Diagnosing bad outputs
Worked Repair: The LinkedIn Rewrite, Fixed
Remember Karim's bad LinkedIn rewrite from Module 4? Let's pretend he didn't have the patience to write a long, perfect second prompt. Instead, he stayed in the same chat and walked through the four diagnostic questions. We'll watch the repair happen.
Karim's first prompt vs his repaired prompt — toggle to compare
The starting point
Karim's first prompt was the vague one:
Make my LinkedIn About section better: "Software engineer with 5 years
experience. Love building things. Looking for new opportunities."
The output was the bad one — buzzwords, placeholders, padding, no length control. (Full output is in Module 4 lesson 3.) For reference, here's a snippet of what he got back:
**Transforming ideas into elegant software solutions for 5+ years.**
I'm a software engineer who thrives on the entire journey—from that
first spark of an idea to watching users interact with a polished
product. Whether I'm architecting scalable backends, crafting intuitive
interfaces, or debugging complex systems at 2 AM, I'm driven by one
thing: building software that matters.
[continues with placeholders and a "Tips for customization" section]
Captured from Claude Sonnet 4.5 (claude-sonnet-4-5) on 2026-04-27. Re-runs may differ slightly.
Run the four questions
Karim runs the diagnostic questions on what he just got:
| # | Question | Karim's answer |
|---|---|---|
| 1 | Is the format wrong? | Yes — too many sections; LinkedIn About is one block. |
| 2 | Is the length wrong? | Yes — about 250 words; he wants 110-140. |
| 3 | Is the content wrong? | Yes — placeholders instead of real facts; he hasn't given the model his actual experience yet. |
| 4 | Is the tone wrong? | Yes — clichés ("thrives on", "polished product", "codes with purpose"). |
All four are wrong at once. Good — he can fix them in one follow-up.
The repair prompt
Karim writes a single corrective follow-up. He doesn't restart. He stays in the same chat and writes:
Three problems with that draft:
1. Length: should be 110–140 words, not 250.
2. Format: a single block of 3 short paragraphs, not multiple sections.
3. Tone: drop the clichés ("thrives on", "polished product", "codes with purpose").
Also, here are my real facts to use instead of placeholders:
- Building checkout flows at a fintech (handled $2M/day).
- Worked across React, TypeScript, Postgres.
- Mentored 2 juniors.
- Looking for a senior role on a product team that ships often, ideally remote.
Rewrite from these. First-person.
This single prompt does four things:
- Names the format problem.
- Names the length problem with a specific target.
- Names the tone problem with three banned clichés.
- Adds the missing context as a clean bullet list.
This is the same content that produced the captured "fixed" output back in Module 4 lesson 3 — but framed as a follow-up correction rather than a fresh re-prompt. The fixed output, as a reminder:
I build payment systems that move money reliably at scale. At my current fintech, I designed checkout flows now processing $2M daily across three countries.
My technical edge is in React and TypeScript—I architect frontend systems that handle complex state without breaking. I also mentor effectively: two junior engineers I've guided now own features independently and ship with confidence.
I'm looking for a senior engineering role on a product team that values frequent releases and direct user impact.
Captured from Claude Sonnet 4.5 (claude-sonnet-4-5) on 2026-04-27. Re-runs may differ slightly.
Three short paragraphs. No clichés. Real facts. Length still came in at 79 words — under the 110-word target — but the model flagged that itself, so Karim has the option to expand or ship.
What just happened, structurally
Read what Karim did:
- He stayed in the chat. No restart, no new conversation.
- He ran the four questions in his head. 30 seconds.
- He wrote one follow-up that addressed all the problems at once. Not a re-prompt — a correction stacked on top of the existing output.
- He added missing context that wasn't in his first prompt. (Module 4 was the "got the prompt right the first time" version. This is the "iterate on your way to the same place" version. Both routes work.)
This is what professional iteration looks like — fast, structured, and almost mechanical once you've practised it.
Next: the small set of cases where you really should restart from scratch instead. :::
Sign in to rate