A few months back I asked ChatGPT to “tell me everything I should know about closed guard” and got back about 1,400 words of confident, well-organized, slightly-wrong text. The errors weren’t dramatic. They were the kind only someone who actually plays closed guard would catch. That’s roughly the shape of using ChatGPT for jiu-jitsu in 2026: useful in narrow ways, dangerous in confident ways, and most of the time you can’t tell which mode you’re in until you’ve already spent a week drilling bad advice.
This is the more specific cousin of my general AI-tools post. That one covers the broader landscape. This one is just ChatGPT — the actual prompts, where it earns its keep, and where it has led me wrong.
What I use it for now
Two things, both unglamorous.
The first: I paste in transcripts of post-roll voice memos and ask for a clean restructure. The output isn’t new information — it’s my own content with the conditions called out and the cause-and-effect made explicit. Sometimes ChatGPT surfaces an ambiguity in my own thinking that I wouldn’t have spotted rereading my own text, and that’s the real value. It works as a mirror, not as a coach.
The second is vocabulary translation. My coach calls a position one thing, the YouTube instructional calls it another, and ChatGPT is reliably good at answering “is this the same position?” That’s genuinely faster than hunting for the alternate name yourself.
Everything beyond those two has bitten me at least once, which is why the list stays short.
Where it’s bitten me
The most expensive failure mode is what I’d call confidently averaging contradictory sources. Two instructionals disagree on a detail, and ChatGPT smooths the disagreement into a middle answer that’s wrong in both contexts. I’ve made this mistake at least twice, and both times I only caught it when a drilling partner pointed out that what I was doing made no sense.
It also can’t tell me what to work on. Whenever I’ve asked, the suggestions are plausible-sounding and almost never right, because the model doesn’t know what I’m bad at. Diagnosing that is my job, and outsourcing it never went well.
The pattern underneath both failures took me a while to articulate: when ChatGPT writes a paragraph that sounds like it understands jiu-jitsu, that’s because the paragraph sounds like jiu-jitsu writing, not because there’s understanding underneath. The fluency is convincing in a way no other failure mode quite matches.
The prompt I actually use
This is the one I reach for most often:
I just rolled and got passed three times the same way. Here’s what I remember: [paste voice-memo transcript]. Help me structure this into: the position I was in, the move I tried, the counter that worked on me, and a likely cause. Don’t suggest new techniques — I want to understand what happened first.
The last sentence does most of the work. Without it, the model jumps straight to solutions — try this sweep, here’s a counter, here’s a new entry — and I never get to diagnosis. With it, the output stays in the what happened layer, which is where I actually need help.
The same constraint-first pattern works for technique notes:
Here’s a rambling description of a position I worked tonight. Don’t add techniques. Just restructure what I wrote into position → move → response → outcome.
That word “don’t” carries both prompts. The model defaults to being helpful in the wrong direction; it wants to give you more, and what you usually need from it is less.
<a class=“inline-cta” href=“https://grappleflows.com/voice?utm_source=blog&utm_medium=referral&utm_campaign=post-chatgpt-to-study-jiu-jitsu&utm_content=inline-cta&ref=blog”
Try it free A purpose-built tool for turning voice memos into BJJ flowcharts. Create a free account →
What it can’t replace
The list is short and hasn’t changed in three years: a coach, a training partner, logged mat hours, and a camera. ChatGPT can format the artifacts around those things — clean up notes, restructure rambling, translate vocabulary between lineages. It can’t substitute for any of them, and the moment you treat it like it can, progress quietly stalls.
If you’ve been studying jiu-jitsu with ChatGPT for a month and your game hasn’t moved, that’s the signal to look at. You didn’t pick the wrong prompts. You skipped the part where the thinking has to happen between your ears, on the mat, in real time.
