You've seen the posts. "🏀 5 KEYS TO BUILDING A WINNING CULTURE 🧵" followed by five paragraphs that could describe any sport, any level, any decade. That's what happens when someone asks a chatbot for basketball content and posts the first thing it says.
The problem isn't the tool. It's that a generic question gets a generic answer, and most coaches are asking generic questions. Below is what changes when you stop doing that.
The one rule that fixes 80% of it
Never ask for a thing. Ask it to react to your thing.
"Write me a practice plan" produces slop, because the model has no idea that you have 75 minutes, two baskets, eleven kids, three of whom can't go left, and a game Thursday against a team that presses. Nothing useful can come from that prompt.
"Here is my practice plan, here are my constraints, tell me what's wrong with it" produces something worth reading. You have supplied the specifics that only you know. The model is doing the one job it's good at: pattern-matching against a huge amount of material and pushing back.
Try this instead
Here is my plan:
[paste your plan with time blocks]
Tell me: what's the weakest 10 minutes of this plan, what am I repeating that I don't need to, and what am I not touching that I should be given the weakness I named. Don't rewrite it — critique it.
That last sentence matters. "Don't rewrite it" stops the model from handing you a sanitized version of your own plan and calling it an improvement.
Four jobs it's genuinely good at
1. Turning your scattered notes into something you can hand out
You watched film and typed nineteen fragments into your phone. Feed them in and ask for a one-page scouting sheet organized by personnel, with a "three things we must do" section at the top. You're not asking it to scout — you already did that. You're asking it to format.
The "don't add anything I didn't say" clause is the whole trick. Without it you'll get plausible-sounding tendencies that no one observed.
2. Arguing with you before your assistant has to
Coaches get isolated. Ask it to take the other side of a decision you've already made.
That last clause turns an argument into a checklist you can actually verify on Thursday.
3. The parent email you don't want to write
Playing time conversations, a kid getting cut, an incident at practice. These are hard because the emotional register is hard, not because the words are. Give it the facts and your intended tone, then rewrite the output in your own voice — never send it raw.
"No sports clichés" is doing heavy lifting. Without it you get "at the end of the day" and "buy in" and every parent knows exactly what wrote it.
4. Explaining a concept six different ways
Your best teaching tool is having eight explanations of the same idea for eight different kids. Ask for the same concept explained for a player who learns by seeing, by doing, by analogy, by numbers.
Three jobs to never hand it
Anything with a number attached. It will confidently produce stats, records, and player measurements that are wrong. If you wouldn't state it without checking, don't let a model state it for you.
Evaluating a specific player it can't see. No amount of description substitutes for watching a kid play. It will happily generate an assessment anyway, which is worse than saying nothing.
Your voice with your team. Players know. The moment a message to the group chat sounds like a press release, you've spent credibility you'll want later.
Where the tools fit. AI is good at reacting to what you already have. It's bad at knowing what happened in your gym. That's the job of actual tools — diagram the set in the play designer, log where shots actually fell with the Shot Chart in the app, then bring that real data to the conversation. A model reasoning about your actual numbers is useful. A model guessing is noise.
The tell
If what comes out could have been written about any team in the country, you asked the wrong question. Go back and add the constraint you left out — the gym time, the roster, the kid who won't pass, the game on Thursday. The specifics are the whole point.