AI Study Plans: What the Table Does Not Tell You
Ask a chatbot to build you a revision schedule and you get a clean table of dates. It looks right. That is the problem. A good study plan and a bad one are the same object: a tidy grid of dates and topics. Reading it tells you nothing about which one you have.
This site spends a lot of time on AI answers that are wrong in ways you can check, a fabricated citation, a price that has changed, a law that says the opposite. Plans are a different failure and a harder one, because there is nothing in the output to verify. The dates are real dates. The topics are your topics. The arithmetic adds up. What you cannot see is whether the shape is any good.
What you would need to know to judge it
There is a body of research on this, and one study does most of the work. Nicholas Cepeda, Edward Vul, Doug Rohrer, John Wixted and Harold Pashler taught over 1,350 people a set of facts, brought them back for a single review after a gap of up to three and a half months, and tested them up to a year later. Psychological Science, 2008.
Their finding is more specific than “space it out”: “At any given retention interval, an increase in the inter-study gap at first increased, and then gradually reduced, test performance. The optimum gap value was about 20% of the test delay for delays of a few weeks, falling to about 5% when delay was one year.”
Three things fall out of that, and a plan can quietly get any of them wrong.
The right gap depends on how long you need to hold the material, not on how much runway you happen to have. Their table is blunt about it: for retention intervals of 7, 35, 70 and 350 days, the best gaps were 1, 11, 21 and 21 days. It grows in days, shrinks as a share, and flattens out around three weeks. Stretching four sessions across five months because five months is what you have is not spacing, it is forgetting with a calendar attached.
The curve is lopsided. It climbs to a peak and then, in their words, is “gradually reduced”. Too wide costs you a little. Too narrow, which is what cramming is, costs you a lot. So if a plan errs, you want to know which direction.
And the whole finding is about repeating the same material by retrieving it. A schedule that puts chapter one on Monday and chapter two on Thursday is a timetable. Rereading your notes on four spread-out evenings is still rereading your notes.
The test worth running
None of that is visible in a table of dates, which makes this a good exercise in checking a kind of AI output that usually goes unchecked.
Ask your chatbot for a revision plan, then put the numbers into the tool below. It will tell you where the gap it proposed sits against the research, and whether it errs in the cheap direction or the expensive one.
Then look for the two things a plan will almost never volunteer. Does it say anywhere that each pass has to cover the same material? Does it say anything about retrieving rather than rereading? If both are missing, you have a schedule, not a study method, and the schedule is the easy half.
Check a study plan against the research
Nothing you type here leaves your browser. There is no account, no upload and no server: the plan is built on your own device and disappears when you close the page.
The omission you should expect
There is one gap that is close to guaranteed, and it is worth knowing about in advance because it is a property of the question rather than of the model.
You asked for a plan for an exam. So you get a plan that ends at the exam. Every session is arranged to make one morning go well, and the day after, nothing is scheduled because nothing was asked for.
For most subjects that is the wrong objective. The same research is clearest exactly here: the best gap for material you still want in a year was 21 days. If you need this next term, the reviews that matter most are the ones after the exam, and they are short. The tool will schedule them if you tick the box.
This is not the model being careless. It answered what you asked. But the framing of a request quietly decides what gets optimized, and the thing that was not optimized does not appear anywhere in the answer as a gap. It just is not there.
Ask what it optimized for. A plan is a set of tradeoffs presented as a list. The tradeoffs are invisible in the output, so ask directly: what was this built to maximize, and what did it give up to get there?
Ask what it left out and why. Phrased that way, not as “is this good”, because the second question invites a defense and the first invites a list.
Find the one number the plan turns on. Most plans have one: a gap, a ratio, a threshold. That single number is usually checkable against something published, and checking it is faster than evaluating the whole plan.
Treat neatness as a formatting choice. A well-structured table is evidence that the model can format a table. It is the easiest part of the job and carries no information about the rest.
What the research does not settle
The study taught facts and tested facts. Whether the same curve governs learning a proof, an essay technique or a language is a reasonable guess, not a demonstrated one, and the authors do not claim it.
It also says nothing about how long a session should be. The minutes field in the tool exists so the output looks like something you can follow, not because a number belongs there.
And to be straight about what the tool does: it spaces your sessions evenly across the time you give it, which is the ordinary approach. What comes from the research is the judgment it puts next to that, and the reviews it adds after the exam. The scheduling itself is not clever, and a tool that claimed otherwise would be doing the thing this article is about.
In a nutshell
A study plan from a chatbot is a good test case for AI advice, because a right one and a wrong one look identical: both are a tidy table of dates. What you would need in order to judge it is the shape of the spacing curve, and that is in the literature rather than in the output. The gap should scale with how long you need to remember the material, roughly 20 percent of the delay for a test a few weeks away and about 5 percent for one a year away, flattening around three weeks. Erring wide is cheap and cramming is expensive. Every pass has to cover the same material and involve retrieval. And because you asked about an exam, the plan will end at the exam, which for most subjects is the wrong thing to optimize.
Nicholas J. Cepeda, Edward Vul, Doug Rohrer, John T. Wixted and Harold Pashler, Spacing Effects in Learning: A Temporal Ridgeline of Optimal Retention, Psychological Science, 2008, read at the source for the sample size, the quoted abstract, the shape of the curve and the table of optimal gaps at retention intervals of 7, 35, 70 and 350 days. The tool checks your gap against those figures; it does not use them to place your sessions, and the section above says so.
For the same problem applied to facts instead of plans, how to catch AI when it is confidently wrong goes through the checks, and AI for students covers where the line is on using it for coursework at all.
