ChatGPT or Claude costs $0 to $20 a month, and used well it covers two of the things you pay the most for in a $2,000 prep course: explanations whenever you need them, and drilling on the concepts you're weak on. Used badly, it teaches you wrong numbers and makes you more confident than you should be. Below are six ways I'd use it, with prompts you can copy, and four things I wouldn't use it for.

I run FreeFellow, and a chatbot works fine alongside any question bank, FreeFellow's included.

The general rule is to use the chatbot to explain things, quiz you and critique your work, and not to treat it as the source of truth. That means not using it to write your question bank, check your math or tell you whether you're ready. I covered why AI-written practice questions are a problem in the seven tells of an unaudited AI question bank.

Six ways to use it

1. Explain it back

This is the one I'd use most. After you study a topic, close the book, write your own explanation from memory, and paste it in:

I am studying for [exam]. Here is my from-memory explanation of [concept]: [your explanation]. Find every error, gap, or imprecision, ranked by how likely it is to cost me points. Do not restate the concept; critique my version.

You're pulling the material out of memory and then getting corrected, and that works better than rereading or rewatching. A chatbot is also better at this than a study partner, since it won't let a small mistake slide to be polite, as long as you ask for ranked errors and not encouragement.

2. Go through every question you miss

When you miss a practice question, don't stop at the right answer. Figure out why you picked yours:

I answered this practice question wrong. Question: [stem and choices]. I chose [X] because [your actual reasoning]. The correct answer is [Y]. Name the specific misconception in my reasoning, state the general rule I violated, and give me one new scenario where the same mistake would trap me again.

The new scenario at the end is what tells you whether the fix stuck. FreeFellow's notes on each answer choice work on the same idea, since each wrong choice is labeled with the mistake that leads to it, so on FreeFellow questions you can paste the note in and go a level deeper.

3. Have it quiz you on a topic

For a topic you've already learned and want to keep fresh:

Drill me on [learning objective] for [exam]. Ask one question at a time, wait for my answer, then tell me if I am right and why before the next question. Escalate difficulty as I get answers right. Mostly conceptual; make me do no arithmetic longer than one step.

The limits in the prompt matter. One question at a time makes you recall the answer instead of recognizing it. Keeping the math to one step keeps the model away from what it's worst at, so it's less likely to hand you a wrong "correct" answer.

4. Break down a formula

If you only memorized a formula, a reworded question can throw you. This prompt helps you understand it instead:

Take [formula]. Walk through what each term is doing economically, what happens to the result when each input rises, and what breaks if each term is removed. Then give me the two most common ways exam questions disguise this formula.

Check any numbers against your formula sheet or textbook. The models are good at explaining how a formula works, less so at the arithmetic.

5. Translate a dense passage

Accounting standards, regulations and actuarial study notes are written for people who already work in the field. Paste the passage and ask for a plain-English version and a small worked example. Then have the model quiz you on the terms in the original passage, so you end up knowing the real wording and not just the simplified one.

6. Get feedback on written answers

For essay and written-answer exams (CFA Level III, the actuarial written-answer exams, CPA simulations), a chatbot that has the actual grading rubric is the closest most candidates will get to a personal grader. You need the rubric. Without it, the model grades on how good your answer sounds and tends to be generous. I wrote more about this in Can AI Grade Your Essay Exam Answers?, including how FreeFellow gives you a prompt with the rubric built in for free.

Four things I wouldn't use it for

  • Your question bank. AI-written questions that nobody reviewed have statistical tells and answer keys nobody checked. Practice with a fixed, reviewed bank so your score means something.
  • Checking arithmetic. Models get numbers wrong and sound sure about it. Check against the worked solution or your calculator, not by asking the same model again.
  • Current-year figures. Tax thresholds, contribution limits and updated standards get mixed up across years. Check the exam body or the current law.
  • Telling you if you're ready. That takes data on how you do on calibrated questions compared with other prepared candidates, and a chatbot has neither your history nor the calibration. Use a readiness score built on real answer data or a full timed mock, and use the chatbot to explain the gaps it shows you.

Putting it together

The setup I'd use in 2026 costs almost nothing: a fixed, reviewed question bank to measure yourself against (FreeFellow has open samples, and the full bank is free with an account), the exam body's own materials when you need the right answer, and a chatbot for the six uses above. Between them you get explanations, drilling, feedback and a way to measure progress, which covers most of what a $2,000 course is selling.