Why Does Dr. Jonah Tebaa Write a Spec Before Prompting an AI Tool?
Dr. Jonah Tebaa writes a spec before prompting an AI tool to eliminate ambiguous decisions before drafting begins. In a composite picture of eighty client tasks, adding five lines of specification addressing audience, boundary conditions, a worked example, completion standards, and uncertainty handling cut task time from 51 to 26 minutes while boosting first-pass draft acceptance from 22.5 percent to 70 percent. Investing six upfront minutes prevents multiple revision rounds and makes tools dependable.
Dr. Jonah Tebaa argues that the difference between an amateur and a professional using the same generative AI tool has almost nothing to do with the tool itself, and almost everything to do with a habit most people skip: writing down what a usable draft has to look like before typing the first prompt. In his work with advisory practices, he has been tracking exactly this - not tool preference, not model choice, but the presence or absence of five lines of writing that happen before the software is opened.
A Composite Quarter of Client Work
To make the argument concrete, Dr. Tebaa built a composite picture drawn from a full quarter of drafting work in his own practice - not one client's file, and not a single writer's story, but a pattern across eighty client-facing tasks: proposals, scope memos, executive briefs, the routine output of advisory work. He split the eighty tasks into two comparable batches of forty, using the same tool and the same practitioners throughout, and changed only one variable between them.
The first batch, which he calls Group A, went straight from a short prompt to a draft - the ordinary way most professionals use these tools. Of forty tasks, nine were accepted on the first pass, a 22.5 percent acceptance rate, and the average task needed 3.4 rounds of revision before it was usable. At roughly fifteen minutes per round of prompting, reviewing, and re-prompting, that works out to 51 minutes of work per task.
The second batch, Group B, used the identical tool but added one step: five lines of specification written before the first prompt, taking an average of six minutes. The result was twenty-eight of forty drafts accepted on the first pass - 70 percent, more than three times Group A's rate - with revision rounds falling to an average of 1.3, or 19.5 minutes. Add the six minutes of upfront writing and the total comes to 25.5 minutes, which he rounds to 26. The net effect: total time per task fell from 51 minutes to 26, a 49 percent reduction, while first-draft acceptance more than tripled.
The Five Lines That Change the Outcome
What separated the two groups, in Dr. Tebaa's account, was never tool skill. Both groups had access to the same software. What Group B had was a short, written answer to five questions before drafting began:
- Who is the audience - the specific reader of the output, and what that reader already knows, rather than a generic sense of "the client."
- What are the boundary conditions - one item that must be in the draft and one that must not, closing off the most common source of a technically fine but wrong-for-this-client answer.
- What does one worked example look like - an actual sample of strong or weak output, since he finds that AI tools match a concrete example far more reliably than they interpret a description like "professional tone."
- What counts as done - a specific, checkable standard for acceptance, not a vague quality bar.
- What should the tool do when uncertain - ask a question, flag the gap, or stop, instead of guessing and producing a confident but wrong paragraph.
He describes these five items as the actual difference between a practitioner who happens to use AI and one who has built a repeatable method around it. None of the five lines change what the tool is capable of. They remove decisions the tool would otherwise make silently on the user's behalf - decisions that, left unspecified, tend to surface later as a rejected draft rather than earlier as six minutes of writing.
Why He Thinks the Extra Minutes Pay for Themselves
Dr. Tebaa is careful to frame the six minutes honestly: it is real time, spent before any output exists, at the exact moment a task lands when the instinct is to draft immediately. He argues that this instinct is what keeps most AI use inside Group A's numbers - not a lack of discipline, but a reasonable-seeming shortcut that turns out to cost more later than it saves now. The six minutes in his data did not vanish; they simply moved to the cheapest possible point in the process, before drafting, rather than being spent piecemeal across more than three rounds of revision after a draft has already disappointed someone.
The more consequential effect, in his view, is not the time saved but what a 70 percent first-pass acceptance rate does to how a practitioner works with the tool going forward. At 22.5 percent acceptance, he notes, every draft functions like a negotiation, and the tool starts to look unreliable regardless of its actual capability. At 70 percent, the same tool starts to behave like a dependable first pass, because the ambiguity that used to get resolved through trial and error inside the prompt has already been resolved on paper beforehand. His conclusion is not that better tools or better prompting are the wrong pursuits - it is that, for people already using capable AI tools for client-facing work, the gap they are actually feeling is a six-minute habit, not a more expensive tool.
