The context you forgot to give
A disappointing result often points to missing context. Five details can make the next attempt easier to review and use.
When someone tells you AI was not useful for their work, they are often describing the same experience. They asked for something, got back a generic, faintly corporate version of it, and decided the tool does not understand their field.
The tool doesn’t understand their field. It also didn’t need to. What it needed was five things that were sitting in that person’s head and never made it into the request.
The five
1. Who it’s for. Not “for a report”. For whom, with what background, reading it how. “For our accreditation file, read by a reviewer who has never seen our program” gets you different writing from “for the planning committee, who were all in the room.”
2. What you’re actually making. The thing itself, and its real limits. Length. Format. Where it will live. A 200-word agenda blurb and a 200-word abstract are different documents doing different jobs.
3. The source material. The single biggest lever. Anything the tool has to invent, it will invent, and it invents the average version. Give it your data, your notes, your last three examples, your gap statement. There’s a whole module on this: working from your own documents.
4. The limits, including what to leave out. What you don’t want matters as much as what you do. “Don’t use the word robust.” “No claims about outcomes we didn’t measure.” “Nothing that sounds promotional.” People rarely say what to avoid, and it works.
5. An example of good. One paragraph of something you were happy with does more for the result than three paragraphs describing what you want. This is the highest-value thirty seconds available to you, and it’s the one people skip.
Same task, two requests
Write a description for our leadership development program.
You’ll get something serviceable and forgettable. “Our comprehensive leadership development program empowers emerging leaders to unlock their full potential through a dynamic curriculum of…” The writing equivalent of a stock photo. Not wrong. Just no use to anyone.
Write a 120-word description of a leadership development program for our internal catalog. Audience: mid-level managers at a health system, 3 to 8 years into managing people, choosing between four internal programs on one intranet page. They're busy and suspicious of anything that sounds like a retreat. The program: six months, monthly half-day sessions with a cohort of 18, a peer coaching triad, one 360 assessment with a debrief, and a workplace project they choose themselves. Must get across: the real time commitment, that it's a cohort rather than self-paced, and that the project is their actual work rather than a case study. Do not use: "empower", "unlock", "journey", "transformative", "holistic". No exclamation marks. Don't imply it leads to a promotion. Match the tone of this existing catalog entry: [paste one real entry]
The second one gets you something you can publish after light editing. Nothing about it is clever. It’s just a well-briefed request.
When it disappoints, check this list
Don’t rewrite the request from scratch. Work out which of the five is missing.
| What you’re seeing | A likely cause |
|---|---|
| Generic, could be any organization | No source material, no example |
| Right content, wrong tone | No audience, no writing sample |
| Too long, or padded out | No length limit |
| Says things you can’t back up | Nothing about what to avoid, no data |
| Doesn’t sound like you | No example of your own writing |
| Confidently wrong on specifics | You asked it to supply facts instead of working with yours |
Five of those six are fixed by adding one thing rather than starting again. The sixth is a different problem, covered in when not to trust it.
Try this
Take something you wrote in the last month that you were pleased with. Ask for it cold, with a one-line request, and see what comes back. Then add the five things and ask again.
Doing that once with your own writing is more convincing than anything on this page, because you know exactly how far off the first attempt was.