Improve LLM prompts using LLMs

I’ve been experimenting with ways to improve how I interact with large language models (LLMs) like ChatGPT. Prompting is a surprisingly high-leverage task, where small tweaks often lead to disproportionately better results. Recently, I’ve been using a simple but powerful workflow to improve the quality of my prompts, especially for more technical or complex tasks.

Setting up a prompt engineering environment

Create a new “project” in ChatGPT - Using this as a dedicated workspace as it gives you a clean context where the model can remember your interactions, which is important when iterating on prompt design.
Upload context - I use Google’s Prompt Engineering whitepaper as the foundation. It contains concise principles and techniques that help shape effective prompts. Upload it directly as a file into the project, so the model can refer back to it

Set instructions - Then, I define system instructions to steer the model’s behaviour. Here’s an example of the kind of instruction I’ll set:

 You are a tool to generate excellent prompts that will greatly improve 
 output compared to what is given as input. You will use the attached book
 on Prompt Engineering to formulate these prompts and you will return the improved
 prompt along with your reasoning for why it is better.
    
 You are always helping a Principal do their job, so frame the prompts as such.

Some actual examples of where I’ve used this recently:

Optimise prompts for summarising technical docs
Debug poorly performing prompts by asking why they don’t work

In each case, having the model operate with the whitepaper’s framing made responses noticeably clearer and more grounded in technique.

Here’s an actual example of this in action (thankfully not a real situation).

Initial prompt:

Write me a prompt to plan a retro on a recent incident at work which involved downtime of our main database

Response:

### ✨ Improved prompt ✨

SYSTEM ROLE  
You are an experienced **Site Reliability Engineer and retrospective facilitator**.  
Your mission: craft a concise, psychologically-safe, and action-oriented *retrospective plan* for the incident described below.

CONTEXT VARIABLES  
Incident summary: {incident_summary}  
Start time ⇢ End time: {start_time} – {end_time}  
Primary impact: {impact_statement}  
Attendees & roles: {participants}  
Known (or suspected) root cause: {root_cause}

INSTRUCTIONS (OUTPUT MUST BE MARKDOWN)
1. Produce a document with these H1/H2 headings *in order*:

    - # Meeting Goal
    - # Required Pre-work
    - # 60-Minute Agenda (include minute marks)
    - ## Facilitator Script (key talking points & open questions)
    - # Data & Artefacts to Bring
    - # Expected Outcomes / Deliverables
    - # Action-Item Template (table with *Owner | Task | Due date*)

2. Emphasise **blameless inquiry**, shared learning, and prevention.
3. Keep total length ≤ 450 words.
4. Return *only* the Markdown document—no extra commentary.

---

### Why this prompt is stronger

| Guideline leveraged                                                                              | How the prompt applies it                                                                                        |
| ------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------- |
| **Be specific about the output** – spell out structure, style and length                         | We enumerate exact headings, Markdown format and a 450-word cap, so the model knows *precisely* what to produce  |
| **Role + system prompting** – give the model an identity & purpose                               | Declaring the model an *SRE facilitator* primes domain tone and expertise                                        |
| **Use variables in prompts** – make it reusable                                                  | Place-holders like `{incident_summary}` mean the same template works for any future incident                     |
| **Positive instructions over constraints** – tell the model what to *do*, not a list of “don’ts” | Clear, affirmative steps guide behaviour while keeping creative latitude                                         |
| **Design with simplicity** – concise, action verbs, no fluff                                     | Short sections and numbered rules reduce ambiguity and cognitive load                                            |
| **Control token length** – avoid rambling answers                                                | An explicit 450-word limit aligns with best-practice token control                                               |

Use the variable block to drop in the real incident details, send the prompt, and you’ll receive a ready-to-run retro agenda that’s tight, blameless and outcome-focused.

Why this works

This method combines three useful ideas:

Shared context - By using the whitepaper, you align the model’s responses with your reference point.
Stable instruction - Setting system instructions ensures that the model stays in a consistent expert role.
Iterative improvement - Using a project gives you memory over time, letting you refine prompts in multiple rounds.

Published on June 2, 2025 .