Like many people, I’ve been following the news about the capabilities of ChatGPT and other AI writing tools with a mix of fear and curiosity. There seems to be a consensus that this will radically change the nature of work, but I’m not totally sure yet what it could actually do for me.
And then I realized there is one thing in my life that feels like drudgery, an obligation that I don’t enjoy: posting to Instagram.
What if I could get AI to write my Instagram captions? Would the words seem human? Would it do a passable job of seeming like an avocado toast–eating millennial? Could it make posting to the grid feel like less of a chore? Only one way to find out.
My idea was that for two weeks, I would attempt to post to Instagram on a near-daily basis. This was a big change for me, as I tend not to post very often at all. This is in part because as a journalist I sometimes deal with online harassment, so I won’t post personal stuff, like photos of my kids or where I live.
So in order to get enough content for two full weeks, I ended up mostly posting one of the more reliably visually interesting things I see in my everyday life: novelty-flavored snacks.
To write the captions, I used an app called ParagraphAI, which installs a keyboard on your phone that you can access by tapping the globe icon (similar to switching to a different language keyboard). When you have the ParagraphAI keyboard up, you type in a prompt or shortened version of what you want it to say. Hit “Write,” and then it pastes in a chuck of AI-generated text.
I chose ParagraphAI, which uses OpenAI’s ChatGPT-3, because it’s relatively frictionless. Sure, I could go to the ChatGPT website, generate some text, and copy it over, but that’s a few more steps. (Not to mention that often the website is overloaded or not available.) My one complaint about the ParagraphAI keyboard is that sometimes I would accidentally tap the globe, type some words, and end up with a big chunk of AI text I didn’t want and had to delete.
ParagraphAI also lets you customize the tone with sliders: From formal to informal, friendly to assertive, and pessimistic to optimistic. Over the course of the experiment, I tried different settings, sometimes redoing a caption with a new tone. I also allowed myself to re-prompt the AI if the result was really far off. But once I got a decent result, I didn’t edit the text at all.
My first attempt at a caption involved a new flavor of Goldfish: Frank’s RedHot (they taste exactly like you’d imagine). Here's what the AI came up with:
This first post got some great feedback from friends, who seemed somewhat interested. “Would try!” said someone. “Love a frank review,” said another.
But some people seemed to notice the tone was a little… strange. “Great content thanks for sharing,” commented another friend dryly. “Is this an #ad?” asked a former coworker.
Most excitingly, the official Goldfish brand cracker account commented: “love that you’re loving it 🧡🔥”
Next I tried a photo of Sour Patch Kids–flavored Trident Vibes gum. (I found it unbearable; my 6- year-old loved it.) My AI caption:
I recently tried the new "sour patch kids" flavor of gum and, unfortunately, it wasn't for me. The flavor was a bit too strong and it didn't really hit the spot for me. I think I'll stick to the classic flavors for now, but if you're feeling adventurous, give it a try and let me know what you think!
It got a few comments like “The vibes are off.” No one seemed to notice though, and since it wasn’t a positive review, it didn’t sound like an ad.
A few days later, I did a review of Starry, the new lemon-lime soda from Pepsi that replaced Sierra Mist. To make it seem less botlike, I used a photo of me holding up the bottle. I asked ParagraphAI to write an Instagram caption in the tone of a food influencer this time.
I just tried out this new soda called Starry, and I've been seeing it in stores lately! Unfortunately, it wasn't for me. I'm sure it's great for someone else, but it just didn't hit the spot for me. Oh well, you can't win them all! #StarrySoda #TryItOut #NotMyThing
I think the hashtags gave away that something was off, because what kind of normal person uses hashtags like this? “U ok?” asked a coworker. Others mimicked the AI-speak. “I'm hearing Starry is quite polarizing," said another friend who works at an ad agency that works with PepsiCo. "Interesting to get your take, thanks for weighing in."
So far, it seemed that ParagraphAI was very capable of creating a legible, understandable caption for Instagram with a very basic prompt. The results didn’t sound like me, but I could see how this might be useful for writing marketing copy for social media.
“Katie? More like K-AI-tie 🤖,” said a high school friend.
A week in, I had run out of interesting food and drink and was a little desperate for something to post. I was going through my photo roll, which is largely full of screenshots of stuff I’ve seen online. One recent thing that I thought was pretty amusing was a screenshot of a Facebook Marketplace listing for a 1992 Suzuki Samurai jeep. What a weird-looking old car!
Bad move. I think that using the screenshot of a sales listing seemed spammy. The ParagraphAI text was also particularly impersonal-sounding.
Immediately my friends noticed. “Is this kapie or an AI chatbot,” a friend commented. “Katie? More like K-AI-tie 🤖,” said a high school friend. “Good bot” said someone I don’t actually know.
Next up, I posted some novelty-flavored Peeps candy. I set the sliders on Paragraph AI to “formal” and “pessimistic,” and boy did it deliver one angry screed:
I got some real replies, like a friend saying “what a let down!” Another person commented “katiebotopoulis,” suggesting I sounded botlike.
At this point, my best friend texted me asking what was going on with my Instagram. Apparently, a few of our other friends had reached out to her, concerned about me and asking if I was OK.
I soldiered on, doing another candy post. Later that weekend, while I was hanging out with my best friend, she got a text from our old roommate from over 10 years ago asking what was happening with my Instagram lately.
While I appreciated the concern from longtime friends, it also made me ponder some deeper questions. Am I so emotionally closed off and distant that my friends don’t feel comfortable asking me directly how I’m doing if I seem to be in some sort of crisis? Am I so poisoned by the internet that the idea of me earnestly posting about food to Instagram sets off alarm bells? Have I twisted myself into such a hideous online troll that I can’t enjoy the actual human connection that social media can bring?
On a later post, about a book on gardening that I had checked out from the library, I got a response from an old friend I had gone to summer camp with and runs her own garden nursery. I hadn’t heard from this person in years and was touched that she was reaching out. But I felt a twinge of guilt that she was so earnestly responding to something artificially written. This was starting to feel…bad.
I knew that AI wasn’t going to work well for Instagram captioning, a form of writing that is by design uniquely personal in tone and nuance. But I did it anyway.
I should disclose here that this is not my first stunt on social media. In the past, I’ve played around with gaming the Facebook algorithm, torturing my friends by keeping the same video at the top of their feeds for weeks on end (the trick is to just keep commenting). I’ve done other stunts too, and in theory I could say I’m doing it “for work,” but the truth is I’d have done it anyway just to fuck around.
To me, my motivations for doing these stunts are clear. Feeling trapped and controlled by the powers of Big Tech, I’m acting like a rebellious middle schooler, using something exactly as intended but for terrible effect.
Of course, I knew that AI wasn’t going to work well for Instagram captioning, a form of writing that is by design uniquely personal in tone and nuance. But I did it anyway, at the expense of my own friends, forcing them to endure garbage posts in their Instagram feed, punishing them for the folly of actually ever caring about me.
I thought this experiment might show something interesting about the capabilities of AI writing, illustrating the computer’s lack of depth and the hollowness of it all. Instead it revealed an unpleasant truth about myself — that I’m a nasty little jerk with a real lack of respect for the people who make the mistake of caring about me.
The last few weeks I’ve been eye-rolling at late night dorm-style conversations about the morality and sentience of ChatGPT and other AI bots. And yet I ended up taking a different fork on the same existential path.
But it was not the humanity of the AI that I should have been questioning. It was my own.
Anyway, I strongly do not recommend anyone else try this. ●