Fear no chatbot

3/1/2023

Artificial intelligence is entering the writing space. But for branding and marketing, chatbots aren’t necessarily the enemy.

 

People can’t stop talking about a chatbot. 

Beginning with its public-facing launch in December 2022, OpenAI’s ChatGPT has commanded headlines across major news outlets and every platform with room for a think piece about this latest feat of artificial intelligence. 

Will ChatGPT destroy or transform education? Will it render Google Search obsolete, or just break it altogether? Can ChatGPT become irrationally self-aware or be weaponized against democracy? And is it, per one online discussion, merely “Mansplaining as a Service?

Fascinating conjecture, to be sure. But how about a more pragmatic question? 

What might this technology, with its ability to communicate in plain, human language, mean for marketing and branding firms, and for the very talented, very human people who create the work? 

Here at Ologie, we took ChatGPT for a spin. And yes, we have thoughts.

 

A look under the hood.

Quite simply, ChatGPT operates like an artificial brain. It’s a closed neural network that exists apart from the internet, and has been trained to sort, rank, and connect enormous amounts of information. Register an account, ask it questions through a simple chat interface, and if ChatGPT deems your question answerable (it balks at subjective, illegal, and vulgar queries), it will generate a coherent, plausible-sounding, but not necessarily accurate reply. 

With each prompt — and there have been literally hundreds of billions of prompts so far — this generative pretrained transformer (the “GPT” in ChatGPT) gets smarter. Whether it will ever know everything — or even “know” anything at all — is a heavy question. 

But one question weighed on our minds: Is ChatGPT a threat to our beloved human writers? 

Short answer: No.

Troubling, longer answer: No time soon. And probably never. Probably.

 

Putting ChatGPT to work.

First things first: ChatGPT is a thrilling time waster. 

If you can think of a zany prompt that avoids its internal filters, ChatGPT will happily oblige. (It refuses to write a haiku about boogers, but it will compose one about mucus: Snotty and thick drops / Slime trails on nostrils, cure / Cold with tea and rest). 

Ask it to compose a help-wanted ad in the style of Jack Kerouac (We seek the free-spirited, the nonconformists, the wild at heart) or a 1920s clubman (Gather ’round, kiddos! We’re lookin’ for a few good eggs to join our little flock!) — and it will do so, but always, oddly, in rhyming couplets, from which this author has mercifully spared you.

Fun and games aside, though, how might ChatGPT fare in branding and marketing? Can it handle the toughest — and perhaps foolhardiest — task in the biz: writing a higher ed tagline? 

To put ChatGPT to this test, we began with a general prompt: “What are some taglines for a small, Northeastern liberal arts college?”

It promptly generated generic results such as:

  • Discover your potential.
  • Where traditions thrive. 
  • Inspiring excellence.

We adjusted the prompt to include “…with an emphasis on experiential learning?” and it turned out results such as:

  • Experience. Learn. Grow. 
  • Discover your potential through experience. 
  • Where the classroom meets the world.

We felt like we were getting somewhere. We added “…with over 200 years of history?” to the previous prompt, and got:

  • Experience traditions. Embrace progress. 
  • A legacy of learning through experience.
  • A college with roots as deep as its reach.

Better. Next, we added “…of fighting for social justice?” and got back:

  • A legacy of justice, a future of change. 
  • Learn. Lead. Make a difference. 
  • Building a just future, one student at a time.

Then we piled on “…and athletic excellence?”

  • Experience. Compete. Champion social justice. 
  • Experience-driven education, championing athletics and social justice. 
  • Champions on the field, change-makers in the community.

Then, “…and research?”

  • Inquiry and action for a better world.
  • Experience-driven education, championing research, athletics, and social justice.
  • Education at the intersection of knowledge, athletics, and justice.

Then, “…and a welcoming community?”

  • Experience-driven education, championing athletics, research, and social justice in a welcoming community.
  • Building bridges, breaking barriers.
  • Education, athletics, research, community: A path to empowerment.

Then, “…in a beautiful, dynamic location?”

  • Education, athletics, research, community, surrounded by natural beauty.
  • Thriving in a community rooted in nature and justice.
  • A legacy of community and beauty, a future of justice and opportunity.

And finally, “…and the highest standards of academic rigor?”

  • Education at the intersection of nature, community, and rigor.
  • Where excellence meets experience.
  • Education, athletics, research, community, nature, rigor: A path to growth and empowerment.

 

So in case you lost track, the final prompt was, in total:

“What are some taglines for a small Northeastern liberal arts college with an emphasis on experiential learning with over 200 years of history of fighting for social justice, athletic excellence, research with a welcoming community in a beautiful, dynamic location and the highest standards of academic rigor?”

 

Let it be said: ChatGPT is a trouper. 

With all the responses it generated (curated by the author), it never stopped checking the number of boxes we presented, while gamely offering up options that tried not to say everything. Which raises the point that ChatGPT never raised a flag: it never challenged the absurd number of messaging points within the prompt. However, when pushed to generate dozens and dozens of taglines with the same prompt, it did warn that it would deliver less distinct and more repetitive results. That’s about as close to self-awareness as ChatGPT gets (one would hope), but knowing when to say when is a virtue.

It’s worth noting that ChatGPT has a surprising handle on ear-pleasing devices — alliteration, parallel structure, iambic pentameter, and other rhetorical catnip. It also demonstrates a broad and tasteful vocabulary, not merely strip-mining the thesaurus, but panning for relative and appropriate gold. All it needs is the guidance of a writer with a modicum of taste. 

 

The good, the bad, and the useful.

As a technology that’s constantly iterating on itself, the ChatGPT of our experiment will necessarily be different from the ChatGPT you know. But we can make some generally durable observations. 

The good: 

ChatGPT is an unbreakable writing intern. It will tirelessly generate lists of words along specific parameters like length, tone, letter combinations, you name it — almost like a smarter, automated OneLook. It can do basic research, generate thoughtful filler copy, and (as we’ve seen) spin endless variations for headlines and taglines. It could just as easily crank out sample language for A/B testing or for creative workshops, toggling the tone of messages from conventional to unorthodox. 

ChatGPT trains us to ask better questions. Working with a chatbot — feeding it prompts, assessing results, narrowing criteria — can be an exercise in asking better questions, not unlike how parents of toddlers learn to navigate the never-ending litany of “But why?” ChatGPT asks nothing of us, but will only answer what we ask of it. By design, the chatbot’s responses improve over time; therefore, so must our queries.

The bad:

ChatGPT is a hot beta. And that, again, is by design. It’s still learning, and the more we use it, the smarter and more useful it will become. But for the foreseeable future, it will remain a sophisticated tool that’s of little use without an artisan’s skill, experience, and instinct.

ChatGPT still lives in the uncanny valley. Technically, it can squeak by the Turing Test. But while it can synthesize information and speak something that resembles plain English, it still needs us to navigate the gullies and byways of storytelling — as indicated by its inability to write jokes. ChatGPT can assemble odd things with the shape and rhythm of jokes, but with none of the soul or surprise of actual human humor.

The useful:

ChatGPT is a total (awesome) tool. At its most useful, ChatGPT exists within a positive feedback loop with its user, encouraging something like healthy dialogue — the more thoughtful the chat, the more productive the session. Additionally, ChatGPT can be helpful at alleviating some of the arduous, exploratory steps in the writing process, providing more jumping-off points, earlier on. As long as we remain cognizant of where the bot’s capabilities end and our writing and storytelling abilities begin, ChatGPT can be a force for good. Or at least better.

ChatGPT can produce the ultimate first draft. Guided by a clear vision and a curator’s sensibilities, the chatbot can produce high-quality raw material for a writer to work and refine. On its own, however, ChatPGT is pure, uncut Dunning-Kruger effect — it not only doesn’t know what it doesn’t know, but it also can’t help putting it all out there. (Unless, of course, it has to do with building bombs, which musical genre is definitively the worst, or haikus about boogers.) 

 

So what about us?

As writers in branding and marketing, we’re safe from ChatGPT for now, and probably better off for it. The chatbot is powerless without a curator. Its AI is responsive, but only as good as your prompt. 

And probably most important: without a human involved, ChatGPT is hopelessly out of its depth. It only sees what it’s shown and gets what it’s been told. It can extrapolate, but it can never assume. It’s far from perfect, but pre-corrects for the slip-ups that reveal truths. It’s oblivious to its lack of humanity, for better and for worse.

ChatGPT needs us more than we need it. As a tool, it’s exactly as good, bad, or useful as we want it to be. But if we use it well, it can make us better at what we’re best at: telling authentic stories that move actual people. And if an AI chatbot helps us get there, all the better.