Prompt

In plain English

Everything the model can see when it starts writing. Your question, the conversation so far, and any material you pasted in.

Because the model predicts a continuation of what it was given, the prompt does not merely ask for output — it sets what kind of output is probable.

A worked example

Two prompts requesting the same thing, and why the second produces usable work.

Thin:  Write about climate policy.

Thick: I am writing a 1,200-word undergraduate essay arguing that carbon
       pricing has underperformed in the EU ETS between 2005 and 2012.
       I have these three sources [pasted]. Draft an outline only, with
       the counter-argument given its own section. Do not invent
       citations; leave a gap where I need to find one.

The thin prompt makes generic output the most probable output, because nothing in it distinguishes this request from a million similar ones.

The thick prompt constrains scope, supplies the actual sources, names the deliverable and forbids the failure mode. Everything specific in it came from the writer.

Note what the second prompt does for a disclosure: it is a record of exactly what was asked for and what was withheld, which is far easier to explain than a chat log of "write my essay".

Why it matters for AI detection

Prompts are evidence. If you are ever asked to account for how a piece of work was produced, saved prompts show the sequence of your thinking in the same way drafts do, and they show what you asked the tool not to do.

They also determine how much of the result is yours. A prompt containing your argument, your sources and your structure produces something you directed; a prompt containing the essay title does not.

Where disclosure is required, naming what you asked for is usually more informative than naming the tool. "I used it to outline and to check my grammar" describes the work; "I used ChatGPT" does not.

Commonly confused with

Prompt engineering

Prompt engineering is the practice of designing prompts systematically, usually for a repeated task in software. Writing one good prompt for one essay is not that, despite the term being applied to both.

Fine-tuning

A prompt changes what the model does this time. Fine-tuning changes the model itself so it behaves differently every time. Prompting is free and temporary; fine-tuning is neither.

Read next

• Prompt library
• How to show you wrote it
• AI disclosure generator