AI can read a folder of a small business's client calls, emails and proposals and return a list of the problems customers mention most, in the words they used. That is the whole method, and what it changes is how you write about your own work.
Your vocabulary is the wrong one
You know your service by its name, so you describe it by its name. Your customers do not know the name. They know the symptom. For example, a bookkeeper might sell "month-end close support", while her clients keep saying that they dread the day the accountant emails asking what a payment was. A homepage built on the second phrase gets a nod from the visitor; one built on the first gets a polite skim.
The Beachway headline shows how large that gap can be. The copywriter Joanna Wiebe could not get straight answers from past patients, so she read the reviews of six books about addiction, and wrote down the readers' own wording. The line she picked, "If you think you need rehab, you do", beat the old headline by more than 400% in clicks on the sign-up button. It was one test on one site, so treat it as an example, not a rule. It worked because readers were already asking themselves whether they needed rehab, and the line finished their thought. Nobody writing from the inside would have chosen it, because it is blunter than a professional would dare to be.
AI makes the reading cheap, and the wording fragile
The slow part of that method was reading. Her review pass took a few hours for 500 reviews. A consultant with twenty call transcripts would face an afternoon; with AI the first pass takes minutes, so an owner can do it every year instead of never. The Fast Company piece gives a prompt for it. Ask for the problems mentioned most often, the outcomes people want, the hesitations about hiring you, and the phrases that keep coming back. Group them by theme, and require word-for-word quotes with the source of each.
The quotes are the point, and they are also where AI is weakest. Left alone, an assistant writes a tidy summary, and the tidy version is the opposite of what you want, because it replaces the customer's blunt phrase with a smooth one. Asked for quotes, it can start with real words and drift into words nobody said. So search your transcripts for a distinctive phrase from each quote you plan to use, and delete any quote you cannot find. For twenty quotes, that is about twenty minutes. Remove client names before you upload anything, especially if a contract limits how client data goes into AI tools.
Include the people who did not hire you. Their objections are the clearest record of what makes a stranger hesitate, and they are the lines you can answer before the proposal is even sent.
What to do with the result
The document is a source for several jobs, not a piece of copy. The phrases that appear in most conversations belong in the first screen of your website. The hesitations become the questions you ask on discovery calls and the answers in your proposals. The same wording also goes into email sequences and landing pages.
Because customers change how they talk, the document ages. A yearly refresh with the same prompt costs an hour or so, and it is how you notice that the phrase customers used last year has been replaced by a new one. Owners who do this will end up sounding like their customers; owners who do not will keep sounding like themselves.