Within two months of its launch, Open AI’s ChatGPT-3.5 hit 100 million monthly active users — shattering the record for fastest-growing consumer application and becoming the talk of every classroom, newsroom and trading room. And the buzz has only continued with the release of an updated model – GPT-4 – just a few months later.
This represents a major shift for content marketing. Before ChatGPT, artificial intelligence never possessed the ability to produce such natural-sounding content so quickly. Now, with the release of this advanced version, many companies are left wondering what this means for their marketing communications programs.
Naturally, at INK, where we focus on content marketing, we’ve long been exploring the impact of AI on integrated communications strategies. So, over the past few weeks, we decided to put the virtual chatbot from OpenAI to the test to see just how good it is and how to best incorporate it into our workflows. What we found speaks to why this technology has been met with both excitement and concern.
In a nutshell: ChatGPT packs a good punch, but it still has some ways to go. As such, we view it as more of a helpful copy assistant rather than a full-out replacement for human writers. That is, at least for the foreseeable future.
SO WHAT EXACTLY IS CHATGPT, AND HOW DOES IT WORK?
ChatGPT harnesses machine learning and natural-language processing to instantly generate responses to human-led prompts.
More technically speaking, the software’s generative, pre-trained transformer language model sifts through massive datasets that have been pulled from the internet. Then, the system’s neural network (which predicts what word will come after the previous in the response) crafts the best average answer. The result is a response that sounds surprisingly conversational, because it’s an aggregate of content written by humans. (Hence all the hubbub.)
GPT-4, while only currently available to ChatGPT Plus users and developers, builds on this core function with improved and added capabilities. Some of the biggest new features include:
- Now accepts both text and image inputs – Whereas GPT-3.5 was only able to accept text.
- Capable of processing 8x the words of GPT-3.5 – OpenAI states that GPT-4 can generate responses up to 25,000 words in length, a big jump from the previous 3,000 word limit.
- Can develop a website from an image input – During OpenAI’s GPT-4 Developer Livestream, it demonstrated how the updated model could take a hand-drawn sketch of a website and deliver a functioning website and even populate it with relevant content.
Many other tech firms have followed suit. In addition to Microsoft’s ChatGPT-backed Bing chatbot, Google introduced its own AI technology named Bard, and we’ve seen a surge in third-party tools and extensions built to work with these AI platforms. But with a first-mover advantage, both in its initial launch and release of an update, ChatGPT is currently getting most of the oxygen.
WHY DOES CHATGPT MATTER TO CONTENT MARKETING?
Unlike search engines that provide content marketers with endless links to sift through as they begin looking for information to inform their writing, ChatGPT skips to producing original-seeming content on its own. We’re talking about everything from poetry to research papers to screenplays, along with outlines, editorial calendars and SEO optimized content.
At INK, that has made us wonder: What does bot-generated content mean for the future of content marketing? And how might we adapt to continue delivering content that drives results for our clients?
Well, who better to ask than ChatGPT itself …
Phew. For now, we can all breathe a sigh of relief.
3 REASONS WHY HUMANS ARE HERE TO STAY
Still, we wanted to dig a little deeper — understanding what other limitations the chatbot might reveal about itself. In the process, we gathered some key reasons why human-produced content is as valuable as ever.
1. Humans can have an original thought
Because ChatGPT is a neutral AI language model — meaning it repurposes content from around the web and doesn’t generate totally unique opinions — the program is a far cry from what would be considered genuine thought leadership.
Put another way, if you’re developing content that is truly forward-thinking (as all good thought leadership and original research should be), you need writers and editors who understand your industry and business deeply and can weave in fresh insights.
Take this comparison of a LinkedIn post written by our CEO Starr Million Baker, next to one written by ChatGPT. Even when we tried to feed the AI a more nuanced prompt, ChatGPT’s content fell short.
Takeaway: Don’t sell yourself short with an overreliance on AI. Our world needs more independent and original thinkers.
2. Humans have a certain je ne sais quoi
One of the pillars of developing quality content is writing with a distinct voice, free of jargon and clichés. In doing so, you can connect more personally with your audience. And in a media landscape buzzing with commodity content, writing solely with AI just adds to the noise.
Zooming in, the details make a big difference. In longer-form content, for example, ChatGPT tends to use throwaway phrases like “in conclusion.” Meanwhile, in shorter social posts, the tool fakes human sentiment with statements like “I’m excited,” which just sound disingenuous.
Zooming out, we wondered: Could ChatGPT actually capture a unique human voice, if given the right prompt? We put the tool to the test, asking for a blurb about the Inflation Reduction Act written in the voice of Matthew McConaughey. Below is what we got back.
* Cringe *
The less-than-impressive result — with an “alright, alright, alright” lopped on at the end — represents just how limited AI can be in capturing the full range of an individual’s personality.
3. Humans build trust by telling the truth
While ChatGPT is trained on data sourced from the internet, you should know that the platform is not actually connected to the web in real time. Even with the introduction of GPT-4, the chatbot’s most current data set (as of March 2023) is from September 2021. In other words, any stats or information the chatbot gives you are outdated by about one and a half years.
Also, if you think back to research 101, you know to cite your sources. But more often than not, bot-generated answers have no hyperlinks or sourcing. We’ve found that you have to ask the program to cite itself if you want to know where the information is coming from, which presents a Catch-22.
ChatGPT is also prone to factual errors. Although GPT-4 brought advanced capabilities, it still has similar limitations as GPT-3.5. In particular, it can sometimes “hallucinate” facts and makes reasoning mistakes.
In an interview with WTTW, Kristian Hammond, director of the Center for Advancing Safety in Machine Intelligence at Northwestern University, notes that ChatGPT is simply a language model; that is, it does not know truth from falsehood. Hammond points out that the program pulls averages across sources, so the information it spits out is often accurate only from a statistical — but not specific — point of view.
To illustrate this point, Hammond asked ChatGPT to write a blurb about a master’s program in AI at Northwestern. The result was copy that mentioned a two-year program, because most master’s degrees take two years to complete. The problem: The Northwestern program lasts just 15 months.
Blindly trusting AI content means you’re rolling the dice with potential misinformation — a game marketers can’t afford to play, especially when trust is so critical to building and nurturing relationships. Our world is already plagued by misinformation, particularly on social media, so the responsibility for verification is all the more paramount.
HOW TO ADD CHATGPT TO YOUR CONTENT MARKETING TOOLKIT
Despite its limitations in producing complete pieces of content, we don’t have to view ChatGPT and other AI tools in an all-or-nothing light. Just as mathematicians use calculators and data scientists use Excel, we marketers can use AI to make our content programs scale faster. When used correctly, these tools can also set our content outputs up for greater success.
Learn what not to write
If you want to break through the noise, you can use ChatGPT to determine what’s already been said on a certain topic. Think about it like this: If a chatbot could produce a piece of content by searching through existing information, why would you want to produce something so similar? When starting any piece of content — web copy, social media posts, blog articles, or bylines — ask ChatGPT what it would write. Then zig where the tool zags.
Kick-start your content
When writing about complex topics like B2B tech and energy, expertise comes with time. For content marketers still developing their knowledge, or for anyone who wants a jumpstart on their writing, we’ve found AI to be a helpful brainstorming partner.
For example, here’s how we used ChatGPT to write the blog you’re reading now.
Having a framework relieves some pressure from the task at hand. That’s not to say, however, that we then just kicked back and called it a day. A lot of research, ideation, buildout, editing, and fact-checking still went into fully fleshing out this article. (And we truly hope you can tell it was not written by a bot.)
Structure your articles for SEO success
Perhaps because ChatGPT is pulling from content around the web, we’ve also observed that it might have a useful function from an SEO perspective.
For one, you can use the tool to help with keyword research. Ask for related keywords to the topic you’re writing about, and the chatbot will suggest a helpful list.
Or, if you’re using the tool for starter copy, you might get an article that is well structured for performance in search, because it’s drawing from a ream of content driven by audience intent. We’ve seen this firsthand, with a top-performing article that was aided in part by ChatGPT — albeit with a ton of human oversight and reworking, not to mention a heavy-handed edit.
Another idea: You can use ChatGPT for something as simple as optimizing a title. Just command the tool to “write an optimized title on (topic X),” and you’ll likely get something that’s straightforward and well written. Again, because the program scours the internet for information, it has a good pulse on how people think and what sort of top-level answers they’re expecting.
LEAN ON AI; DON’T DEPEND ON IT
As technology evolves, we should try to see the opportunity before us instead of just the potential threat. Sure, the peril of robots taking over continues to loom. But in the end, we can’t discard the importance of a human touch — and just how hard it is to replace. The union of AI and human ingenuity is evolving fast. And frankly, it’s exciting to be a part of, especially in the creative space, where technology empowers us with even more ways to tell new stories to the world.