Why Does Everything Sound Like ChatGPT Now? What the Research Says

You're Not Imagining It

If LinkedIn posts, cover letters, product pages and even some conversations have started to blur together, there's evidence behind the feeling. Researchers have measured AI's fingerprints in academic papers, peer reviews and spoken language. The effect comes from three directions at once.

Reason 1: A Lot of What You Read Was Drafted by AI

The simplest explanation is volume. A study of more than 15 million PubMed abstracts found that at least 13.5% of 2024 biomedical abstracts were processed with large language models, and up to 40% in some groups of papers. The researchers spotted it through "excess vocabulary": the word "delves" appeared 28 times more often than expected, "underscores" almost 14 times and "showcasing" almost 11 times. They found the vocabulary shift larger than the one caused by Covid. (Kobak et al., Science Advances 2025)

Peer review shows the same pattern. A Stanford-led study estimated that between 6.5% and 16.9% of peer reviews at several major AI conferences could have been substantially modified by LLMs. In one conference's reviews, "meticulous" appeared about 35 times more often than before. (Liang et al., 2024)

If that's true in science, where writing is scrutinized, it's at least as true in marketing copy and social media posts. For a full list of the words that give it away, see words ChatGPT overuses.

Reason 2: People Are Starting to Talk Like AI

This is the surprising part. Researchers at the Max Planck Institute for Human Development analyzed hundreds of thousands of YouTube academic talks and podcast episodes and found that words ChatGPT favors, such as "delve," "meticulous" and "showcase," became more common in human speech after ChatGPT's release. Many grew 25% to 50% a year. (Yakura et al., arXiv)

The latest version of the paper adds an experiment: after a short interaction with a chatbot, people started using its words themselves. So even text written entirely by humans now carries some AI flavor, because the humans have absorbed it. It's one reason careful human writers get falsely flagged by detectors. (AI detector false positives)

Reason 3: AI Makes Different Writers Sound the Same

AI help doesn't just add a few words. It pulls writers toward a shared middle. In a study published in Science Advances, writers given AI-generated ideas produced stories rated more creative individually, but the stories became more similar to each other. The authors called it "an increase in individual creativity at the risk of losing collective novelty." (Doshi & Hauser, 2024)

A Cornell study found something more pointed: AI writing suggestions made Indian and American participants' writing more similar, mostly by pulling Indian writers toward Western styles. Suggestions defaulted to pizza and Christmas, and descriptions of Diwali lost their specific details. (Cornell Chronicle)

Reason 4: The Models Are Trained Toward One Voice

Why do ChatGPT, Gemini and Claude often sound alike? Part of the answer is how they're tuned after initial training. People rate model responses, and the models learn to produce what gets rated highly.

• Less variety. Research on reinforcement learning from human feedback found it "significantly reduces output diversity" compared with earlier training stages. (Kirk et al., ICLR 2024)
• A pull toward the familiar. A 2025 paper traced "mode collapse" partly to typicality bias: raters systematically favor familiar-sounding text, so models learn to produce it. (Zhang et al., Verbalized Sampling)
• Agreeable by default. Anthropic researchers showed that assistants tend toward sycophancy, telling users what they want to hear, partly because people prefer it in ratings. (Sharma et al.) That's where much of the upbeat, validating tone comes from.
• Different models, same answers. A NeurIPS 2025 study dubbed it the "Artificial Hivemind": not only does one model repeat itself, different models give strikingly similar answers to open-ended questions. (Jiang et al.)

Put together: models learn a polished, agreeable, familiar style; millions of people use them; and the style spreads into human writing and speech. Wikipedia editors now keep a detailed catalog of its tells. (Wikipedia: Signs of AI writing)

The Sound Itself

Here's what the convergence produces, in practice:

• Inflated significance: everything is "a testament to" something or "a pivotal moment."
• Negative parallelisms: "It's not just a tool, it's a mindset."
• Groups of three, constantly.
• Even rhythm: sentences of similar length, paragraph after paragraph. (The case for short sentences)
• Dashes and tidy summaries: an em dash in every paragraph and a closing line that restates the point. (Why ChatGPT uses so many em dashes)

How to Make Your Writing Not Sound Like Everyone Else's

• Lead with something only you know. A detail, a number, a real example. The convergent voice is generic by design; specifics break it. (What makes writing sound human)
• Say what you actually think. Agreeable, balanced filler is the model's default. An opinion is not.
• Vary the rhythm on purpose. Short sentence. Then a longer one that takes its time.
• Teach the AI your voice if you use it, rather than accepting its default. See how to make ChatGPT write like you or build a Claude writing skill.
• Check before you send. Run our 10-point self-check, or paste the text into our free AI content detector to see which signals stand out.

And if you're cleaning up an AI draft that sounds like everything else, Simply Humanize rewrites for varied rhythm and plainer phrasing. Then add what only you can.