new material into them. Output as a table:
week, date, topic, deliverable.

This returns a schedule you can copy into a calendar or print out. Check two things by hand: whether the most time went to the topics where you're losing the most points (the model sometimes splits time evenly instead), and whether the practice tests aren't all stacked at the end — you need them spread throughout, or you get no feedback along the way.

A weekly rhythm you can actually keep

A five-month plan looks nice, but it's kept week by week. A rhythm that works: three to four shorter practice blocks (30–45 minutes, one topic each), one block for reviewing old mistakes, and a full practice test every other week. Short blocks spread across the week beat one long Sunday session — that's covered in spaced learning.

Here's my weekly prep plan for the entrance exam:
[paste the week from the table]

Break it down into concrete blocks by day. My schedule:
[when I have clubs/activities, when I get home from school,
when I have practice/training]

Conditions:
- no block longer than 45 minutes
- every block has a clear start and end: what I'll do and how
  I'll know it's done (e.g. "10 area problems, checked against
  the answer key")
- one block a week is set aside for revisiting old mistakes
- schedule nothing on [day], that's a day off

For each block, write one sentence on why it's there.

This returns a week broken into blocks you can check off. The last point matters psychologically: a block you understand the reason for is harder to skip than an item on a list.

Phase 2: official practice tests

This is the single most effective part of prep, and also the place people most often get wrong — either they skip practice tests entirely, or they take them the wrong way.

Where to get real questions: from the exam provider, not from a chat

Never generate test questions in a chat. Not because it's forbidden, but because it wrecks your calibration: a model will produce questions that look like exam questions but have a different difficulty level, a different topic distribution, and a different way of phrasing things. Training on an imprecise copy teaches you something other than what will actually be on the test.

Where real questions actually live:

• National entrance exams: the official exam provider (Cermat, in the Czech case) publishes past questions and answer keys — the single most valuable material there is, because it's written by the same provider that will write yours.
• Colleges: most faculties publish sample or past test versions, often with solutions — look in the applicants' section of the site.
• Aptitude tests: the test provider usually offers sample versions and practice tests.
• Workbooks and prep courses are a supplement, not a substitute — quality and difficulty vary widely.

Download everything you can get your hands on and save it in one folder. Use part of it for practice now, and save part of it for the last month as fresh tests you haven't seen yet.

How to actually take a practice test

A practice test is only worth something if it resembles the real exam. That means: at a desk, paper and pencil, a timer set to the official time limit, your phone in another room, no interruptions, no peeking at your notes. If you take it comfortably with your resources handy, you're measuring something you won't have on exam day.

When you finish, set the test aside for ten minutes, then grade it against the official answer key. And most importantly: before you look at the key, go through the test one more time and note, for every question, how confident you were — sure / guessed / had no idea. That one note makes your mistake analysis twice as valuable, because it separates “I know this but botched it” from “I had no clue.”

Scheduling your tests over time

I have [12] weeks left before my exam and [8] official past tests
available ([subject/subjects]).

Schedule when to take each one so that:
- the first one happens this week as a baseline measurement
- the tests are spread out evenly, not stacked at the end
- the last two are ones I've never seen before, and fall in
  the final three weeks before the date
- there's always at least a week between tests to practice
  whatever showed up as a weakness on the previous one

For each test, tell me what to write down afterward and how
I'll know I'm improving. Output as a dated list.

This returns a test schedule you can put in your calendar. Check one thing: whether there's enough time between tests to actually practice. Practice tests by themselves don't teach anything — they only measure. The learning happens in between them.

Phase 3: mistake analysis by root cause

This is the core of the whole guide. Most people grade a test, think “oh, I got that one wrong,” redo the problem, and move on. That solves nothing, because they never asked why the mistake happened.

Four causes, four different fixes

Almost every mistake on a test falls into one of four categories, and each needs a completely different fix:

• Not knowing. I didn't know how to do it. Fix: learn the material and drill it. This is the only category most people even address — and it's usually the smallest one.
• Carelessness. I knew how, and botched it: mistyped a number, mixed up a sign, rounded wrong, wrote the answer on the wrong line. The fix isn't “study more” — it's a checking ritual and slowing down at the critical points.
• Time. I knew how, but didn't get there. The fix is test strategy — question order, minute budgeting, the ability to walk away from a question.
• Misreading the question. I could calculate, but I calculated the wrong thing. A missed “not,” “at least” instead of “at most,” an answer in the wrong units, two answers picked where only one was called for. The fix is a technique for reading the question carefully.

The distinction matters enormously: someone losing fifteen points to carelessness and time doesn't need another workbook — they need a different relationship with the test. Deciding “I'll just study more” without sorting out the causes means months over a workbook with no movement on the actual test.

A prompt that sorts your mistakes for you

I just took a practice entrance exam ([subject]). Here's my
breakdown, question by question:

[for every wrong answer, write: question number, what it was about,
my answer, the correct answer per the key, how confident I was
(sure/guessed/no idea), and whatever I remember going through
my head]

Sort all my mistakes into four categories:
A) KNOWLEDGE — I don't know this material
B) CARELESSNESS — I knew it and made a slip
C) TIME — I didn't get there, or rushed it
D) MISREADING — I calculated something other than what was asked

Output as a table: question | topic | category | why you think so.
Below the table, write:
1. How many points I lost in each category
2. Which category is my biggest hole
3. Three specific things to do over the next two weeks — ranked
   by how many points each one is likely to recover

Where my descriptions aren't enough to classify something, ask
a follow-up question instead of guessing.

This returns a table and a breakdown of lost points by cause, which tends to be a surprise — for most people, “knowledge gaps” turn out to be a smaller share than they expected. The last line of the prompt matters: without it, the model guesses the category, and a guess based on your two-word note isn't reliable. If it asks, answer it.

A mistake log, and coming back to it

Analysis is only worth doing if you return to it. Keep one notebook or file — a mistake log — where every mistake gets a row: date, question, topic, category, what you'll do about it. After two weeks, try the same type of question again. The best measure of progress isn't “how many points did I get” but “does a mistake I already logged show up again on a new test.” If it does, your earlier analysis was just talk.

Here's my mistake log from the last [6] weeks — every mistake
from practice tests and drills:

[paste the log: date, question, topic, mistake category,
what I did about it]

Analyze it:
1. Which types of mistakes keep recurring (same topic or same
   kind of slip across more than one test)
2. Which ones I've actually fixed — they showed up early on
   and haven't come back
3. Where I keep making the same mistake even though I already
   logged it — and what to do differently this time
4. A recommendation for the next two weeks: what to practice
   and in what order

Be specific and don't go easy on me. If the log shows I haven't
moved at all on some topic, say so.

This returns an overview of what you're actually fixing versus just checking off. Point 3 is the valuable one — a mistake that recurs after being logged usually means the fix you wrote down was too vague (“be more careful”).

Phase 4: drilling your weak spots with AI

Once you know where you're losing points, it's time to drill. This is where AI is at its most useful — a patient tutor who explains something a third time just as willingly as the first, and never sighs.

A step-by-step explanation, not just the answer

The worst thing you can do is have AI just tell you the answer. You already know the answer from the key. What you need is a method — one you can pull off yourself next time.

I couldn't solve this entrance-exam question. I'm [a 9th grader /
a senior], and I generally struggle with [topic].

The question:
[copy the question exactly; for geometry, attach the image too]

My attempt and where I got stuck:
[what I tried, where it stopped making sense]

Explain it to me this way:
1. First tell me the key word or piece of information in the
   question I should have noticed, and how to recognize this
   type of question
2. Then a step-by-step method — for EVERY step, explain why it's
   done (not just what's being calculated)
3. Flag the step where people most often mess this up
4. Finally: how I can check my answer a different way

Don't use terms we haven't covered in school, and if you do,
explain them. Keep it simple.

This returns an explanation you can reuse next time, since it includes a way to recognize the question type, not just the steps. Check your result against the official answer key — models do slip up on intermediate calculations, especially in geometry and multi-step problems. If the result disagrees with the key, the key is right; ask to be shown exactly where the method diverged. More on this technique in explaining material with AI.

Let it walk you to the answer yourself

Even better than an explanation is a dialogue where you do the work: add a rule to your prompt like “ask me one question at a time, don't reveal the method or the answer, and if I answer wrong, guide me with a question instead of correcting me.” It's slower than reading a finished method, and it sticks in your head far better. After a few exchanges, the model tends to slip back into explaining — remind it of the rules.

Generating similar problems — and the mandatory check

Once you understand one question, you need five more of the same type. This is where AI saves a lot of time, but a hard rule applies: AI-generated problems have to be verified. Models routinely make up numbers for math questions that don't lead to a clean answer, drop a condition that makes the problem solvable in the first place, or write a correct question paired with a wrong solution.

Here's a question from an official entrance exam that I couldn't
solve:

[question]

Create 5 problems of the same type and similar difficulty for
me to practice. Do this:
- vary the numbers and context, but drill the same skill
- make the first two a bit easier, the last two a bit harder
- for each problem, give: the question, the correct answer,
  and the complete method
- for each, note what slip is easiest to make on it

Then check each problem yourself: solve it again from scratch,
independently of your first method, and tell me whether you get
the same result. Where the results disagree, throw the problem
out and tell me.

This returns a practice set plus a self-check. That check isn't a formality — having the model redo the problem a second time catches a good share of errors. Even so, the rule stands: for math, always ask to see the method and go through it. If the method doesn't add up on some problem, or the result doesn't come out clean, throw the problem out instead of spending half an hour on it. A bad question reveals itself by simply “not working out.”

Your first language: spelling, forms, and reading comprehension

Your first-language exam works differently than math. Spelling patterns need lots of sentences to drill on, and reading comprehension needs to be trained on real texts.

I keep losing points on these issues in [subject] tests:
[e.g. subject-verb agreement, commonly confused homophones,
capitalization rules, identifying parts of a sentence]

Prepare an exercise set for me:
1. For each issue, first a short explanation of the rule in plain
   language, with a trick for remembering it and where the most
   common trap is
2. Then 10 fill-in-the-blank sentences for the first issue, 10 for
   the second, etc. — make the sentences ordinary, not artificial,
   and include some where the rule doesn't apply straightforwardly
3. Put the correct answers all the way at the end, not next to
   individual sentences, so I can do it honestly
4. For each answer, add one sentence on why

Don't use examples that look like they're from an elementary
school textbook.

This returns a set you can print out. Spot-check the rules against a reliable style/usage reference — models occasionally get edge cases wrong and confidently explain a rule that doesn't actually exist. For things you just need to know by heart (spelling patterns, prefixes, irregular forms), flashcards work well — see the Anki flashcards with AI guide.

Aptitude tests: training how you read a question

For general aptitude-style tests, you're not studying material — you're training how you read a question and how fast you recognize its type. The verbal section rests on reading relationships between concepts precisely, the analytical section on converting text into a diagram or table, the quantitative section on estimation and working with numbers without lengthy calculation.

This is a question from an aptitude test where I got stuck or
it took me way too long:

[copy the question and all the answer choices]

My answer was [X], the correct one is [Y] (per the official key).

Break it down for me this way:
1. What type of question this is and how to spot it at a glance
2. The fastest method — and how much time it should take
3. Exactly where I probably went wrong, given that I answered [X]
4. How to quickly eliminate answer choices without fully solving
   the question
5. One rule I should take away from this question for next time

Don't talk vaguely about "logical thinking" — I want a concrete
method.

This returns an analysis focused on technique, not knowledge. Point 4 is often more valuable than the correct method on these tests — eliminating three nonsensical choices is usually faster than honestly solving the question.

Phase 5: time strategy and the big day

Knowledge is only half the result. The other half is how you handle the test during those ninety minutes.

A minute budget and three passes

Work out how many minutes you get per question on average, and treat that as a limit. Then run the test in three passes. First: solve everything you can right away — a question where you don't see a way in within thirty seconds gets flagged and skipped. Second: go back to the flagged ones. Third: check your work and fill in whatever's left. The first pass is the hardest one to keep disciplined and pays off the most: the most common way to lose points is getting stuck on question eight for ten minutes and never reaching question twenty, which would have taken a minute.

I'm taking [exam name]. The [subject] test runs [X] minutes and has
[Y] questions worth [Z] points total. Wrong answers [do / don't]
lose points.

My practice test results by topic:
[list where I'm strong and where I'm weak, and how much time
I typically run short by]

Suggest a time strategy:
1. A minute budget: how much time per section, including a
   buffer for checking
2. What order to go through the questions based on what I'm
   good at
3. When it makes sense to walk away from a question — a specific
   signal, not "when it's not working"
4. What to do with whatever minutes are left at the end
5. What to do with questions I don't get to — given whether wrong
   answers lose points

For each point, write one sentence on why. At the end, give me
a 5-bullet strategy card I can memorize.

This returns a strategy built around your profile. Always check point 5 against the official instructions — the model sometimes guesses at the point-deduction rule, and “guess on everything” is bad advice when a wrong guess costs you points.

Practicing under pressure

You need to test your strategy before the real exam. Take your last two practice tests under exactly the conditions you'll face: same time of day, same length, no breaks. And if possible, at least one in an unfamiliar setting — handling stress at a desk that isn't yours is its own skill.

Practice the first thirty seconds too: catching your breath, reading the instructions, flipping through the whole test. Anyone who dives straight into question one usually has no idea the last three questions are easy.

The last week, the day before, and the morning of

No new material gets studied in the last week. You review what you already know, take short sets of questions to stay sharp, and above all, sleep. A sleep-deprived brain can't retrieve even what it's already learned — sleep in the week before the exam is more effective prep than another two hours over a workbook.

The day before the exam: go over your test strategy and your mistake log, pack your things (admission ticket or ID, pens and drawing tools, a snack, water), check the route and add a buffer. After that, nothing — studying new material the night before leaves you with nothing but the feeling that you don't know anything. Eat a normal breakfast in the morning, arrive early, and don't chat with other kids outside the building about who knows what. Two free things help with nerves: slow breathing with a longer exhale, and remembering that nervousness is just your body getting ready to perform.

Phase 6: the parent's role

This section is for parents, and it's short on purpose, because it fits in one sentence: your job is to hold the conditions and the calm steady, not to police the performance.

What helps and what hurts

Practical support helps: a quiet place to study, food, a sleep schedule, transportation on the day, applications handled on time, printed official past tests, and a willingness to explain something when your kid asks. And above all: be curious about the process, not the numbers. “What did you get?” puts a kid on the defensive; “what gave you the most trouble on that test?” starts a conversation you can actually learn something from.

What hurts: drilling beyond the plan, comparisons to classmates, sentences like “if you don't get in, then…,” and your own anxiety rubbing off on your kid. Be careful with tutoring added at the last minute too — once a kid already has a plan and a rhythm, three new people with three different methods in the final month tend to do more harm than good.

I'm a parent. My kid is in [9th grade / their final year of high
school] and preparing for [exam]. [X] weeks remain.

Our situation at home: [describe it — e.g. my kid studies in
bursts, is anxious, refuses help, or on the contrary wants me
to quiz them every evening]

Advise me on:
1. What I can concretely do to help their prep go better —
   practical things, not general advice about "being supportive"
2. What I should avoid, even with good intentions
3. How to ask about practice test results without it turning into
   a fight — suggest three specific sentences
4. How to tell the difference between ordinary nervousness and
   a situation where we need outside help (a school counselor,
   a doctor)

Keep it short and practical.

This returns concrete behavioral guidance instead of generic phrases. Take point 4 seriously: long-term sleep or eating problems, or significant anxiety, aren't things a better study plan fixes — they belong with a professional. AI's job here is organizing the prep, not handling health.

And a note on privacy: when you upload scans of notebooks, report cards, or evaluations, remove names, ID numbers, and the school's name. Sensitive personal data belongs, at most, in a paid account with contractual data protection, and even then in anonymized form.

Common mistakes

• Practicing what you already know. It feels good and accomplishes nothing. What matters is time spent on what's actually tripping you up — and you only find that out from mistake analysis, not from a feeling.
• Training on questions generated in a chat instead of official questions. The difficulty and style of questions are different, and you'll learn something other than what's actually on the test. Generated problems are a supplement for drilling one type of question, not a source of tests.
• Having AI show you the answer instead of the method. You already know the answer from the key. Without a method and a way to recognize the question type, the same kind of problem will trip you up again next time.
• Not checking AI-generated problems. Models invent questions with no clean solution and pair correct questions with wrong answers. Half an hour spent on a badly written problem costs you both time and confidence.
• Taking practice tests comfortably. In bed, with your phone nearby, no timer, and the option to peek at your notes — a test like that measures something you won't have on exam day.
• Skimping on sleep in the final week. A sleep-deprived brain can't retrieve even what's already in it. The last night before the exam spent over a workbook is a pure loss of points.

The best tools

• The exam provider's official archive (Cermat, a faculty's website, the aptitude-test provider) — the only source of questions worth training on for real.
• Claude or another AI in a chat — step-by-step explanations, Socratic guidance to the answer, sorting mistakes by cause, generating practice sets. Reading photos of notebooks and handwritten solutions works, so you only need to snap a photo of the question.
• A project or persistent context — a place where you write once what grade you're in, what exam you're preparing for, and what's giving you trouble, so you don't have to repeat it with every question.
• Anki or another flashcard app — for what simply has to be memorized: formulas, spelling patterns, definitions. See the flashcards with AI guide.
• Paper, a pencil, and a timer — irreplaceable. The real test is taken by hand on paper, so that's exactly what you should practice with.

What you get out of it

• Time: instead of randomly flipping through a workbook, practice aimed at your actual weak spots — you get substantially more done in the same four hours a week.
• Points: the biggest gains usually come from carelessness, time, and misread questions. Those are points you get without learning any new material, just by handling the test differently.
• Calm: someone who's taken eight practice tests under real time conditions walks into the real one like it's the ninth. The nerves don't disappear, but nothing catches you off guard.
• Quality of prep: a mistake log and a plan built backward from the date mean you always know what to do and why. That's the difference between a system and panic.

Pro tip

The most valuable half hour of the entire prep process is the one right after a practice test, while you still remember what was going through your head on each question. Don't put off the analysis until evening — notes like “I saw what to do here, but didn't want to finish the calculation” evaporate within an hour, and without them you can only guess at the mistake category.

And a rule to follow through the whole process: the official questions and the official answer key are always right; AI is only there to explain. When a model's result disagrees with the key, the model is wrong — even when its method looks completely convincing. If you're preparing for final exams at the same time, this guide connects to the year-long plan for seniors.