AI Writing Sucks
Photo by Andrea De Santis on Unsplash
While artificial intelligence may very well evolve to overtake humanity, we should be grateful that we presently live in that sweet spot of its infancy when it’s harmless and helpful, like a Roomba. Content is consumed faster than ever before, and if the crush of influencers and Facebook how-to videos has proven anything, it’s that said content doesn’t necessarily have to be nutritious, it just has to have texture. Time-saving generation tools like AI copywriting are growing in popularity, but can they be trusted as a reliable source of content in the long term? I sampled a few of the top AI writing programs to see how they stacked up, and what I found was eye-opening.
Jarvis and Writesonic are two of the most popular AI-based content creation tools available. Both ask users to pay monthly fees to access their program (with limits on word count that can be extended with higher payment tiers) and both more or less work the same way. They have templates for creating blogs, search engine optimized headlines and titles, advertisement copy, technical content, and all manner of formats for which words would be useful.
If you’re tech-savvy, both companies feature programmable customization options users can tinker with to tell the algorithm how to make the words it generates flow along more specific subject trenches. Essentially, a user plugs in a desired title or subject line and provides a summary of the points they’d like the AI to hit in its creation of a body of text. You can toggle the length, tell it to incorporate certain keywords, and when everything is lined up--hit the blast-off button. Both Jarvis and Writesonic work quickly, one of their selling points, and I was legitimately impressed with the way both programs handled the basics like topic sentences, transitions, and stringing words together in general in a matter of seconds. Upon closer inspection, however, thoughts often trailed off into incomplete sentiments, introductions of new topics were a little jarring, and the more challenging keywords I wanted woven in were ham-fistedly shoehorned as their own standalone sentences with no connective tissue. I also noticed the repetition of phrases and idioms that were clearly triggered by the words I was typing. For example, when I asked Jarvis to write me a sales pitch for a new line of socks I’d invented, it kept repeating the phrase, “I have socks older than you are.” It gave me the impression that certain words triggered go-to sentences without consideration for the topic. Overall, I liked how intuitive these programs were at identifying “copy that sounds like copy” and peppering the subjects into these cookie-cutter formats, but there wasn’t a lot of original thought to be found.
ArticleForge was my next stop on this journey. It boasted a much higher monthly ask, but its sticking point was its quality of content. I tested the “blog article creation” tool, which functioned similarly to Jarvis and Writesonic—I provided a keyword followed by sub-keywords that I wanted the article to hit along the way and told it to “go.” While the other algorithms took seconds to blurt out content, ArticleForge took a solid minute to generate an entire article for me, which is an eternity in the eCommerce world, but I let it do its thing and I wasn’t entirely disappointed. ArticleForge seemed to work by scouring Google for similar articles and topic phrases to then assemble a 250-word block that could stand next to any other blog post and look legitimate.
For this exercise, I asked ArticleForge to provide a sales pitch for why I should be hired to rewrite your screenplay. It started strong, stating that being a “Hollywood screenwriter” took time and dedication and that you needed to look at other writers’ stories to learn how to do it. Then it fell off the rails by providing not-so-true facts about Drew Barrymore (not a writer) and Aaron Sorkin (who didn’t write a movie the AI claimed he had written). It concluded that script rewrites, across the board, cost $100 and could be fixed with one sentence. As someone who’s spent almost his entire life studying this craft, I couldn’t agree less!
ArticleForge, to me, felt like an alien wearing someone’s skin, jerking its limbs erratically and flapping its mouth to make the sounds a human makes while remaining woefully unaware of the meaning behind them. Even the “better” sales pitch parts of my AI-generated text felt rudimentary, like an aggregate of thousands of hours of focus group research for how to sound the most like a timeshare seminar. Useful for generating quick ideas that a human being could tweak and expound upon, sure, but I wouldn’t trust it to lead my brand’s marketing campaign.
An amusing aside, in my research I stumbled upon a company called “Books By AI.” Their catchphrase was “tired of books written by authors?” What I found was twenty or so AI-generated novellas, complete with procedurally-generated covers and testimonials from robot critics with faces generated by thispersondoesnotexist.com (a website that cobbles together realistic faces of nonexistent people by using a proprietary algorithm). If you want to feel like you’re having a stroke, please check out booksby.ai and pick up one of their paperback masterpieces.
Grammarly, a program aimed at correcting grammatical errors and fixing syntax, is one of the more useful tools I found in my travels. Its free software (all you need is an email address) allows users to plug in up to 15,000 words (no cap!) and receive corrective instruction. It’s handy for fixing tenses, finding double-up words, and doing an overall sweep of a large body of text, but it’s also not the best at interpreting context. Beyond that, a lot of their higher functions (tone interpretation, sentence flow, etc.) were locked behind a paywall. I was curious to see what Grammarly thought of the nonsensical dust cover summary of Breath Catcher by Tom Griffitzer, one of Books By AI’s featured authors, and found that the algorithm grading the other algorithm gave it a perfect score on clarity, delivery, and engagement. Just to give an idea of the content, here’s an excerpt:
“They’re too long ago, a man wants to leave her laughs.” Stunning.
AI writing is a cute gimmick and arguably pretty useful for generating large blocks of text no one’s going to read--technical manuals where creativity and independent thought isn’t a requirement, for example. I could also see it being useful for large-scale topic generation and kindling for bigger ideas that a writer could finesse and run to the finish line, but for consumers looking to incorporate an impactful body of text with heart, the best solution is still a living, breathing human being. Brand messaging, after all, hinges on an all-important story, and while the robots might take everything from us one day, they can’t take our innate ability as storytellers.
