What makes an AI sound like one?

If you don’t instruct an LLM, it will produce plain “AI-coded” texts

<aside>

Language and wording

• Excessive use of empty fancy words: "Plethora,” “Myriad of,” "Facilitate,” “Optimal/optimize,” “Robust,” “Leverage" (as a verb), "Utilize" (instead of "use”), "In terms of,” "It goes without saying,” "Needless to say,” "Crafting" (unless talking about actual crafts), "Synergy,” "State-of-the-art,” “Engage with" (instead of simpler verbs), "Impactful,” “delve,” “notable,” “intricate,” “realm,” “multifaceted,” “unleash,” “mosaic,” etc. </aside>

<aside>

Structure

• Long chains of long sentences
• Perfectly matching in size paragraphs
• Too symmetrical structure
• Like written by A+ high school student without personality </aside>

<aside>

Syntax and punctuation

• Long, complex-compound sentences in non-academic writing
• Perfect punctuation
• Em-dashes—like these one

Note: some people—me including—are just “punctuation pedants”

</aside>

More ideas

I asked ChatGPT o3 model to come up with ideas on making LLM’s prose sound less like an LLM.

TL;DR – Think like a slightly distracted, opinionated human: inject personal quirks, mild imperfections, concrete lived details, and rhythmical surprises. Below is a buffet of 40+ concrete moves (with ready‑to‑paste prompt snippets) that collectively blur the “LLM gloss” without making the prose sloppy.

Syntax & Rhythm

1. Micro‑sentences + monsters Prompt: “Mix 1‑ to 4‑word sentences with 40‑word monsters.”

2. Fragments on purpose Prompt: “Allow deliberate sentence fragments (e.g., ‘Strange, but true.’).”

3. Mid‑sentence parentheticals “Drop clausal asides (– like this –) that interrupt flow.”

4. Anacoluthon (broken train of thought) “Occasionally abandon a sentence and—well, you get it.”

5. Right‑branching vs. left‑branching shuffle “Alternate ‘Because …, I …’ with normal SVO order.”

6. Inconsistent enumeration “Use (a) letters, 2) numbers, • bullets – randomly.”