Skip to content
busyless
Book a call

One prompt is not a system

Why 'write me a blog post about X' tops out at fine, and what the good version does instead.

7 min read

Level 0

Curious

0 XP
150 XP to ApprenticeNo streak yet0 badges0% done

Open a chat box. Type write me a blog post about inventory forecasting. Hit enter.

You get a blog post. It reads fine. It has an intro, some headings, a conclusion. And it will never rank, because a thousand other people got the same fine article this morning.

Here is what that one prompt actually did:

chat
>Write me a blog post about inventory forecasting.
# Inventory Forecasting: A Complete Guide In today's fast-paced business landscape, inventory forecasting plays a crucial role...
No sources. No idea who reads it. Nothing checked. It wrote from memory, and memory is where old prices go to die.
This is the whole session. Two lines. The model never left the room.

What it never did

  • It never looked at what already ranks for that phrase.
  • It never checked a single price or statistic.
  • It never read your product notes, so it guessed what you sell.
  • It never read anything you published before, so it repeats you or contradicts you.
  • It never checked its own work.

None of that is a prompting problem. You cannot fix it with a better prompt, because the tool has no hands. It cannot open a file, fetch a page, or run a check.

Claude Code has hands

Claude Code is the same model with a terminal attached. It can read files on your machine, fetch live pages, run scripts, and edit its own output. That one difference changes what you can ask for.

Same request, run through a real pipeline:

claude
>Write the inventory forecasting guide. Client is VOS, primary keyword "inventory forecasting".
Loading the client folder first. I never describe a product from memory.
Read(Clients/VOS/_system/knowledge-base.md)
⎿Product, pricing, proof points, promotion policy. Verified 2026-07-30.
Read(Clients/VOS/_system/style-overrides.md)
⎿Sentence-case FAQ headings, US spelling, soft-sell only.
Skillwrite-non-listicle9 phases
WebSearch(inventory forecasting)
⎿15 SERP results. Shortlisting the 10 most relevant.
WebFetch(10 competitor pages)
⎿Heading maps extracted. Gap analysis done.
It has now read what actually ranks. That is the difference between writing about a topic and writing against a SERP.
Read(references/research-library.md)
⎿Checking which sources we already over-used.
Outline is ready. Writing section by section.
Bash(python references/article-check.py draft.md)
⎿0 em dashes, 0 curly quotes, 3 body lists, 2 tables, 5 FAQs. Clean.
Fifteen or so tool calls before a single paragraph gets written. That is the product.

One prompt

  • Writes from memory
  • Invents prices and stats that sound right
  • Has no idea what already ranks
  • Repeats your last three articles
  • Checks nothing
  • Ceiling: fine

A pipeline

  • Reads your notes before it writes
  • Verifies every number against a live page
  • Reads the ten pages that beat you
  • Compares against everything you shipped
  • Runs a script that fails on defects
  • Ceiling: publishable without a rewrite

The whole course in one line

You are not learning to prompt better. You are learning to build a pipeline and write down its rules, so the model has something to follow when you are not watching.

Challenge

Put the pipeline in order

60 XP+15 perfect

Seven stages. Tap them in the order they have to run. Get the order wrong and you will do the same work twice, which is exactly how most people burn a week.

Tap in order

    Your pipeline shows up here.

    Do this now

    1. 1

      Install Claude Code and open it in any folder that has your writing in it.

      npm install -g @anthropic-ai/claude-code
cd ~/my-content
claude

      It reads the folder it is opened in. Point it at real work, not an empty directory.

    2. 2

      Ask it to describe your own writing back to you.

      > Read three files in this folder and tell me what rules the writing follows.

      This is the fastest way to see the difference. A chat box cannot do this at all. Watch it open real files.

    3. 3

      Write down the three rules it named that you agree with.

      Those are the first lines of your CLAUDE.md. You build that file in Module 2.

    Finished reading?

    Mark it done to bank the XP and keep your streak alive. Any challenges on this page score separately, so you can come back for a perfect run later.