The Problem Isn't
Your Prompt.

Why your AI output is average

You use these tools:

And the output comes back generic. Your instinct is to write a better prompt. But the prompt is not the problem. The problem is how you're handing the model its context.

We read plain text linearly, top to bottom. A model does not. It scans the structure first — headings, lists, bold text, code blocks. Anything that isn't structured gets treated as one generic blob.

  ── Human Reading ──────────────────────────────────────────────
  Top → bottom, word by word, linear

  ── AI Processing ──────────────────────────────────────────────

  Step 1: Structure scan
    # Heading found    → HIGH PRIORITY section marker
    ## Sub-heading     → SECTION BOUNDARY
    **bold text**      → CRITICAL INFORMATION
    - bullet point     → RULE / CONSTRAINT (follow this)
    ```code```         → TECHNICAL SPEC (literal, not paraphrase)
    | table |          → STRUCTURED DATA (process as unit)

  Step 2: Unstructured text
    "big paragraph with everything in one block"
    → AI treats as BACKGROUND / LOW PRIORITY blob
    → Important details inside? Likely missed.
            

Rule 1 — headings High priority signal

The model reads headings as section boundaries. Whatever sits under a heading becomes the context for that section. Anything with no heading above it is low-priority floating text.

In practice: if you're asking the model to do something complex with several distinct parts, break those parts into headings. It processes each section independently and follows it far more accurately.

Rule 2 — bullets and numbered lists Rules the model follows

A rule written inside a paragraph can be skipped. A rule written in a list gets followed. That difference matters most when you need the model to respect specific constraints or steps.

Numbered lists are especially effective when sequence matters. The model treats them as ordered steps and works through them one at a time.

Rule 3 — code blocks Literal spec signal

Inside a code block the model expects literal code and nothing else. Put only code in there, no explanation. Mix the two and it conflates them, and the output becomes unpredictable.

This is a common mistake: people put explanation and code together inside one block. The model then starts paraphrasing the code instead of interpreting it literally. If you want the code followed literally, the code block must contain code only.

Rule 4 — bold text Critical priority signal

The model treats bold text as the highest priority signal. Without bold, everything looks equally important — and when everything is equally important, the model guesses at what actually matters.

Use bold specifically for the things the model must never miss: constraints, hard requirements, exceptions, and critical output format instructions.

Rule 5 — tables Structured data signal

Whenever you're comparing things or handing over structured data, use a table. A comparison written in plain text only gets partially processed. In a table the model takes the whole context in at once, because a table is a self-contained data structure.

Tables are especially powerful when the model has to compare across multiple attributes, or when different cases need different defined behavior.

Copy-paste template

Use this structure for every complex AI task. Fill in the blanks:

# [Task Name]

## Context
[Brief background — 1-3 sentences. What is this, what does it do?]

## Task
[What exactly do you want AI to do? One clear sentence.]

## Requirements
1. [First requirement — use numbered list for ordered steps]
2. [Second requirement]
3. [Third requirement]

## Constraints
- **[Non-negotiable constraint — bold it]**
- **[Another hard constraint]**
- [Softer guideline — no bold needed]

## Input
```[language]
[paste your code / data / config here — ONLY the code, no explanation]
```

## Output Format
[What should the output look like? Table? Code? List? Explain clearly.]

| If comparing options | Use a table like this |
|---------------------|-----------------------|
| Option A            | Description           |
| Option B            | Description           |