← Back to Field Guides
Prompting · #2 Field Guide

Bad Prompts vs
Good Prompts

Same task. Two ways to ask. Three real pairs that show what changes when you stop typing keywords and start writing prompts.

5 Min Read


The Crash Course covers how to think about prompting. This guide shows the same idea in practice. Three tasks a program manager runs into every week. For each one, the bad prompt is what most people actually type. The good prompt is what gets you something usable.

The outputs below are real — the kind of thing you'd get from Claude or ChatGPT today, not cherry-picked. The bad outputs aren't bad because the model is broken. They're bad because the prompt didn't give the model enough to work with.


Framework

The Three Pieces

A mental model you can reuse

Before the pairs, the framework that explains why one works and the other doesn't. Every good prompt has at least two of these three pieces — and when you can't manage the first two, the third is your escape hatch.

01

Situation

Who you are, what you're trying to do, who the output is for. Without this, the model can only return a generic template — that's the safest answer to a question with no context.

02

Shape

The form you want the output to take: format, length, tone, structure. "Three bullets, under 20 words each, for someone joining the project" produces something different from "summarize."

03

Or, Ask the Model to Interview You

When (1) and (2) are too much to front-load — you're starting cold, you don't know what's relevant yet, the task is fuzzy — flip it. Make the model gather the context. It knows what's missing better than you do.

Copy-Paste

Before you help me with this, interview me one question at a time until you have everything you need to give me a useful answer. Don't draft anything until you've asked enough.

Used at the start of a chat, this often beats trying to write a perfect prompt cold. You'll see it in action in Pair Three.


Pair One

A Status Update for Your VP

No context, no draft

Bad Prompt
Write a status update for my project.

What You Get Back

Here's a status update template you can fill in:

Project: [Project Name]
Status: [On Track / At Risk / Off Track]
Key Updates:
• [Update 1]
• [Update 2]
Blockers: [List any blockers]
Next Steps: [What's coming up]
Good Prompt
Act as my comms partner. I'm a senior PM at a B2B logistics company writing a Friday status update for my VP. My notes from this week:

- Vendor contract finalized; kickoff Tuesday
- Two engineers out sick; sprint slipped 3 days
- Need her sign-off on the revised Q3 timeline

Three short paragraphs covering: where we are, what's blocking, what I need from her. Direct, no fluff, under 150 words.

What You Get Back

We finalized the vendor contract this week and kickoff lands Tuesday — on the timeline we agreed in our last 1:1. The team is in good shape going in.

Two engineers were out most of the week, which slipped the current sprint by three days. We're absorbing it without changing the milestone date, but it eats the buffer we had.

I'd like your sign-off on the revised Q3 timeline before Wednesday so we can communicate it to stakeholders by EOW. I'll send the one-pager Monday morning.
Why It Changed: Without a situation, the model can only give you a template — that's the safest answer to a question with no context. Your job is to give it the situation.

Pair Two

Catching Someone Up on a Long Email Thread

The verb is the problem

Bad Prompt
Summarize this email thread.

[15 emails pasted in]

What You Get Back

On Monday, Sarah emailed the team to flag a delivery delay. James replied asking for the carrier's contact. On Tuesday, Sarah followed up with the carrier's information. James then forwarded the thread to Operations, who replied that they were investigating. On Wednesday, Sarah asked for an update…

[continues for 11 more sentences]

Good Prompt
Summarize this thread for someone joining the project this week. Three bullets, under 20 words each:

(1) what's been decided
(2) what's still open
(3) the immediate next step

If anything is ambiguous, flag it.

[15 emails pasted in]

What You Get Back

Decided: Operations owns the carrier escalation; root cause review due by Friday.

Still open: Whether to switch carriers for the rest of the quarter — Sarah and James disagree.

Next step: Carrier post-mortem Thursday at 2pm. James to send invite.

Flag: cost impact wasn't quantified anywhere in the thread.
Why It Changed: "Summarize" is too open. The model picks something — usually a chronological recap, because that's safe. Specify what kind of summary, for whom, in what shape.

Pair Three

Drafting Q4 OKRs for Your Team

When you can't front-load context

Bad Prompt
Help me write Q4 OKRs for my team.

What You Get Back

Here's a sample OKR framework you can adapt:

Objective 1: Improve team productivity
• KR: Increase output by 20%
• KR: Reduce cycle time by 15%

Objective 2: Enhance team collaboration
• KR: Hold weekly cross-functional syncs

[continues with generic placeholders]

Good Prompt
Before you help me draft Q4 OKRs, interview me one question at a time until you have everything you need: my team's role, current priorities, what we shipped last quarter, what's on fire, and what success looks like for Q4.

Don't suggest anything until you've asked enough.

What You Get Back

Got it. Let's start with the basics — what does your team do, and how big is it? (E.g. "8-person platform team owning the internal data tooling.")

[Then it asks the next question after each answer. By question 5 or 6, it has the situation. The OKRs it drafts are about your team, not a template.]
Why It Changed: When you can't front-load all the context yourself, don't try. Make the model gather it — interview-style prompts work because the model knows what it needs better than you do.

Recap

Same framework, three different tasks.

Situation. Shape. Or interview.

Build the habit once, get it on every prompt. The pairs above all came down to the same three pieces — applied differently each time.

Try It

Confused by a term? See the Glossary →