AI slop
AI words and phrases everyone recognizes, and what to write instead
The words and lines that give away ChatGPT writing, what research found about them, and the plain words to use instead in your posts and replies.
Short answer: Chatbots lean on a small set of words, such as delve, tapestry, testament, pivotal and underscore, and on a few sentence shapes, like It’s not just a tool, it’s a movement or three adjectives in a row. A study of 15 million research abstracts found “delves” used 28 times more often than expected in 2024. Readers have learned these words, so they read as AI on sight. Use the plain word instead, and say one specific thing.
The words chatbots overuse
Wikipedia’s editors keep a guide to spotting AI writing, and it lists words by the years chatbots overused them. The list changes as the models change.
| Years | Words chatbots overused |
|---|---|
| 2023 to mid-2024 | delve, tapestry, testament, pivotal, underscore, intricate, meticulously, vibrant, landscape, garner |
| Mid-2024 to mid-2025 | align with, fostering, enhance, showcasing, highlighting |
| Mid-2026 on | leverage, multifaceted, quietly, steady, prioritize |
Source: Wikipedia, Signs of AI writing, read in October 2026.
None of these words is wrong on its own. They read as AI when they show up where a plain word would do, and when several show up together.
What the research found
The best known study looked at more than 15 million biomedical abstracts in PubMed, from 2010 to 2024, and counted words that suddenly became far more common after ChatGPT came out.
Less common words with strong excess usage included delves (r=28.0), underscores (r=13.8), and showcasing (r=10.7)
In plain terms, “delves” showed up 28 times more often in 2024 than the trend before chatbots would predict. The same study estimates that at least 13.5% of the 2024 abstracts were processed with an AI model.
A study of peer reviews for AI conferences in 2024 found the same thing in another place: between 6.5% and 16.9% of review text may have been substantially changed by AI models. Its authors watched a few adjectives climb, among them commendable, meticulous and intricate. You can read it on arXiv.
The sentence shapes that give it away
Words are easy to spot. The shapes of the sentences give it away just as fast.
| Shape | What AI writes | Write this instead |
|---|---|---|
| The fake contrast | It’s not just a tool. It’s a movement. | Say what the tool does: “It saved me an hour a day.” |
| Three in a row | Fast, simple, and powerful. | Keep the one that’s true: “It’s fast.” |
| The cheerful opener | Great question! | Start with the answer. |
| The setup | Here’s the thing: | Say the thing. |
| The big word | This is a testament to your hard work. | “This shows how hard you worked.” |
| Serves as | This serves as a reminder that... | “This is a reminder that...” |
| The sign off | I hope this helps! | Stop when you’re done. |
In replies, the lines are even shorter
Under posts on X, Threads, Bluesky or LinkedIn, AI replies tend to be one of a handful of lines:
- Great insights! Thanks for sharing.
- Couldn’t agree more.
- This is a game changer.
- Love this perspective!
- What’s your take on the future of AI?
Each one could sit under any post. That’s what gives them away, more than any single word: they say nothing about the post they answer. A reply that mentions something only that post has, like a number from it, a word the author used or what’s in the picture, can’t be pasted anywhere else.
On X, replies like these are also the kind that end up behind Show probable spam.
Should you stop using these words?
No. “Not all text featuring these indicators is AI-generated,” as Wikipedia’s guide puts it. People wrote “pivotal” long before chatbots did. Two habits make your writing read as AI:
- Using the fancy word where a short one works. “Shows” beats underscores. “Look into” beats delve into.
- Stacking them. One vibrant is a word. A vibrant tapestry that underscores a pivotal moment is a chatbot.
The em dash belongs on this list too, and has its own article: why AI uses so many em dashes.
How Twick writes replies
Twick drafts replies with 100% no AI slop, each one about the post in front of you and in your own voice. You pick one and post it yourself. See it on the Twitter reply generator, or read what AI slop is for the bigger picture.
Questions
What words does ChatGPT use the most?
The best known are delve, tapestry, testament, pivotal, underscore and intricate. Newer models moved on to others, such as leverage and multifaceted.
Why does ChatGPT say delve?
Nobody has shown why. The studies above measure how often, and they agree the jump starts in 2023 and 2024.
Is delve still a sign of AI in 2026?
Less than before. Each new model has its own favorite words, so the list keeps moving.
Do these words prove a text was written by AI?
No. They make it more likely when several appear together, next to other signs like stock openers and replies that could sit under any post.