cat writing/prompt-is-the-first-page.md
Your Prompt Is the First Page
We type into ChatGPT like it's a Google search bar, then wait for a miracle. A prompt isn't a question. It's the first part of a page, and the model writes the rest.
Most of us type into ChatGPT the way we type into Google. One line, press Enter, wait for a miracle.
“Write me a post about marketing.”
That’s not a prompt. That’s a wish.
A search bar finds. A model writes.
Google’s search bar looks for pages that already exist. A few keywords are enough, because the page is out there.
A model has nothing to find. The answer doesn’t exist until it writes it, and it writes from one place only: the text in front of it.
That’s the definition I work with:
The prompt is all the text the model reads before it replies.
All of it. Your message, the earlier messages in the same chat, and any instructions the app adds that you never see.
So you’re not asking it a question. You’re writing the first part of a page, and it completes it. Under the chat window, that’s literally what happens: the model reads everything in front of it and predicts what comes next, one small piece at a time. The chat just makes it feel like a conversation.
Why the wish gets a boring answer
Read “write me a post about marketing” as the top of a page. The page is almost empty.
A post for whom? Selling what? Arabic or English? How long? Serious or light?
Most of the time the model won’t stop to ask. It guesses, and it picks the safest, most typical option every time. That’s why you get the post you’ve seen a thousand times: a big opening line, five tips, a row of emojis, a question at the end.
It didn’t fail. It completed a blank page the only way a blank page can be completed.
Three things on every good page
Any good prompt has three things inside it:
- Who. Who is speaking, and to whom.
- What, exactly. One clear task.
- What the answer looks like. Length, language, format, and what to leave out.
Put “write me a post” next to the example from the reel:
“You’re writing for the page of a company that sells software to mid-size companies in Egypt. Write an 80-word post, in colloquial Egyptian Arabic, that explains one feature only. No emojis.”
Read it again with the three in mind. The company, its buyers, the market: that’s who. One feature only: that’s what. Eighty words, colloquial, no emojis: that’s the shape.
Same model. Same subscription. Two completely different answers.
The difference is before Enter
When the answer comes back weak, the first reflex is to blame the model. Switch to another one, or pay for a bigger plan.
Sometimes a stronger model does help. But in that example, nothing changed except the page. The difference wasn’t in the model. It was in what you wrote before you pressed Enter.
Try it
Take the last one-line prompt you typed and rewrite it on this skeleton:
You're writing for [whose voice: a company, a person, a page], and the reader is [who will read it].
[The one task, stated exactly. One thing only.]
Keep it to [length], in [language], as [format]. No [what to leave out].
Then open two fresh chats with the same model: the old one-liner in one, the rewrite in the other. Fresh on purpose, so the first answer doesn’t leak into the second page. If your app has memory switched on, use a temporary chat for both. Compare the answers side by side.
Before you send any prompt, one test: if you handed this page to a sharp freelancer who knows nothing about you, could they do the job without asking a single question? If not, the model can’t either. It just won’t ask. It will guess.
It was never a search bar. It’s a page, and you write the first part.