Twenty Minutes Before You Prompt

FREE WORKING SESSION . FACULTY

Twenty Minutes Before You Prompt

Twenty minutes, by hand, on one real piece of your own writing. You finish holding a Distinctiveness Brief you paste above your next prompt, so the model compresses your material instead of the average of your field.

Nothing here is generated. Nothing is saved, nothing is sent, and there is no account. Everything stays in this browser tab.

A PNAS study published August 11, 2026 found that proposals with heavier language model involvement were less semantically distinctive, sitting closer to work the agency had already funded. The model was not failing. It was returning the center of everything already written, because the center is all it was given. This session is the twenty minutes that give it something else.

Before you start

Pick one real thing you are about to write or have just written: a specific aims page, an abstract, a course description, a program review section. One. Not a category. Have it open in another tab. Answer every box about that piece.

Write in fragments. This is not prose. Nobody reads it but you and the model.

SEED . What only you know

The training data contains what has been published. Start with what has not.

1. The unpublished observation

One thing your lab, your classroom or your caseload has seen that is not written down anywhere yet. A pattern, a result, a recurring failure.

2. The number nobody else has

A count, a rate, a duration from your own setting. If you do not have one yet, write the number you would need to collect and where it lives.

SPROUT . Why this problem is yours

A model can describe the problem. It cannot say why you are the one standing in front of it.

3. The constraint you work under

Budget, population, calendar, equipment, accreditation, contract. The limit that shapes your version of this work and is not in the literature.

4. Who is actually in the room

The specific population you serve, described the way you would describe them to a colleague, not the way a demographics table would.

GROW . What you already ruled out

Negative results are the single largest body of knowledge that never reaches training data. Yours are a competitive advantage.

5. The version that did not work

The approach you tried and abandoned, and the one thing it eliminated. A model will propose it back to you unless you close the door first.

6. The consensus you disagree with

Name the position your field currently holds that you think is wrong, and the reason. This is the field’s center. Naming it is how you stop drifting into it.

HARVEST . The sentence only you could write

One sentence. If a colleague in your field could have written it without knowing you, go back to box 1.

7. The one sentence

Build my Distinctiveness Brief

Your Distinctiveness Brief

Paste this above your next prompt, before the request. Then ask for the draft.

Copy the brief

That is one.

Twenty minutes bought distinctiveness for a single piece of writing. Now count the rest of your year. The specific aims page. The two abstracts. The program review. Three syllabi, and the course description for the new prep. Every one of them starts at the same blank box, and every one of them drifts to the same center unless you do this again.

Doing it by hand every time is the honest answer, and it is the answer most people abandon in week three.

The Course Record is the version you write once. It is a five minute profile of what you teach, who is in the room, and how you sound, and every tool in the Harvest Kernel Faculty Toolkit reads it before it drafts a single line. Same principle as the brief you just built, except you do not rebuild it from scratch on a Sunday night. Build your Course Record.

Source

Qian, Y., Wen, Z., Furnas, A. C., Bai, Y., Shao, E., and Wang, D. “The rise of large language models and the direction and impact of US federal research funding.” Proceedings of the National Academy of Sciences, August 11, 2026. doi.org/10.1073/pnas.2601439123