World events like war changes the way we speak & write…
…but with the Release of ChatGPT in November 2022, human writing underwent an unprecedented change…
Word events change the way we speak… World War II soldiers carried home words like “G.I.” and “jeep,” terms born in the chaos of battle yet today woven into the general fabric of American life…
Gritty military and war shorthand speech becomes expressions for life’s general messiness…with civilians using it with their own meaning attached to it. Sans any understanding of the origin…we use phrases like gung-ho, eager beaver, head honcho, debrief and more…
Our writings shows the increase use of certain words, and a decrease of others…this is not not caused by some world event, but Generative AI…
No, Generative AI steps into this process of changing, crafting phrases and slang echoing the way returning soldiers once gifted us with new expressions.
World War II still lives on in our words, AI’s introduced a fresh vocabulary and changing the meaning of words; prompt, token, hallucinate, inferenceand more…
Innovation meeting human voice, language evolving through history’s tides
Back to the study
So, how much Generative AI/ChatGPT is in research in general, and in studies…well…being able to track the introduction and use of GenAI in research papers is already a good indication…
These concerns have prompted researchers to investigate their presence in academic texts.
These efforts to measure how often LLMs are used in scientific papers have led to three distinct approaches…
How deeply have LLMs infiltrated scientific writing?
LLM Detectors
The first approach uses specialised detection tools — essentially black boxmodels designed to sniff out LLM-generated text. These detectors are trained on a mix of human-written and LLM-produced samples, learning to spot subtle differences in style or structure.
Word Frequency Modelling
The second method takes a statistical angle, analysing word frequency patterns in scientific texts.
Researchers model these patterns as a blend of two sources — human writing and LLM output — using known examples of each to estimate the mix.
Marker Word Tracking
The third approach is more rudimentary but tangible and practical…
The third strategy zeroes in on specific marker words that LLMs tend to overuse — like delve or intricate, which don’t tie directly to a paper’s content.
By counting these telltale words, researchers can flag texts that might have been polished (or written) by Generative AI, offering a simpler yet clever way to spot LLM influence.
The Telltale Words: How ChatGPT Sneaked into Academic Writing
A group of researchers did when they dove into 14 million PubMed abstracts spanning 2010 to 2024.
To spot the fingerprints of ChatGPT and other large language models (LLMs) in academic writing.
Using a clever analysis of excess vocabulary — words that suddenly spiked in use after LLMs hit the scene.
The result is a fascinating glimpse into how technology is reshaping human behaviour, one word at a time.
The Word Explosion
The study found, around 2022, certain style words began popping up more often in scientific abstracts.
Words like delve, intricate, and meticulous surged in frequency.
Why?
These are the kinds of words LLMs love to sprinkle into text to sound sophisticated. By comparing abstracts from the pre-LLM era (2010–2021) to those written after ChatGPT’s introduction, the researchers estimated that at least 10% of 2024 abstracts had been touched by an LLM.
In some fields, countries, or journals, that number soared to 30%.
At least 10% of 2024 abstracts had been touched by an LLM
Scientists are quietly outsourcing bits of their writing to AI. It’s not plagiarism in the traditional sense, but a subtle shift in how ideas are crafted and expressed.
This isn’t just about word choice — it’s about human behaviour adapting to a new tool.
Why These Words?
So why delve and not, say, dig?
LLMs are trained on vast swaths of text, often leaning toward formal or academic tones.
They don’t just write; they embellish. A human might say, We looked at the data, but an LLM might change that to, We meticulously examined the intricate dataset.
It’s a stylistic flex — one that betrays its AI origin.
The study suggests this isn’t random; it’s a pattern baked into the AI’s DNA.
This shift reveals something deeper about us. Humans are drawn to efficiency and prestige. LLMs offer both: they save time and make us sound smarter. But they also nudge us toward a homogenised style.
The Behavioural Ripple
The numbers are striking, but the behaviour behind them is even more intriguing. The study found differences across disciplines — fields like computer science embraced LLMs more than, say, physics.
Countries with heavy academic pressure or non-English-speaking researchers showed higher spikes, hinting that LLMs might level the linguistic playing field.
One can argue that this isn’t just about cheating or laziness. It’s about adaptation. Scientists aren’t ditching their expertise; they’re augmenting it.
But there’s a catch: LLMs can churn out inaccuracies or amplify biases. The more we lean on them, the more we risk diluting the raw, messy humanity of scientific discovery.
What’s Next?
The researchers argue that LLMs have already left a bigger mark on academic writing than global events like the COVID-19 pandemic.
A chatbot outpaced a world-altering crisis in influence.
As AI tools evolve, so will our words and eventually the way we speak and our habits.
For now, this study is a mirror…it shows us how we’re bending to technology — and how technology bends us back.
Next time you read a paper that feels a little too intricate, you might just smile and wonder: who — or what — wrote this?
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.