Remember that feeling when you first heard about ChatGPT? Mind. Blown. Right? It felt like something shifted in the world of AI writing. Suddenly, anyone could whip up text in seconds – poems, emails, even code. But beyond the initial buzz, a really important question bubbled up: How much is this stuff actually being used? Like, in the real world, day-to-day, when it comes to AI writing?

Well, buckle up, because a new study just dropped that’s got some seriously eye-opening answers. And honestly? It’s a bit of a wild ride.

Hold Up – AI is Writing Our… Consumer Complaints?!

Yep, you read that correctly. A team of researchers from Stanford and other top universities just published a massive, in-depth analysis looking at “how much Large Language Model (LLM) assisted writing has seeped into different corners of society”. And the scope? Let’s just say it’s bigger than you probably think.

They didn’t just scratch the surface either. We’re talking about a deep dive into four major areas:

• Consumer Complaints: Think about those times you’ve had to vent your frustration about a dodgy product or service. Turns out, AI might be lending a hand (or, well, algorithm) in crafting those angry emails.
• Corporate Communications: Press releases, investor updates – all that polished, official language coming from big companies? Yep, LLMs are in the mix there too.
• Job Postings: Searching for a new gig? That enticing job description you just read? It could have been touched by AI.
• United Nations Press Releases: Even on the global stage, in the realm of international diplomacy and policy, AI is making its mark on how information is communicated.

The scale of this study is honestly mind boggling. We’re talking about analyzing hundreds of thousands even millions of documents from January 2022 all the way to September 2024. That’s a lot of data. And what they found? Prepare to have your expectations… adjusted.

ChatGPT: The Starting Gun for an AI Writing Race

Here’s the thing: it’s not like LLMs popped up overnight. But let’s be real, when ChatGPT hit the scene in November 2022, it was like someone flipped a switch. Suddenly, these powerful AI writing tools were accessible to, well, practically everyone. And the study confirms this adoption of LLM writing skyrocketed right after ChatGPT’s release.

Let’s talk numbers, because they paint a pretty clear picture. Take financial consumer complaints for example. By late 2024, the researchers estimate that around 18% of those texts appeared to be touched by LLMs. That’s nearly one in five! And it wasn’t just in one place either; this trend was broadly spread across different regions, maybe even a tad higher in cities.

Then there are corporate press releases. Hold onto your hats, folks, because here the numbers jump even higher. Up to 24% of corporate communication texts showed signs of LLM assistance. Twenty-four percent! That’s nearly a quarter of all those official announcements, investor updates, and company statements.

Job postings weren’t immune either. Especially in smaller companies and get this younger companies. It seems the newer, more nimble businesses are quicker to jump on the AI bandwagon. While overall job postings showed slightly lower LLM usage (just under 10% in small firms), it’s still a significant chunk.

Even at the United Nations showed similar trends. Nearly 14% of UN press release content was flagged as being generated or tweaked by LLMs.

Now, here’s a twist. While the initial climb was super steep, the study suggests that growth started to level off in 2024. Does this mean we’ve hit peak AI writing? Maybe. Or, as the researchers wisely point out, it could be that LLMs are just getting better at sounding human, making it harder to detect their digital fingerprints. We’ll circle back to that sneaky thought in a bit.

Location, Location, Location… and Education? The Surprising Geography of AI Writing

You might think tech adoption is always a straight line: cities first, then it trickles out to rural areas. Right? Well, when it comes to AI writing in consumer complaints, the study throws a bit of a curveball.

They looked at complaints filed with the Consumer Financial Protection Bureau (CFPB) and found some fascinating regional differences. States like Arkansas, Missouri, and North Dakota showed the highest adoption rates. Meanwhile, states like West Virginia, Idaho, and Vermont were lagging behind.

But here’s the really interesting part: major population centers like California and New York showed less variation in adoption levels. It’s almost like the AI writing wave is hitting smaller states harder, or at least, showing up more clearly in their consumer complaints. Could it be that in these smaller states, people are finding LLMs particularly helpful in making their voices heard? It’s definitely something to chew on.

Leveling The Playing Field

Get this: the study also looked at urbanization levels and education. Surprisingly, areas with lower educational attainment actually showed slightly higher rates of LLM adoption in consumer complaints. This flips the traditional tech adoption script on its head! It hints that these tools might be leveling the playing field, empowering folks who might not have always felt confident in their writing to advocate for themselves. That’s a pretty powerful idea, isn’t it?

Corporate Speak, AI Style: Press Releases Get a Digital Makeover

Let’s swing back to the corporate world and those oh-so-important press releases. The study drilled down into three major distribution platforms: Newswire, PRWeb, and PRNewswire. And guess what? The AI writing trend was consistent across the board.

Before ChatGPT? Barely a blip. AI-modified content hovered around a tiny 2-3% – likely just the algorithm’s margin of error. But post-ChatGPT? Boom. A significant jump within just a couple of quarters.

Newswire saw the most dramatic leap, peaking at over 25% LLM-assisted content by late 2023. PRWeb and PRNewswire also saw substantial growth, though a bit less extreme, plateauing around 15%. This suggests a pretty sweeping adoption of LLMs in corporate content creation, starting early in 2023.

Even within press releases, there were nuances. “Business & Money” and “Science & Tech” categories showed the biggest increases in AI-assisted writing. Does this mean companies are leaning on AI more heavily for those technical, information-heavy releases? It certainly seems that way.

LinkedIn Job Hunting in the Age of AI

Job postings – another area ripe for AI disruption, right? The researchers dug into LinkedIn job postings, focusing on smaller companies (because, let’s face it, big corporations often have entire HR departments and well-oiled content machines).

Initially, across all LinkedIn job postings, they saw a small but statistically noticeable 3-4% of content that was LLM-modified. But when they zoomed in on small companies, the picture got clearer.

For these smaller players, there was a gradual but definite increase in AI-assisted job postings. Post-ChatGPT, the estimated percentage rose to around 5-10% across various job categories. Engineering and sales roles seemed to be leading the charge, hitting closer to 10% AI-modified content.

And here’s another twist: younger companies – those founded more recently – were way more likely to use AI in their job postings. Firms founded after 2015 reached 10-15% AI-modified text in some roles. Older companies? Lagging behind, often below 5%. It’s like the digital natives of the business world are much quicker to embrace AI in their recruitment game.

This raises a fascinating point: could AI be helping smaller, younger companies compete in the talent market? By lowering the cost and time of writing job descriptions, these tools might be giving them an edge. But… there’s always a “but,” isn’t there?

The Million-Dollar Question: Are We Losing Our Human Touch in Writing?

This study is packed with insights, but it also throws up some really important questions. As AI writing becomes more widespread, are we sacrificing something along the way?

The researchers touch on this delicate balance. On one hand, AI can boost efficiency, cut costs, and potentially even democratize access to content creation. That’s all undeniably positive. Think about non-native English speakers, smaller businesses, or individuals trying to navigate complex systems – AI could be a powerful ally.

But on the other hand, over-reliance on AI-generated content could have downsides. In corporate communications, for example, are press releases becoming more generic, less nuanced? Are job postings losing their unique company voice, making it harder for job seekers to differentiate between opportunities? And what about trust? If we know AI is increasingly involved in the information we consume, does it erode our faith in the authenticity of that information?

The study authors wisely point out the trade-off: cost efficiency versus credibility. It’s a balancing act that organizations, and society as a whole, are going to have to grapple with.

The AI Detection Dilemma: A Limit to What We Can See?

Now, let’s circle back to that crucial point about detection, because it’s a major caveat in this whole conversation. The researchers themselves acknowledge a key limitation: “Perhaps the biggest limitation in our study is that we cannot reliably detect language that was generated by LLMs, but was either heavily edited by humans or was generated by models that imitate very well human writing.”

Let’s unpack that for a second. The AI tools used to detect LLM-generated text are constantly playing catch-up with the creators of LLMs. As models get more sophisticated, their output becomes harder and harder to distinguish from human writing. And if someone takes an AI-generated draft and then heavily edits it, those digital fingerprints become even fainter.

What does this mean for the study’s findings? It suggests that the adoption rates they’ve uncovered might actually be underestimates. The true extent of LLM-assisted writing could be even higher than the already significant percentages they’ve reported.

Think about it, if AI is becoming stealthier, flying under the radar of detection tools, how can we really grasp its full impact? It’s like trying to measure the wind, you can see its effects, but pinning it down precisely is tricky.

This detection challenge isn’t just a technical hurdle; it has broader implications. If we can’t reliably tell what’s AI-generated and what’s human-written, it blurs the lines of authorship, accountability, and even truth itself. It’s a bit like entering an era of digital ambiguity, where we have to question the origins of the text we encounter.

Looking Ahead: Navigating the New Writing Landscape

This study isn’t just about numbers and percentages. It’s a wake-up call. It shows us that AI-assisted writing isn’t some future possibility, it’s here, it’s now, and it’s becoming deeply woven into the fabric of our communication.

So, what do we do with this knowledge? Here are a few key takeaways and questions to ponder:

• For Businesses: Embrace AI’s efficiency, but tread carefully. Don’t sacrifice authenticity and genuine human connection for the sake of speed. Think critically about where and how AI writing tools are best applied.
• For Job Seekers: Be aware that AI might be shaping the job descriptions you read. Look beyond the polished language and try to get a real sense of the company culture and the actual role. And maybe, just maybe, consider using AI to help your application stand out in a world increasingly influenced by AI-written content.
• For Policymakers: It’s time for thoughtful conversations about AI ethics, transparency, and regulation in the realm of communication. How do we ensure equitable access while safeguarding against potential risks like misinformation and erosion of trust?
• For Researchers: The detection challenge is real. We need to keep pushing the boundaries of AI detection technology to better understand the evolving landscape of AI-human collaboration in writing.

Ultimately, this research highlights a fundamental shift. Writing, a skill once considered uniquely human, is now increasingly becoming a partnership with AI. The question isn’t whether AI will write, it already is. The real question is: how do we embrace this new reality in a way that maximizes the benefits while preserving what truly matters – the human voice, genuine connection, and trustworthy communication?

It’s a question we all need to answer, together.

| Latest From Us

• 
  Forget Towers: Verizon and AST SpaceMobile Are Launching Cellular Service From Space
• 
  This $1,600 Graphics Card Can Now Run $30,000 AI Models, Thanks to Huawei
• 
  The Global AI Safety Train Leaves the Station: Is the U.S. Already Too Late?
• 
  The AI Breakthrough That Solves Sparse Data: Meet the Interpolating Neural Network
• 
  The AI Advantage: Why Defenders Must Adopt Claude to Secure Digital Infrastructure