We work with Claude and ChatGPT every day now but getting them to write well has always been a challenge. They’ve trained on the content of the Internet - good and bad - which has led them to write in some curious forms.

Em—dash is by far the most famous of the AI writing woes. Although, technically it’s not really a writing woe at all, given writing has used it for centuries. However, it’s become such a tell-tale sign we generally remove it.

But punctuation marks aside there are some clear repeating writing patterns which are just a bit… ugh.

Fletch, a product marketing agency, created this helpful graphic:

So we decided to turn this into a system prompt. Copy and paste to projects, custom GPT or custom instructions on Claude and ChatGPT:

** Annoying Sentence Structures — and how to fix them **

- Avoid choppy fragment pairs like "All the X. None of the Y." This is extremely choppy and tries too hard.
All the features. None of the friction.
# To fix, convert the second fragment to a prepositional phrase:
Get all the features with none of the friction.

- Avoid the "It's not X. It's Y." construction. It's a little too cutesy. The staccato vibe is out.
Good leaders ask for help. It's not weak, it's wise.
# To fix, change the first sentence to a dependent clause:
While weak leaders won't ask for help, the best ones know it's wise.

- Avoid the "No X. No Y. No Z." construction. It reads as very twee, approaching children's book territory.
It's next-gen automation. No waste. No guessing. No restarts.
# To fix, start with two bad things and convert the third into a positive:
Avoid waste and guesswork with automation that gets it right the first time.

- Avoid stacking single-word adjectives or short phrases as sentences like "Good thing. Good thing. Good thing." It comes off dated and salesy.
Our strategies? Simple. Effective. Easy.
# To fix, skip one in the list, modify a noun with the other, and convert the last to a verb:
Trust us to deliver simple strategies that work.

As an aside, why not have a watch of our em—dash video? It explains why certain AI writing patterns become a thing:

Reply

or to participate