Engineers and support leads who hold the sharpest knowledge about churn, objections and broken promises rarely publish, and AI now separates knowing from writing. OpenAI's September economic research, covering more than 1.5 million work-related ChatGPT messages from April through July, found repeat cross-field use among roughly 6,200 workers climbing from 13.1% to 25.9%. For business this matters because the public voice of a company can shift from marketing to the people who answer the hardest customer questions.
How experts become authors with AI help
Most startup content is still written by whoever can write, not whoever knows the subject. A marketer interviews an expert for twenty minutes, then spends a week polishing notes into something vague, which the expert approves without enthusiasm because real work waits. Companies with budgets solved this with ghostwriters and communications teams, while others left sharp thinking locked in Slack threads, call notes and hallway conversations. The new pattern is different: an engineer talks through a problem for fifteen minutes, receives a clean draft, corrects errors and adds an example only they would know.
The mechanism described in the source keeps authorship with the expert and assigns craft to the model. The expert supplies knowledge, usually by talking, brings examples and background, and puts their name on the claims. AI turns that input into a draft, and the expert then checks the result and sharpens the argument. OpenAI data supports this behavior: when people asked for help outside their field, they were more likely to provide context and ask the model to verify output. Drafting stops being the bottleneck, while standards and verification become the core work.
The background for this shift is the rise of borrowing skills across occupations. Between April and July, workers increasingly turned to AI for tasks outside their own occupation, with the metric measured against their in-role AI activity. Writing fits this use well because many experts think clearly out loud, explain a problem to a customer in five minutes, yet freeze at a blank page. Earlier, knowing something and writing it down required two different skills, and most specialists had time to build only one. Now the second skill can be rented from a model.
What expert-led content means for business
For companies that publish a blog, newsletter or founder LinkedIn, the practical change is a wider pool of authors and faster production. A support lead with two hundred tickets behind them can produce a post that marketing alone could not write, with specific objections and trade-offs. Small firms gain what only executives at large firms once had: a draft without hiring a ghostwriter. Large firms gain scale, turning operators and engineers into regular voices without pulling them away from core tasks for a week.
The limitation is that AI carries only the point of view it receives, and one line about an industry produces text it would give to anyone. Generic posts of identical shape appear when people publish with nothing to say, and writing still exposes holes in weak ideas. The source recommends an editor role: check every fact and number, cut anything the expert would not say out loud, push back where the argument is thin, and hold the line on company positions. That editor can be a founder, a marketer or a fractional hire, but the function must exist separately from drafting.
The marker to watch is whether cross-occupational AI use keeps rising beyond the 25.9% level recorded by July and whether expert-signed posts replace marketer-interviewed posts in company channels. If support and engineering bylines grow while content keeps specific cases and numbers, the model of editor plus AI draft is working. If feeds fill with smooth but interchangeable articles, companies borrowed the tool without moving the knowledge.
