Why AI Adoption Is Narrower Than It Looks: Jon Benson

Every leadership team buying AI tools is asking a version of the same question: is this market already crowded, or is serious use of these systems still confined to a small group? The volume of AI products suggests the first. Jon Benson’s answer is the second.

Benson created the video sales letter in 2006, the long-form video-and-voiceover format used to sell a product, and now runs BNSN.AI. Put plainly, the platform turns information about a business into a finished marketing campaign. You give it a website address or a book, answer questions for about 15 minutes, and it decides which funnel the offer needs and writes the copy, ads, emails and sales assets to run it.

What follows sets out why he treats the glut of AI content and the real depth of adoption as unrelated, why he built specialised agents instead of using one general model, and what he takes the reliable return to be. The decision underneath is a live one: whether to buy general-purpose tools and point them at your own work, or to buy systems built by people who already know your trade.

From Sales Letters to Software That Writes Them

Before any of the AI work, Benson invented a format, and nobody was selling with long-form video, words and voiceover when he started. Having come to marketing sideways as a fitness author, he found himself a copywriter without planning on it.

By 2010 his attention had moved to software, for a practical reason: video sales letters are hard to write, and the finished script runs to something like 7,000 words on his account. If the structure could be encoded, other people could use it.

That encoding is what mattered later. Feeding early language models blocks of text laid out like Mad Libs, with whole paragraphs broken into slots for the machine to fill, his team was teaching models the patterns their own software already used. Output stayed poor until the patterns were supplied.

For a founder wondering which part of their business could become a product, the sequence is the lesson. Two decades of manual practice produced the method, and the method, not the model, is what the software sells.

Why One General Model Cannot Write a Whole Campaign

Ask a general chatbot for a video sales letter and it will not sound right, in his view, because no single model is good at every part of it. Different sections go to different models at different temperature settings, each combination mapped by hand.

The customer sees none of that. They paste in a URL or upload a book, and the first agent they meet is a digital version of Benson conducting the interview he once conducted in person. Research then runs in the background on competitors and on what the target buyer says in forums.

Only then does the system choose the shape of the campaign. Over a hundred funnel variations sit behind it, and customers are not asked to pick between them, since on his reasoning most are not marketers and would not know which assets a campaign needs.

Marketing leaders face a sharper test here than whether the writing is any good. The question is whether a system makes the structural decisions, choosing which assets for which traffic, or only writes text once a human has made them.

Why Serious AI Adoption Is Narrower Than It Looks

Asked whether his market is saturating or still open, Benson separates two things that usually get discussed as one. Fatigue is real, driven by the pace of new releases and the quality of most of what they produce. Depth of adoption is another matter.

His evidence here is a recollection. Having run a survey of his own list about coding tools, he came across a graphic he remembers Anthropic publishing on the same subject, a grid of boxes representing its subscriber base:

95% of the boxes were empty they were gray and then you had like six boxes that were yellow and then one box that was red and that one box is where we are.

What he took from it was that the field looks far larger than it is, and that most people have no idea what these tools can do.

The visible surface keeps filling with generic output. He points to the slop accumulating on YouTube and Meta and to the dead internet argument, noting that much AI-made video now feels and sounds identical. His comparison is the dot-com bust: everybody assumed the whole thing was finished, and yet none of today’s big platforms existed then. The collapse cleared out the junk and the serious operators stayed.

Investors reading saturation off the number of AI products are, on his account, reading the wrong signal. A count of tools says nothing about how many businesses have rebuilt a process around one of them.

Selling the Expertise Rather Than the Hours

A friend of his built something comparable for book writing and made it an agency. As Benson describes that model, “they still charge $15,000 for someone to write a book.” His own target is the person who has $300 and wants a book on Amazon.

That choice determines everything upstream: the software has to work without professional help, with the customer adding their own edits rather than commissioning a draft.

Done-for-you is the one thing he does not sell.

Full in-person deployment gets farmed out to two people he trusts, at 15 to 25 grand for two or three days of immersion. His preference is to teach people to run it themselves, which takes a do-it-yourself mindset.

Owners weighing a services business against a product business will recognise the trade. Encoding the expertise pushes price down and volume up, and it forces a standard of quality a consultant between customer and work would otherwise absorb.

Capacity, Cost and the Question of Headcount

Nine people work at the company, and on his estimate they do the work of about 90.

That ratio sits behind the comparison he offers to anyone costing subscriptions against payroll.

name me one employee that can work 24/7, 365 for 500 bucks a month.

A claim in that shape naturally raises the question of jobs. His framing is capacity rather than substitution: the tools let a small team carry work it otherwise could not. The only people he has let go were those who said they did not want to learn, among them a developer he rates highly who spent three weeks on something Benson could do in three hours.

On returns he is careful: some customers make far more than others, no financial guarantee attaches to any of it, and what he will commit to is that anyone applying it properly gets time back.

Technology executives inherit the sharper end of this. Whoever runs technology today, in his view, has to be willing to own AI as well, since the people who resisted hardest read coding assistants as a threat to their role. His head of tech of 15 years is going nowhere, having taken the tools up early enough that he reckons he has written a thousand lines of code in the past year.

A Private Assistant Trained on Your Own Record

The third thing he describes is neither a content generator nor an agent workflow. It is an assistant running on his own machine, trained on his own material, and he has named it Isaac.

To build it he went further than most people would. In his words, “I had it read every email that I ever sent or received for 25 years.”

The purpose was not writing. Because it knows his history, he can put a decision to it, such as which of three projects to sell first, and get an answer shaped by patterns in his own behaviour. A browser chatbot, on his account, cannot do that.

Setup he describes as a copy-paste into terminal, then permissions covering email and calendar access, plus a rules file of what it must always and never do. To get up to speed he suggests about 60 days.

For executives the distinction worth carrying away is between AI used to generate output and AI used to inform judgement. The second only works on private material, which makes it harder to copy and far more sensitive to how a company governs its data.

What separates one company from another here is less whether it uses AI than what it feeds it. Generic tools on generic tasks produce the output the market is already learning to discard. Point the same technology at twenty years of proprietary method, or at a private record of how a business makes decisions, and it produces something a rival cannot buy with the same subscription.

Jon Benson set out his reasoning at greater length in The Future of AI Marketing Digital Clones, Automated Funnels and Human Expertise with Jon Benson on the GrownLearn podcast.

Listen on Spotify Apple Podcasts SoundCloud YouTube