ChatGPT can dazzle us with insightful responses but often falls short when generating complex content like articles or code. Even for humans, crafting such content is difficult—that's why we have editors and peer reviews.

A surprising aspect of ChatGPT is its ability to critique itself, a process that—as we'll discover—can significantly enhance its output quality.

Craft the Perfect Critique Prompt

There are many ways to ask ChatGPT to criticize. Here is one template that I often use when writing:

• As a X,

• Criticize and suggest improvements for the following Y,

• Targeting Z.

Here is a concrete example (this advice taught me a lot about writing!):

As a highly regarded editor of the New York Times blog section,

Criticize and suggest improvements for the following atomic essay (250 words, straight to the point),

Targeting the style of Dickie Cole or David Perell.

These steps are key:

• your role sets the tone,

• what you want improved dictates the focus,

• and the target you aim for guides the language.

Tell ChatGPT to review and fix its own code

Because ChatGPT using its own output as its input, self-critique taps into data from actual review sessions in its training corpus. These will often contain helpful advice which, once present in the context window, will further improve ChatGPT's output.

Here's a programming prompt I often employ (note how the review gets reviewed too!):

'Assume the role of an experienced programmer. Review and refine the previous code, focusing on bugs and clarity. Mention and disregard irrelevant issues.

Asking ChatGPT to critique and correct itself significantly improves its output —making it an indispensable asset in your toolkit.