Like many knowledge workers, CCI’s staff in recent months have experimented with the use of AI for both administrative and editorial tasks to varying degrees of success. Editorial director Jennifer L. Gaskin details some of those experiments along her journey from skeptic to adherent and back again.
The editorial team here at CCI is quite small, and like many small offices, we have in the past couple of years begun using AI for a variety of tasks. We’ve given it administrative work like listening to and summarizing meetings, researching other compliance industry media and drafting job descriptions. We’ve also given it editorial tasks like analyzing speeches by government officials, cleaning up interview transcripts and rewriting press releases into news briefs.
AI completed most tasks at least passably well, though in no case did the chatbot do a better job than a human being would have done if they had the time. Still, there are only so many hours in a day and our early experiments suggested that I could lean on AI to help me get more work done in the same number of hours. I somewhat grudgingly started giving AI (I prefer Anthropic’s Claude, for those who are curious) more and more tasks, like analyzing industry survey data and writing news briefs, evaluating contributed articles to see if they were worth publishing and even writing headlines.
Because I stopped doing these tasks, I got worse at doing them, more disconnected from the reasons why the tasks needed doing in the first place and less enthusiastic about my job, all of which made me a much less happy human being.
Claude does everything I ask of it with great gusto and confidence, and I suppose a generous interpretation of the situation would say that I could be forgiven for believing that Claude was doing these things not just with great gusto and confidence but with the level of journalistic rigor and professionalism our readers deserve.
I, like all of you, am inundated with advertising, marketing and reporting that declares how unbelievably advanced generative AI is (and how scared I should be as a knowledge worker). AI can analyze medical images and aid in diagnosis, it can pass the bar exam, it can create fake security camera footage, it can edit Hollywood movies. Anything a person can do, AI can do better, or so say the companies opening up the money faucet full blast, telling their teams to be “AI-first.”
Not only is that not true, but by putting my faith in AI to supposedly decrease my workload (which it didn’t really, but more on that later), I stopped doing things I loved, things that lift my spirit and give meaning to my very existence.
Flying Blind on AI: The New Normal for Compliance Teams
With the Senate's decisive vote against a state AI regulation moratorium, compliance officers face a stark reality: Most organizations are using AI, but not everyone has policies governing the technology as it keeps on advancing. Jennifer L. Gaskin reports on how teams are building risk-based
Read moreDetailsAdoption to rejection: My AI journey
CCI is far from the only company experimenting with AI; in fact, I’d likely have an easier time tallying the number of companies not using AI. We aren’t spending very much money on these experiments, but many companies, especially large enterprises, are shelling out enormous sums. In fact, a Boston Consulting Group survey at the beginning of the year said one in three companies will spend upwards of $25 million this year on AI.
Increasingly, corporate AI initiatives are running into problems. Stack Overflow’s annual developer survey found that positive sentiment for AI tools has gone down as usage has gone up. A Pew Research Center survey in April showed that the general public remains highly skeptical, both about the long-term impact of AI on their lives (43% say AI is more likely to harm than benefit them) and how it will affect their jobs (only 23% of people expect a positive impact). A study conducted by Stanford University’s AI center surfaced a disconnect between how workers say they want to use AI and what the tools are actually capable of doing. One survey even found that many workers feel pressured to use AI, so much so that 16% pretend to. And evidence is mounting that companies are simply lighting money on fire: A new study found that 95% of generative AI pilots in enterprises realized no return on investment.
When AI performed adequately at one or two tasks, I gave it more things to do and perhaps it became a victim of its own success, another example of the Peter Principle, Claude rising to the level of its own incompetence. The more I used it for tasks that required real nuance, the worse it got at doing those tasks.
I wrote a 2,500-word prompt that included multiple examples of well-edited articles with compelling, informative headlines. When it generated recommendations that were fit for purpose, I praised the chatbot and incorporated the good thing it had done into the prompt. I communicated with Claude as I would with a fledgling editor or journalism student; I was clear but helpful in my guidance, explaining not just how I wanted things to be done but why I wanted them done that way.
Yet it would routinely make mistakes that were directly counter to explicit instructions. CCI mostly follows the stylebook published by The Associated Press, which means we do not use the Oxford comma, a topic of great debate in writing circles and even in my own home. Every prompt I give Claude includes this note, but despite my clarity, the chatbot is remarkably inconsistent and routinely includes Oxford commas in its output, particularly when it’s been asked to do multiple instances of the same task, like writing headlines for a few stories in a row. I’ve similarly explained our guidelines around capitalization of words in headlines, the use of percentage signs and the proper formatting of em-dashes (yes, humans do use em-dashes). These rules are routinely broken, meaning I have to fix what Claude screws up, which takes time away from other tasks and, on its face, defeats the purpose of using AI to begin with.
I know that if I were to give these tasks to that fledgling editor or journalism student, they would sometimes make mistakes, too. But the difference is the more they did the same task, the better they would become at the task. That’s because they would eventually learn, which Claude does not and cannot, at least not in the way people do. In humans, learning creates physical changes in our brains. It might take a while, but that young journalist over time would reflexively stop using the Oxford comma because the learning process would rewire their brain, and if they did happen to type a stray one for some reason, they would likely see the issue and delete it on their own.
In addition to obvious mistakes, Claude frequently generates text that suggests it wholly misunderstands what CCI is or even what journalism is. One of my main tasks is writing headlines. Writing a good headline is difficult and usually involves multiple rewrites, but it represents the very essence of journalism: informing and engaging readers with as few words as possible. Roy Peter Clark of the Poynter Institute for Media Studies has called headline writers “the poet among journalists.” A good headline has nothing superfluous but everything the reader needs. I have won awards for how well I write headlines. AI does not write headlines well because it isn’t really reading and absorbing these articles; it’s just guessing at what the next word is. In my experiments, that often led to unclear, vague or misleading headlines that I would have to spend time rewriting, again defeating the purpose of using AI in the first place.
This AI fatigue I’ve experienced puts me in good company but has also reignited my deep skepticism about this technology. I am convinced that nobody but a human being can write headlines to my standards.
When I ask Claude to help me write a headline, here’s what it’s not doing that a person would be doing: Reading the material multiple times, drawing out nuance, picking out interesting turns of phrase or rhetorical connections the author has made. Making an attempt to craft a headline that summarizes the material, draws the reader in, plays on a cultural theme. Deleting everything and starting over. Opening a new browser tab and going to Thesaurus.com or IMDb.com for inspiration. Settling on something that will work but isn’t great and moving along to another task. Having a sudden burst of inspiration and realizing that’s only because it read that headline on another article and it became lodged in its subconscious. Walking around the office a few times, thinking out loud about the article. Going back to the version that wasn’t great before deleting everything and starting over again. Eating lunch. Realizing this can only go on so long before a deadline gets missed and riding that mini-wave of adrenaline to finally, at long last, identify and type the exact right words.
If that sounds like a torturous process, it is. And I love it. It’s like putting together a puzzle. I’m lost, nowhere, stuck in the weeds — until I’m not. Sometimes a headline writes itself, and sometimes authors write their own really great headlines. And sometimes that burst of inspiration comes just as I’m about to hit “publish” on the not-so-great headline and I then have to redo other elements of the article. But that process of trying, failing and succeeding is energizing for me, something I didn’t realize I would miss.
Dumber, worse at my job & moody
Research shows when people rely on AI, they use less of their brains, which is part of its allure and one of those benefits AI companies won’t stop talking about. Offload some rote tasks to AI and you’ll be freed up to use your brain in other ways, AI proponents say.
But it’s not really that simple. An MIT study published this year showed marked differences in brain connectivity between groups of writers based on their use of AI. And with deference to the researchers, who implore journalists not to use words like “stupid” when describing their findings, my experience was exactly that — Claude made me dumber and worse at certain aspects of my job.
The lengthy prompt I mentioned earlier was designed to help me scale some of the tasks I do for CCI, namely writing headlines, subheadlines (or deks, as we call them) and introductory blurbs. Using Claude for these tasks saved me hours every week, there’s no doubt about that. I could do more with my time using Claude than I could just on my own. But those time savings had a series of downsides. For one, I got so used to using the bot that I began to forget how to do things.
One day when Claude was down and I was forced to do my work unaided, I froze and for a while spun my wheels because I had gotten so out of practice. It took me some time to remember that I do know how to do this work and do it better than any machine ever could because I am a human being.
I have always done creative work, save for some college jobs where I worked retail. That’s not by accident. I believe that writing, editing and designing information for people is what I am meant to do with my life, at least the part of my life that’s about making money.
I’d go beyond that and say that not only is creative work the best way I know how to earn a living, it’s also self-care. Research has tied creativity to well-being, and my own experience reinforces that positive relationship. When I write a clever headline or perfectly succinct blurb, the happy chemicals my brain produces elevate my mood. By offloading those cherished-but-difficult tasks, I had unwittingly made myself much less happy.
AI-last policy
Nobody forces me to use AI professionally; there is no corporate mandate at CCI that we become AI-first. But I do feel deeply for people who work at companies where that is the case. It’s easy for me as a creative type to push back against AI because, at this point, the bots, while they might work faster, don’t produce better work. But there are jobs where that argument is harder to make, whether because the job actually can be done well by AI or because the people who decide how the job gets done believe that to be true and have spent, say, $25 million on AI, so it must be true.
Just as I love the sometimes-painful process of writing and editing, I am sure that many people reading this are at this very moment being forced to try to train AI to take over a task they enjoy doing and are quite good at, a task that perhaps is one of the reasons why they got into their career in the first place.
AI can’t do my job as well as I can, so it’s pointless to bring up the fact that it can do it faster. But I think this is what proponents of indiscriminate use of AI must believe: Why do something well when you can do it fast, and if a machine can do something faster than you, why should you do it at all?
I won’t say the name of the company, but the other day I got a sales pitch in my inbox from a vendor who wanted to show CCI how we could automate publishing. They were sure to say editors would make the final call, of course, but they seemed to believe the thing we should care most about is speed, which is the rallying cry of many of those pushing AI the loudest.
But speed isn’t the only metric, and it’s not the one that matters most to us. Like a student who hurries through a test, marking the middle option on questions without even reading them, AI generating a bad headline quickly isn’t a success. A task done sloppily is an incomplete one, and so far, most of what I see with AI is slop.
CCI will continue to experiment with how we can use AI, particularly for non-creative tasks, but it’s safe to say I’ll spend much less time trying to trick a large language model powered by a water-guzzling data center into appearing human.
Jennifer L. Gaskin is editorial director of Corporate Compliance Insights. A newsroom-forged journalist, she began her career in community newspapers. Her first assignment was covering a county council meeting where the main agenda item was whether the clerk's office needed a new printer (it did). Starting with her early days at small local papers, Jennifer has worked as a reporter, photographer, copy editor, page designer, manager and more. She joined the staff of Corporate Compliance Insights in 2021. 
