A different prompt for analysis, research, drafting or brainstorming

A brand coordinator needs a name for a new line of protein bars. She reuses the prompt template that works for the monthly sales analysis: a precise list of constraints and “the best name, with its justification.” She gets one name. She asks for alternatives and two nearly identical ones arrive. The creative director calls them “safe and forgettable.”

The prompt wasn’t badly written. It was written for another task type.

Four tasks, four requests

Task	What you’re after	What the request needs
Brainstorming	Breadth	Many options, varied, from different angles; filter later
Research	Reliable information	Sources, and that it flags what it couldn’t confirm
Drafting	A text for someone	Audience, purpose, tone and format
Analysis	A conclusion from data	The data, the criterion, what to do with gaps, and that it shows its reasoning

Brainstorming: open first, close later

Asking for “the best” with all the constraints at once closes the search before opening it. To generate ideas, split into two steps:

1. “Give me thirty names from different approaches: descriptive, invented, evoking plant origin, playing on energy. Don’t filter them yet.”
2. “From that list, keep the ones under three syllables that work in English and Spanish.”

Adding more constraints to the first step narrows the range even further. Asking it to score its own name from one to ten doesn’t generate variety.

Research: separate from drafting

A project manager asks for “a polished two-page memo on the market for courses in a neighbouring country, with size, competitors and accreditation rules.” The memo reads well, but gives a market size with no source, names a competitor that closed two years ago and sums up a legal requirement in one confident sentence.

The same style works for her when she drafts internal updates from her own notes, because there she supplies the content. When the content has to be found, the request changes:

• First, research: “Gather findings on market size, active competitors and accreditation rules. For each one, the source. Flag separately what you couldn’t confirm.”
• Then, verify the figures the decision rests on.
• Only then, draft the memo with what was confirmed.

Asking for the finished text up front mixes searching and writing, and whatever couldn’t be confirmed is disguised by the prose. Adding “make sure everything is accurate” doesn’t change that.

Drafting and analysis

For drafting, everything from the domain’s first lesson applies: who it’s for, what it’s for and how it should come out, plus an example if tone matters.

For analysis, give it the data (don’t ask it to assume it), state the criterion (“compare by cost per customer, not total spend”), what to do when a figure is missing, and ask it to show the calculations or steps, so you can follow how it reached the conclusion.

Many jobs combine types: a proposal may need research, then analysis and finally drafting. Each stage with its own request, chained.

Common traps

• Using the same template for everything because “it works.”
• Asking for a single “best” answer in a brainstorm.
• Asking for the final document when the task is research.
• Responding to unsourced figures with a more cautious tone or a shorter text. Fewer claims or more “maybe” don’t make them verifiable.
• Adding “be accurate” or “be creative” instead of changing the shape of the request.

On the exam

Notice what type of task the scenario describes and whether the prompt used belongs to another type: a common clue is that “the same style works well for something else.” For ideas, look for the option that asks for many varied alternatives and filters later. For research, the one that asks for findings with sources and unconfirmed points flagged, before drafting. Rule out the ones that only adjust tone or length or add adjectives.