Studying for the Philippine Bar with AI, without learning law that does not exist

Three weeks before the Bar, a candidate reviews an AI-generated summary of just causes for termination. It lists five grounds, quotes Article 297, and reads exactly like the reviewer she paid for. One of the five is phrased slightly wrong, in a way that flips an exception into a rule.

She will not notice tonight. She will memorize it, because memorizing is the whole point of the season, and AI-polished text is frictionless to memorize. The error surfaces in September, inside an essay answer, in front of the one reader who knows the provision by heart. Study tools that cannot be checked do not just fail to help; they install mistakes with excellent production values.

Used well, AI is the study partner previous generations did not have. It explains a provision three different ways until one clicks. It generates endless practice questions from a topic you name. It plays examiner: give it your essay answer and ask what a strict grader would attack. It turns dead commute time into recall drills.

All of these share one property: they work FROM material you give it or check against sources you hold. The failure modes live on the other side of that line, where the AI supplies the law itself: summaries of cases you never opened, doctrines recalled from training data, mnemonic lists of elements nobody verified. The 2025 results underline how thin the margin is: 48.98% passed, the highest rate in years, and still barely half.

One rule keeps AI on the right side of your preparation: before anything goes into your memory deck, trace it to the codal provision or the reported case. Not to the AI's assurance, to the text.

This is the same claim-versus-passage habit from the 90-second answer check, with the stakes turned up, because Bar review is deliberate memorization. An unverified error in ordinary work misleads you once; an unverified error in your review installs itself permanently and waits for exam day. The trace takes a minute with the codal open, and it doubles as active recall, which is better study than rereading anyway.

The practical setup: keep the primary sources one click away from every AI answer. The Supreme Court E-Library and the Official Gazette are free and authoritative. Your review materials, syllabi, and notes can live alongside them.

This is exactly the shape Tatsulok is built in: curated collections of the Constitution, the codes, statutes, and Supreme Court decisions, where every answer to your question links to the exact provision or holding it came from. Ask it to explain unjust vexation or drill you on the elements of estafa, and the citation to check is already attached. When the collection does not contain an answer, it says so, which during Bar season is precisely the tool telling you "go read the codal, this one is not settled the way you think."

Here is a drill you can run today, with any tool. Pick one syllabus topic. Ask the AI for the elements or requisites, with the source for each. Then, before reading its answer closely, write your own list from memory. Compare the three: your memory, the AI's claim, and the actual provision.

Where your memory and the codal agree, you have confirmed knowledge. Where the AI and the codal disagree, you have caught the machine, which is the best possible reminder of why you check. And where your memory fails, you have found tomorrow's review topic. Three signals from one exercise, and every one of them traced to real law. That is what using AI right looks like when the stakes are a professional license.