Nearly every martial arts school owner who has tried AI for writing has had the same experience: the first draft arrives instantly, reads fluently, and is unusable. It is generic, it is enthusiastic in a way you are not, and it says things about your school that are not true. The problem is almost never the tool. It is that the owner asked for a finished article when what these systems are good at is a different job entirely — and that the input contained nothing only your school could have supplied.
What these tools are actually good at
| Good at | Bad at |
|---|---|
| Turning something you said into something readable | Inventing something you did not say |
| Structure: outlines, headings, ordering an argument | Knowing which argument matters to your market |
| Producing twelve variations so you can pick one | Knowing which of the twelve is right |
| Rewriting the same content for different lengths and formats | Facts, figures, dates and names — which it will produce confidently and sometimes wrongly |
| Making a rough draft consistent in tense, tone and terminology | Your voice, unless you have given it a great deal of your actual writing |
| Asking you questions that surface what you know | Replacing the knowing |
The single most valuable use, and the one owners least expect, is the last row. A school owner who spends ten minutes being interviewed by an AI about why their beginner programme works produces material no generic prompt would ever have surfaced — and that material is the article.
The rule that fixes most bad output
Dictation, not generation
Do not ask for an article about anti-bullying programmes. Talk for four minutes about what you actually do when a parent tells you their child is being bullied — the first question you ask, the thing you tell the child, the thing you tell the parent, the mistake you used to make. Then ask for that to be turned into an article.
The first approach produces something that could have come from any school in the country, which is exactly why no AI system will cite it and no parent will be moved by it. The second produces something only you could have said, written more cleanly than you would have written it.
Your knowledge is the scarce input. The writing was never the scarce part.
What to give it before asking for anything
| Input | Why it changes the output |
|---|---|
| Who the reader is, specifically | “A parent of a seven-year-old who has been meaning to do this for six months” produces different writing from “prospective students” |
| What you actually do | Your programmes, ages, schedule, what is distinctive. Without this, it fills the gap with plausible generalities. |
| Samples of your own writing | Three or four things you have written yourself do more for voice than any instruction about tone |
| What you will not say | Claims you cannot support, promises you do not make, words you dislike. This is the most under-used instruction available. |
| The format and length | A 600-word FAQ answer and a 2,000-word guide are different jobs |
| The one thing the reader should do next | Every piece of content has a next step, and it is usually not “enroll now” |
Writing that AI systems will actually quote
This is where writing for a school website has changed materially, and where the work overlaps with what this journal has covered on local search. Increasingly, parents do not read your page — a system reads it and repeats a sentence from it. That reframes what good writing means.
| Practice | Effect |
|---|---|
| Answer in the first sentence, then explain | A page that spends three paragraphs warming up has nothing quotable in it |
| One question per heading | Headings are how both readers and machines locate the part that answers them |
| State facts as facts | “We teach children from age four” is quotable. “We believe every child can succeed” is not. |
| Be specific rather than general | “We teach discipline” is unfalsifiable and therefore worthless. Describe the mechanism. |
| Use the words parents use | They search “karate for kids near me”, not “youth traditional martial arts curriculum” |
| Connect to real local things | Named neighbourhoods, named schools, named landmarks — where genuinely true |
| Date anything that changes | Undated schedules and prices are treated with suspicion by readers and systems alike |
The line this journal will not cross, and neither should you
There is a genre of advice circulating in this industry that recommends writing school marketing in a third-party, objective-sounding voice — content that reads as though an independent expert were recommending you — and supporting it with statistics, expert endorsements and success stories generated to be persuasive.
The underlying structural insight is sound: content that answers a question directly, cites evidence and reads as informative rather than promotional does perform better with both readers and AI systems. That part is worth adopting.
The execution is where schools get into real trouble. Three specific lines:
Do not invent statistics. “89% of parents report improved confidence” is a fact about your school or it is nothing. If you have surveyed your families, say so and say how many. If you have not, delete the number. AI systems increasingly weight citable claims — which means a fabricated one is a claim you have published under your own name, in a form designed to be repeated.
Do not attribute endorsements that were not given. A named professional recommending your school must have recommended your school. In the United States, the Federal Trade Commission’s rules on endorsements and testimonials carry real penalties, and they apply to a school’s website exactly as they apply to anyone else’s.
Do not present marketing as independent journalism. Writing informatively is good practice. Writing in a way designed to make a reader believe an unaffiliated third party produced it is not, and it is the kind of thing that ends badly and publicly.
Everything legitimate in the approach survives these three rules. Write the answer-first, evidence-led, specific, locally grounded page — and make every claim in it true.
A practical workflow
| Step | Detail |
|---|---|
| 1. Talk, do not type | Dictate four minutes of what you actually know about the subject |
| 2. Ask it to interview you | “What would a parent want to know that I have not covered?” Answer the questions it asks. |
| 3. Ask for the structure first | Fix the outline before any prose exists. Fixing structure after drafting wastes everything. |
| 4. Draft one section at a time | Whole-article drafts are harder to correct than they are to re-commission |
| 5. Strike every claim you cannot support | Do this as a separate pass, reading only for claims. It catches what a general read does not. |
| 6. Read it aloud | The fastest test of whether it sounds like you. If you would not say it, cut it. |
| 7. Add the thing only you know | A specific student, a specific class, a specific mistake you made. This is what the tool cannot supply and what makes the piece worth publishing. |
Frequently asked questions
Why does AI-written content for martial arts schools sound generic?
Because the input was generic. A request for “an article about anti-bullying programmes” contains nothing only your school could have supplied, so the output could have come from any school in the country. Dictating four minutes of what you actually do when a parent raises bullying — the first question you ask, the mistake you used to make — and asking for that to be written up produces something only you could have said.
What should you give an AI tool before asking it to write?
Who the reader is specifically, what your school actually offers, three or four samples of your own writing, an explicit list of things you will not say or claim, the format and length, and the one action the reader should take next. The samples do more for voice than any instruction about tone, and the do-not-say list is the most under-used input available.
How do you write school content that AI assistants will quote?
Answer the question in the first sentence and explain afterwards; give each question its own heading; state facts as facts rather than beliefs; describe mechanisms rather than asserting qualities; use the words parents actually search with; connect to genuinely local places and schools; and date anything that changes.
Is it acceptable to write school marketing in a third-party voice?
Writing informatively rather than promotionally is good practice. Writing so that a reader believes an unaffiliated third party produced it is not. Three lines should not be crossed: do not invent statistics, do not attribute endorsements that were not given — the Federal Trade Commission’s rules on endorsements and testimonials carry real penalties in the United States — and do not present marketing as independent journalism. Everything legitimate in the approach survives those rules.
What is the most valuable use of AI for a school owner who writes?
Being interviewed by it. Asking the tool what a parent would want to know that you have not covered, and then answering its questions, surfaces material no prompt would have produced. Your knowledge is the scarce input; the writing never was.
