AI for Advisors newsletter
As advisors in our AI Powered Financial Advisor program know, I’m a believer in structured prompts, the kind where you spell out role, task, format, context, and always ask AI to ask you questions.
That’s the work of effective prompt engineering, and it matters.
But there are moments when your prompt isn’t a prompt at all. It’s a single word such as “summarize” or “proceed” or “transcribe.”
This is a common scenario: Someone types one word into Claude or ChatGPT and gets back exactly what they needed—tight, focused, a ready-to-use response. Ten minutes later, a different person types the exact same word and gets back vague, generic output. Same one word prompt but completely different results.
Use long prompts to start a project. Use short prompts to steer it.
The difference has nothing to do with the word. It has everything to do with when that word shows up in the workflow.
These aren’t prompts. They’re commands. And what makes them work (or fail) is something called context timing: the relationship between what you type and the context the model already has when you type it.
From prompts to commands
When advanced AI users think about prompting, they think about writing a detailed paragraph, and that’s often the right move, especially at the start of a session. But once you’re inside a working thread and the AI already has the source document, the goal, and the audience, you often don’t need a paragraph. You need a verb.
“Rewrite.” “Shorten.” “Clarify.” “Next steps.”
I call these tactical prompts. They’re not standalone instructions. They’re continuation commands, short, sharp inputs that keep work moving when the context is already in place. Think of them less like a prompt and more like a directive you’d give a colleague who’s been sitting next to you for the last 20 minutes, watching you work.
Context timing in action
A short prompt works when it arrives at the right moment, after enough context has been established for the model to act on it well. The wrong question is “Is this prompt too short?” The better question is “Has the model seen enough to act on it?”
Consider three scenarios:
- You paste a messy meeting transcript and tell the AI it’s for your team’s Monday standup. “Action items” gives you something sharp and usable.
- You’ve written a rough draft of a client email and already described the relationship and the goal. “Polish” does real work.
- You’ve built a long research summary, and the model knows your thesis. “Distill” gets you a crisp paragraph.
In each case, the one-word prompt isn’t doing the heavy lifting. The preceding context is. The prompt is just the trigger.
Here’s what that looks like across a few more typical work situations:
- Drafting a memo? “Formalize.”
- Revising an email? “Shorter.”
- Reviewing a report? “Analyze.”
- Prepping a presentation? “Talking points.”
- Cleaning up research? “Compare.”
- Planning a project? “Sequence.”
- Copying text from an image? “Transcribe.”
- Discussing next steps in a thread? “Proceed.”
The six kinds of context that matter
If context timing is the skill, then it helps to know what kinds of context count. I think of it in six categories:
Source context is the text, notes, file, or transcript already in view. Task context is what you’re trying to get done. Audience context is who this is for. Workflow context is what stage of the project you’re in. Conversation context is what’s already been said in the session. Constraint context covers the specifics—length, tone, format, deadline.
Here’s why these matters: “Rewrite” means something very different depending on whether the hidden context is “for a client email,” “for an executive memo,” or “for a slide headline.” The word is the same. The context around it changes everything.
And not all one-word prompts need the same kind of context. Some transform existing text—“shorten,” “simplify,” “formalize”—and they need strong source context. Some trigger reasoning—“analyze,” “compare,” “evaluate”—and they need data or arguments. Some shape the output format—“action items,” “talking points,” “briefing note”—and they need a clear sense of audience and purpose.
Knowing which type you’re reaching for tells you which context must be in place first.
When short prompts fail
Short prompts break in predictable ways: “analyze” without a document, “rewrite” without a target reader, “prioritize” without criteria like impact, urgency, or effort. The prompt is fine. The context isn’t there yet.
Short prompts are for iteration, not initialization.
Here’s the rule of thumb I give people in my training: Use long prompts to start a project. Use short prompts to steer it.