We delivered an AI Gatekeeper for ourselves — Charleston AI, on Clements Ferry Road — after a viral video took our inquiries from 20 a day to 2,000 overnight. It was designed, built and put to work from a phone in three hours, reads every inquiry as it lands, and separates real Charleston clients from everyone else.

This one is about us. We do not name clients in these stories, and we are not going to start, but the client in this case is Charleston AI, at 405 Jessen Lane off Clements Ferry Road, so the rule does not apply. Charleston AI opened on January 1, 2026. We are eight months old.

For those eight months the Lab has run on a steady, manageable trickle. About 20 inquiries a day, across the website form, the live chat, the phone and email. Some are Charleston people who want an appointment. Some are people from Atlanta or Phoenix who have not read the part about the service area.

A few are vendors. Twenty a day is a number one person can read over coffee, and one person did, every day since we opened. Then a forty-second clip from a session at the Lab went up on a Sunday evening. Nobody here posted it.

By Tuesday morning it had been viewed 38 million times, one of our co-founders had been renamed "Jimmy Rainbow" by several hundred thousand strangers, and the trickle was 2,000 a day. The volume was not the problem. The mix was. Somewhere in those 2,000 were the same 20 or so real Charleston clients as always, and now also a few hundred people from every other state asking for something we do not do, several hundred people who wanted to argue about AI in general, a great many people who only wanted to say "Jimmy Rainbow" to Jimmy Rainbow, a steady stream of people sending the same message 30 or 40 times to see if a human would crack, people impersonating clients to get a reply they could screenshot, and a small number who were simply ugly.

All of it arrived in the same inbox with the same sound. The 20 real ones were still in there. We were losing them. The AI Gatekeeper was built on ChatGPT Work, on a phone, in three hours.

Not on a laptop at a desk, and not over a week of planning, because a viral video does not give you a week. It was Thursday morning. On an iPhone, mostly standing up, between appointments, because the appointments did not stop. The design, the test messages, the rollout to each channel, all of it from the phone.

That was a deliberate choice. By lunch it was reading every channel. If it could not be built that way, we had no business telling a Wadmalaw boat builder or a Shem Creek fishmonger that it could. It sits in front of everything.

Website form, chat, email, and voicemail, which it transcribes. Each inquiry is read as it lands and sorted into five bins: Charleston client, real question, well-wisher, noise, and hostile. Charleston client means a person in the metro describing an actual problem, a restaurant that cannot keep up with reservations, a law office drowning in intake, a retired engineer on Daniel Island who wants to learn. A message that says "can AI do my job" is a question.

A message that says "I run a four-truck HVAC company in Summerville and my dispatcher quit" is a client, and it goes to the top. Every Charleston client gets a reply within the hour, in our voice, with the two things that matter: face-to-face at the Lab, never Zoom, and you pay after, never before. Real questions get a real answer. Well-wishers get one warm line.

Noise is filed and never answered. Hostile is filed and kept, and nobody reads it unless there is a reason to. At 5:30 each day it produces a short list. Charleston clients, ranked.

Questions worth a human answer. A count of everything else. The first day it found 23 real Charleston clients in 2,140 messages. In the five weeks since, it has surfaced 611, and the calendar at the Lab is booked further out than it has ever been.

There is one flaw, and it is ours to own. In the first week a producer from a national morning show wrote in. Then a trade magazine. Then a newspaper in Columbia.

The Gatekeeper filed every one of them as hostile. It read "we'd love to send a crew to your lab" as precisely the kind of thing a troll says, and the more the producer wrote, the more certain it became. Four messages later, she was at the bottom of the pile. We found her by hand, eventually.

We could fix it. We have not. The reporters still get filed with the trolls, and one of us reads that folder on Sunday mornings and calls back the ones worth calling. It is not a bad system.

The Gatekeeper has never once let a real Charleston client slip, and that was the whole point. Everything in this story is what we tell clients every day. Build it small, build it where you already are, and let it do the reading so you can do the deciding. This time we took our own advice on a phone in three hours, on the worst morning to try.

It held.