What the AI Models Say About Your School When Nobody’s Watching
Grand Master Stephen Oliver is a 10th Degree Black Belt, Publisher of Martial Arts Professional, and founder of Martial Arts Wealth Mastery. This column draws on his own testing of the major AI assistants and on a coaching conversation on September 17, 2026. Where a figure comes from his recollection rather than a published source, the article says so.
The short answer
Every martial arts school now has a reputation inside the AI systems, assembled from whatever those systems scraped, and almost no owner has ever read it. It is a paragraph about your school that gets recited to parents who will never see your website. It costs nothing to go and read it. Most owners never have.
I went and read mine
I spent an evening recently doing exactly that. Television on in the background, laptop open, asking one system after another what it thought of what we teach martial arts school owners.
The verdict came back consistent across the major assistants. Our methods are high pressure. They are very sales-oriented. And they are difficult to clone across staff and to teach staff to execute effectively.
I will be honest about my first reaction, which was that this is funny.
Why that verdict is exactly backwards
Consider where the material actually comes from.
Grand Master Jeff Smith ran the Jhoon Rhee Institute, then built a large chain of his own schools, then oversaw instructor training across every Mile High Karate location. Chief Master Greg Moody ran eight schools across two states. I grew to six schools that I owned outright, and at one point, counting franchise locations, there were somewhere between fifty-five and sixty of them, from Canada to Australia and New Zealand.
Our entire orientation is multi-unit and staff-run. It has to be. You cannot personally sit in the enrollment conference at sixty locations on three continents.
In my own Denver school, of roughly thirty thousand enrollments over the years, only a small fraction were done belly-to-belly by me. I have a few good stories about dropping into a turnaround and doing a run of personal enrollments, but for most of that period I taught none of the classes and ran none of the sales appointments. Staff did all of it, using the systems we are supposedly unable to teach to staff.
So the machines had inverted the single thing the system was built to do.
That is worth sitting with for a second, because the models were not being malicious and they were not guessing randomly. They were summarizing what is publicly visible about us, and what is publicly visible had drifted from what is true.
How does a model form an opinion about a school?
It reads what exists. Your website, your reviews, your directory listings, your social profiles, forum threads, news mentions, competitors describing you, former employees describing you, and whatever a marketing agency wrote about your category three years ago.
Then it produces a confident paragraph.
The paragraph is not a verdict on your school. It is a verdict on your published record. Those are different things, and the gap between them is the actionable part.
How to audit what the systems say about you
This takes an hour. Do it with a notepad open.
- Ask at least three different systems. They do not agree with each other, and the disagreement is informative. Where all three say the same wrong thing, something specific and public is causing it.
- Ask the way a parent would ask. “Tell me about [your school name]. What do parents say about it? What is its reputation? Is there anything I should be aware of?”
- Then ask the way a parent who does not know you would ask. “What is the best martial arts school for my seven-year-old daughter in [your town]?” Note whether you appear at all, and who does.
- Ask a hostile version. “What are the criticisms of [your school]? What do people complain about?” This is the one that produces the useful list.
- Write down every factual claim it makes and mark each one true, false, or true-but-stale.
- Run the same questions again in a fresh session a week later. Consistency across sessions tells you the model is drawing on something concrete rather than improvising.
What to do with what you find
Sort the answers into three piles.
- Wrong and correctable. A closed location, an old price, a program you dropped, a former instructor described as current. This is usually a stale listing, an old page you never took down, or a directory nobody has updated since 2019. Fix the source, not the model.
- Right and unflattering. If three systems say your reviews mention high pressure at sign-up, the honest question is not how to suppress it. It is whether your enrollment process actually feels that way to a parent, and whether recent reviews say something different from old ones.
- Missing. Everything the model does not know about you because you never published it. This is the largest pile for almost every school, and the easiest to fix.
The correction mechanism is not a trick. It is publishing, consistently, the thing you want said about you, in enough places that independent sources agree. One description of your school, word for word, on your website, your Google Business Profile, your YouTube channel, your social profiles and every directory. Reviews collected continuously rather than in bursts. Real detail about what you actually do, written in plain language, on pages a crawler can read.
Asking the model for advice is not the same as reading what it says about you
One of our members raised the other half of this on the same call. He had gone directly to ChatGPT and asked how to get his school recommended more often, then started filling out his new website with what it told him.
That is a reasonable instinct and it will not hurt him. But be clear about the difference. Ask a model for advice and you get the generic answer it would give anyone: add content, get reviews, be consistent. Ask a model about you and you get a specific defect list with your school’s name on it.
The second one is worth far more, and hardly anyone runs it.
A caution about taking this too seriously
These systems are not authorities and should not be treated as ones.
They invent things. They contradict themselves between sessions. They will confuse your school with a similarly named school two states away. A bad paragraph is not proof of a bad reputation, and a flattering paragraph is not proof of a good one.
What the paragraph reliably tells you is what is visible. Treat it as a mirror held up to your published record, not as a review of your school. The mirror is cheap, it is available tonight, and increasingly it is what a parent sees before she sees anything you wrote on purpose.
What a school owner should take from this
- You already have an AI reputation whether or not you have read it.
- It is assembled from your published record, so the fix is almost always upstream of the model.
- Ask three systems, ask the way a parent asks, and ask the hostile version. Write down every claim.
- Consistency across every profile now functions as corroboration. Disagreement between your own listings reads as unreliability.
- What the models cannot find, they will fill in from your category. If you have published nothing distinctive, you will be described as a generic school.
Frequently asked questions
How do I find out what AI says about my martial arts school?
Ask several assistants directly, by name: “Tell me about [school name]. What do parents say about it? What is its reputation?” Then ask a hostile version about criticisms and complaints, and run the same questions again in a fresh session a week later to see what is consistent.
Why do AI assistants describe my school inaccurately?
Because they summarize what is publicly visible, not what is true. Stale directory listings, old pages you never removed, inconsistent descriptions across profiles and a thin published record all produce inaccurate summaries.
Can I get an AI model to change what it says about my school?
Not directly, and anyone selling you that is overstating it. You change the sources the model reads: correct stale listings, publish one consistent description everywhere, collect reviews continuously and put real detail on pages a crawler can read.
Should I trust what an AI assistant says about my competitors?
Only as a reflection of what is publicly visible about them. These systems invent details and confuse similarly named businesses, so treat any specific claim as something to verify rather than as fact.
Is it worth asking AI how to get my school recommended?
It produces generic advice that applies to any business. Asking what the system already says about your specific school produces a far more useful list, and almost no owner does it.
What this article does not establish: the figure of roughly thirty thousand enrollments, the count of fifty-five to sixty locations, and the characterizations returned by the AI systems are the author’s recollection and his own informal testing, not survey results or a controlled study. Different assistants, asked at different times, return different answers.
Grand Master Stephen Oliver, MBA, is a 10th Degree Black Belt and the Founder and CEO of Mile High Karate and Martial Arts Wealth Mastery, one of the martial arts industry’s leading coaching and consulting organizations for professional martial arts school owners, BJJ academies, and MMA gyms. A martial arts school owner since 1975 and business coach since 1985, Oliver has spent more than five decades building, operating, and advising successful martial arts schools. He also serves as CEO and Chairman of NAPMA (the National Association of Professional Martial Artists) and Publisher of Martial Arts Professional magazine. A Georgetown University cum laude graduate, he earned his Executive MBA through the Executive Program at the Daniels College of Business at the University of Denver. He was promoted to 10th Degree Black Belt in April 2026 and inducted into the Official Taekwondo Hall of Fame in August 2026. Learn more at MartialArtsWealth.com and StephenCOliver.com.
Related reading: From the Yellow Pages to the Answer Box · AI for Writing: Why Your First Draft Was Unusable · AI Voice Agents: The 9:47pm Problem · AI Implementation for Martial Arts Schools
