What Brand Marketers Need to Know about ChatGPT
Asking ChatGPT what it can do for marketers, and dissecting its answer
In our freshly published Outlook 2023 report, we highlighted the rise of synthetic media, aka media content generated by AI, as an emerging trend that has the potential to revolutionize digital content creation, reshape the landscape of the creator economy, and ultimately redefine our relationship to digital media.
Among all the “Generative AI” tools (a popular term for the software that generates synthetic media) available to the public at the moment, ChatGPT is by far the most well-known. Despite all the controversies and school bans (or perhaps partly because of them), ChatGPT has broken into mainstream consciousness, as evidenced by the countless news articles and memes about it. ChatGPT is estimated to have reached 100 million monthly active users in January, according to a new UBS study cited by Reuters, just two months after its public launch, making it the fastest-growing consumer application in history. Not since the debut of the first iPhones back in 2007 has a new tech invention captured the public imagination in such an undeniable fashion.
Created by OpenAI based on the company’s GPT-3 family of large language models, ChatGPT works like a regular chatbot, but it is unlike any other chatbot in terms of the sophistication in the answers it provides, as well as the vast range of topics and domains about which it appears to be knowledgeable. Instead of a chatbot, it’d be perhaps more apt to label it the first mainstream “text-generating AI tool.” People have been asking it to write school essays, solve complicated math equations, compose work emails, develop apps, create new workout plans, and even plan dinner parties by generating the themes and recipes — but that’s all just the tip of the iceberg.
The Obvious Answers
Needless to say, we’re only at the beginning of this upheaval, and as an active participant in the attention economy, brand marketers should be keeping a close eye on the development of synthetic media and its myriad of potential applications.
So naturally, here at the Lab, we had to ask ChatGPT, “what can you do for marketers?” And here’s its answer:
At first glance, ChatGPT has made quite a convincing sales pitch for itself to brands and ad agencies. The first two use cases — generating personalized content and handling customer service — are the low-hanging fruits whose applications are easy to imagine, and we are likely to see real-life implementations very soon. (The rest of the list amounts to enterprise use cases on the backend, and does not involve consumer-facing brand marketing.)
Content creation is an obvious use case — Ryan Reynolds already used ChatGPT to write an ad copy for his mobile carrier company and read that ad copy aloud in a viral YouTube video. How long before ad copy and social media posts are actually written by AI tools like ChatGPT? It could certainly be a valuable ideation tool to produce a variety of options for decision makers to choose from.
With the rise of programmatic buying and dynamic creative, it stands to reason that tools like ChatGPT would be well-suited to supply digital marketers with endless iterations of ad copy and social media captions that are tailored to each audience segment, no matter how niche they are. If brands’ first-party CRM data is plugged in, then perhaps this type of automated personalization could theoretically be narrowed down to an individualized level at scale.
As for automating customer service, chatbots have been deployed for handling basic tasks in customer service for years. But, what’s exciting about using ChatGPT is that it may finally enable the kind of naturalistic, “can’t tell if it’s a human or a bot” conversations that bring a “personal” touch to automated customer service. Many call center workers today already handle incoming inquiries by following decision flowcharts, so why not let an advanced chatbot take care of that, and free up workers’ time to focus on more complicated customer service questions??
The Tilda Swinton Test & AI Hallucinations
That said, just because those use cases are well-suited to ChatGPT doesn’t mean its implementation will be without challenges or growing pains. While ChatGPT has the capability to generate answers that look professional and correct, there is no guarantee of their factual accuracy or whether they are tainted by discriminatory AI biases. Most machine learning models today tend to have a “hallucination problem” of perceiving things that are not there. Wonder what that means? Take this seemingly random exchange I had with ChatGPT for example:
Despite this seemingly perfect list that ChatGPT confidently responds with, any real Tilda Swinton fan would quickly spot two glaring errors:
- One, Tilda’s performance in Doctor Strange was not widely celebrated, but rather much maligned for the yellowface controversy her casting caused; its inclusion by ChatGPT here seems to reflect the general popularity of Marvel films more than her performance.
- Second, Tilda actually played three characters, not two, in the 2018 remake of Suspiria. She chose to credit one of her roles under a male pseudonym, a fact that was widely publicized upon the movie’s release but somehow was neglected by ChatGPT’s database.
In both cases, the database that ChatGPT is being trained on convinced itself of things that are categorically untrue. Of course, this is simply an open-ended question with no real stakes. But, when hundreds of thousands of marketing dollars are on the line, not to mention the potential legal and reputation issues that any error could cause, it is clear that ChatGPT (and other generative AI tools will be used for marketing purposes) will require human supervision for the foreseeable future.
In other words, the new “don’t trust everything you read on the internet” is “don’t trust everything ChatGPT tells you.”
Update March 15, 2022 — after Microsoft confirmed today that Bing is running on the advanced GPT-4 model, I went and asked Bing the same question about Tilda’s best performances, and received a much better, more accurate answer:
Moreover, for Generative AI tools like ChatGPT to be deployed at scale for marketing content creation and customer service, brands will need to make sure they are properly trained on their up-to-date customer dataset, as well as industry and company guidelines before deployment.
It is important to remember that, as impressively competent as ChatGPT seems, it is far from the type of artificial general intelligence (AGI) with true intelligence capacity to understand or learn any intellectual task that a human being can. All it does right now is remix the content it finds in its database and assemble it into an answer that looks perfect in style. It could beat a lot of chatbots in use today to provide more natural conversations and well-devised answers, but it will not replace human customer service reps or content creators any time soon.
What ChatGPT Is Not Telling Us
Looking beyond the obvious answers, I found it rather fascinating that ChatGPT’s answer does not include anything about its capability to change how people search for information online. Given how much Google’s search ads and SEO practices occupy the minds of every digital marketer, it is rather remarkable that it has neglected to put “product discovery or recommendation” on its list.
Ever since ChatGPT broke into the mainstream, there has been speculation that it is going to change how we search online. 40% of Gen Z already forgo keyword-based search on Google in favor of searching on TikTok and Instagram instead. Now there is this know-it-all AI, ready to put together an answer to just about any question you may have, all without forcing you to parse through ads and click through ten blue links before finding the answer you seek.
After all, the New York Times didn’t call ChatGPT a “Code Red for Google’s search business” for no reason.
Therefore it is not surprising that Microsoft is reportedly working to incorporate ChatGPT into its search engine Bing in the coming weeks, as part of a multi-billion-dollar investment in OpenAI, which would create a more conversational user experience that could make Bing more competitive with Google.
Of course, Google is actively working on its counter-moves as well. A few weeks ago, the New York Times reported that Google founders Larry Page and Sergey Brin have discussed their company’s response to ChatGPT, with plans to launch over 20 AI products this year, including a demo of its own search chatbot.
It’s not hard to imagine how product discovery may change if ChatGPT’s Q&A-style interface were to become a dominant way for online search. After all, you can already ask it to recommend a product today:
All that is left to do is to feed it the latest content from Wirecutter or The Strategist, add some images and clickable links, and voilà, a general-purpose AI shopping assistant is born.
Unsurprisingly, this works for product comparison as well:
Of course, OpenAI is incapable of giving a single answer that is truly original — everything in its answers come from an existing piece of writing in the vast database the language model is being trained on. In a sense, it is the ultimate digital hivemind, synthesizing all the opinions floating around online and packaging them into an agreeable, mainstream consensus. In the age of conversational Q&A search, the SEO will not be just about keywords, but also about the volume of mentions of your brand and product in the database, for the same reason that ChatGPT listed Doctor Strange as one of Tilda Swinton’s best performances.
Of course, different industries will have different use cases of ChatGPT to explore, but alas, you wouldn’t get that from asking it. Compared the following nearly-identical replies from ChatGPT to the two similarly worded questions:
As you can see, the answers are led by four generic marketing use cases with the keywords swapped out, and, in the case of promoting auto brands, followed by an additional nonsensical answer of assisting virtual test drives, which ChatGPT has no way of fulfilling in its current text-only format. Once again, the answers expose the inner workings of its language learning models that spit out safe, pat answers masquerading as intelligence.
That said, that doesn’t mean there won’t be new, unique applications of ChatGPT in various industries. In healthcare, for example, it may be worth looking into how tools like ChatGPT may help talk therapists handle virtual sessions, considering that some users are already turning to it for mental health advice. For travel brands, figuring out how to use ChatGPT to generate personalized itineraries of different destinations to inspire travel booking would be a more specific method of lead generation. It will be a powerful brainstorming tool for ideation for all brand marketers, but only if you know how to ask the right questions.
In other words, if you want to figure out how to leverage tools like ChatGPT for marketing innovations, you’ll have to think outside the box. Ironically, that is something ChatGPT seems rather incapable of doing at the moment. That, as it turns out, still requires true human ingenuity.
What’s Next for ChatGPT’s Budding Marketing Career
Now that ChatGPT has fully taken off, OpenAI is under pressure from investors to start monetizing its service and turn a profit. This week, after much speculation, OpenAI officially launched a premium tier called ChatGPT Plus, charging for unlimited access and additional perks for $20 per month.
Considering that OpenAI also owns Dall-E, a popular text-to-image generative AI service, one could easily envision a foreseeable future where some of the answers from ChatGPT can be fed directly to Dall-E to generate some visuals as well. Forget using ChatGPT to write school essays; the OpenAI suite may generate an entire powerpoint presentation for you based on one prompt one day. Considering how much of marketing creatives centers around visuals, it would have an even greater impact on how brands use generative AI tools, should this pipeline ever be established.
In addition, the company is also reportedly working on an app version of ChatGPT to help further push mobile adoption. No doubt many companies and knowledge workers will be curious to try it out, but the real change won’t happen until some enterprising individual or startup develops an industry-specific version based on the version of generative AI tech that OpenAI is licensing to Microsoft.
Of course, the moves from competitors, especially Google, will be easier to see in the following months. Already, reports say that Google is asking employees to test potential ChatGPT competitors, including a chatbot called ‘Apprentice Bard,’ which uses Google’s conversation technology LaMDA. Reportedly, Apprentice Bard’s answers can include recent events, a feature ChatGPT has yet to add.
From a broader competitive landscape, considering that Google could potentially integrate its generative AI products into its many services like Gmail and YouTube and Android OS down the road, it’d be interesting to see Apple choose to join the Microsoft/OpenAI camp instead (or perhaps developing its own its Generative AI tools for iOS) in this brewing Generative AI war. (Of course, users should be free to use whichever AI tools they prefer, but never underestimate the power of the default.) Moreover, it will be interesting to see which side the big creator platforms like TikTok and Instagram align themselves with.
