One of the more intriguing changes expected to hit the marketing landscape in the coming year is the anticipated growth of generative AI capabilities. An exciting toy to play with in 2023, few marketers were using AI to make meaningful advertising work beyond the PR value of saying, “Hey! We made an ad with AI!” AI was rife with problems from potential trademark and copyright issues to creepy hands rendering on otherwise lifelike AI-generated “people.” But in May of 2023, Google announced the future launch of Gemini, their next-generation large language model (LLM) that promised to change the AI world with enhanced capabilities.
On December 6, members of Google’s “Internal Access” program (early adopters, influencers, employees, and researchers) flooded the web with videos of Gemini as the “CHATGPT slayer.” And then, just as suddenly, it fell silent as the public availability of Gemini was pushed to a later date in 2024. With details of the sudden change limited in online searches, we looked for answers by interviewing the one source we were confident would have the answers, Google Bard.
In this edition of Plain Talk, we’ll find out what Bard had to say about the Gemini launch and the future of AI in marketing applications.
- What Is Google Gemini?
- What Happened to Its Launch?
- Is Google Gemini Better Than ChatGPT 4?
- The Role of AI Prompts in Marketing: “Garbage in Garbage Out”
- Get Expert Help Understanding the Future of AI in Marketing
What Is Google Gemini?
According to Google Bard, Google Gemini is the new LLM from Google. It is unique for several reasons.
First, it was trained on a “massive dataset of text and code specifically related to Google’s products and services.” The purpose of this is to improve the relevance of searches for each user. This makes sense, of course, because Google’s mission, “to organize the world’s information and make it universally accessible and useful,” clearly fits this purpose. But while LLMs like Gemini and ChatGPT are clearly the future of search, Gemini, like Chat, is more than just a search enhancement.
According to our interview with Bard, Gemini offers:
- Multi-modal interaction with users (meaning interaction via audio, text and/or video). This was important because the Gemini announcement on December 6, 2023, came just two weeks after ChatGPT announced similar capabilities in their “Turbo” release.
- Enhanced creative text capabilities. So basically, it can write anything from poems, emails, and scripts to web code and even original music. (Maybe even a generative video soon?)
What Happened to Its Launch?
The December 6 press and social media blitz initially sounded like a public release would happen the following week on December 13. Then something happened. Excited (and skeptical) users on the web began to investigate the December 6 announcement and demos, especially how fast, accurate, and conversational the Gemini chatbot interface was and whether the indicated prompts in some of its demo videos really worked. It turned out that many felt Google had tried to trick them by editing the demo video to make Gemini look faster and easier than it was. This, in turn, kicked up a firestorm of concern in the media that Gemini might not be what was promised. Now, Google is saying that Gemini will be available in “early 2024.”
What Google Bard had to say
We asked Google Bard about what happened. Bard tells us, “Google limited access to Gemini Pro on December 6, 2023. This was done due to some technical issues that were causing the model to generate incorrect or misleading results. Google is working to fix these issues and plans to re-release Gemini Pro to the public in the coming months.” Additionally, a Google spokesperson noted that the video was a “representation” of the results they were getting during testing and that there was no intention to fool the public.
When asked if Google was embarrassed by the launch event mistakes, Bard said, “It’s likely that Google was embarrassed by the mistake of making Gemini Pro available to the public before it was ready. This error could have caused confusion and frustration for users, and it could have also damaged Google’s reputation. However, it’s important to note that mistakes happen, and Google is taking steps to rectify the situation. The company is working to resolve the technical issues that caused the rollback, and it is expected to make Gemini Pro available to the public again in the coming months. In the meantime, Google is providing updates to users and apologizing for the inconvenience.” (We thought that was a refreshingly honest answer.)
Is Google Gemini Better Than ChatGPT 4?
So, let’s assume then that Google just jumped the gun a bit early and that when released, the Gemini products will be as advertised. This is a safe assumption because, after all, they are still Google. Under this assumption, which of these LLMs might be more useful for the purpose of enhancing your marketing? The answer is likely both. Let’s explain.
Apples-to-apples technology
According to Google, Google Gemini outperforms ChatGPT 4 on 17 out of 18 widely accepted logic, text, and multimodal performance tests. While we do not doubt this, these claims do not appear to be independently verified (yet). The nerdiest among our readers are welcome to read the full report here. While the implications of these performance benchmarks are not clearly applicable to your daily marketing efforts, today, they are, if verified, clearly going to be relevant as AdTech catches up to the abilities AI is putting out into the digital world. It’s a horserace, but if Google is right, it will have the superior technology for now.
Google Gemini
According to Google (and Bard), Gemini products are trained on and focused on improving the user experience related specifically to Google products and services. At first, this may sound a little limiting until you remember that Google’s products and services include, well, Google, the omniscient search engine that captures 92% of all global searches, according to StatCounter. Google products and services also include their business and productivity products (Google Ads, Workspace, Cloud, etc.), YouTube (still more than twice as large as TikTok), Google Play and more. So, if your marketing plans rely heavily on organic search, paid search, and online video on Google products, we think it would be a mistake to ignore Gemini.
Google Gemini may be able to help you produce Google-search-friendly content like:
- Social media posts
- Blog content
- Website content
- Native content
But it may be too early to just set Gemini loose on your content needs solo. According to Google Bard, there are some risks.
Bard says that while Gemini has been “trained on a massive dataset,” it is possible “that Gemini could be used to generate spam or low-quality content.” This is because Gemini will know all the “tricks” about SERP but may not understand the difference between high-ranking content and GOOD content. So, Gemini may not always be able to understand the nuances of human language, which also could lead to “errors or misinterpretations.” Bard acknowledges the search-friendly benefits of Gemini are attractive but indicates that Gemini’s ability to produce your content may still be “in the future.”
ChatGPT 4
As it had been pretty candid so far, we decided to ask Google Bard the same question about content created by ChatGPT4. Bard said the following:
“Whether or not content created with ChatGPT 4 will be more search-friendly than content that is not remains to be seen, as this technology is still under development. However, there are a few reasons why it is possible that ChatGPT 4-generated content could be more search-friendly than content generated by other methods.
Massive dataset of text and code
ChatGPT 4 has been trained on a massive dataset of text and code, which includes a variety of publicly available content. This means it has been exposed to a wide range of keywords, phrases, and sentence structures commonly used in search queries. This could make it more likely that ChatGPT 4-generated content will use the terms that people are searching for, which “could improve its ranking in search engine results pages (SERPs).”
Natural and coherent text
ChatGPT 4 is able to generate more natural and coherent text than other language models. This means that the content it produces is likely to be more engaging and informative for users, which could lead to longer dwell times on SERPs. This, in turn, could signal to search engines that the content is high-quality, which could further improve its ranking.
Real-world information
ChatGPT 4 has the ability to incorporate information from the real world through Google Search. This means that it can generate content that is up-to-date and relevant to current events, which “could make it more likely to appear in SERPs for relevant search queries.”
Essentially, both tools are rapidly gaining the ability to help you with content creation, especially to help you with being found in digital search, so wherever you can, it’s important to use LLMs like ChatGPT and Google Gemini (when it is released) to help improve the quality of your marketing communications. But regardless of which LLM you choose (or how many), the tool is only part of the puzzle.
The Role of AI Prompts in Marketing: “Garbage in Garbage Out”
The good news is that AI tools like ChatGPT and Google Gemini are rising to meet us at our respective levels of ignorance on different topics. The bad news is that most LLMs are still heavily reliant on the quality of how we ask them questions. These “prompts” determine how LLMs access data in their massive datasets and assemble it in a response. How well prompts work will also depend on what you are doing with your AI (Is it script writing? Music? Tagline options? etc.) and the dataset, of course on which the AI is trained and how the AI prioritizes your instructions.
A radio spot worthy of Wordsworth
For example, we asked ChatGPT to do the following:
“Write a :15 radio commercial script for Coca-Cola that is in the style of a Wordsworth poem and incorporate relevant pauses for popping open of a can, pouring of soda and for bubbling soda sound effects.”
Pretty specific right? Here’s what we got:
[Soft, melodic music playing in the background]
Narrator (whispering):
Behold the elixir, a nectar divine,
A symphony of bubbles in every line.
Nature’s sweet secret, a dance of delight,
In every drop, joy takes its flight.
[Sound of a can opening]
Narrator (whispering):
Coca-Cola, the muse of our days,
A sip of pure bliss in life’s ballet.
Sparkling whispers, a refreshing song,
In every note, where memories belong.
[Chorus of laughter and clinking glasses]
Narrator (whispering):
Gather ‘round, friends, with hearts full of cheer,
In each shared moment, the magic is clear.
Coca-Cola, a tale in a sip,
In every taste, let your spirit trip.
[Soft music fades out]
Pretty good, right? It is consistent with Wordsworth’s style, too. Unfortunately, our 15 second spot is actually 36 seconds long because the AI decided to prioritize the style and content over the length of our prompt. We also got wildly different options from AI based on suggestions of a style of poetry, a specific poet, length, sound effects, etc. (You’ll want to make sure that your AI did not accidentally plagiarize anything, too.) So, essentially, AI currently gives you options for how to execute a creative strategy and style which (currently) still needs to come from a skilled concept creative.
Human input is still critical
We find the same is true of long-form AI content like blogs. Regardless of the specificity of the prompts, we find that a human is still critical for three reasons. First is the need to be factually correct. Depending on the AI you use and its dataset, we find that some AI blog content can include incorrect or outdated information. Second, to avoid plagiarism. As the number of complaints and even lawsuits grow around the datasets used for training AI, it is important that quotes from others are attributed or hyperlinked to their sources.
Finally, there is tone. If a commercial is written in the style of a Wordsworth poem, there is ample data for AI to pull from to capture that style. Understanding “brand voice” is another thing entirely. We are confident that having a writer who can polish content “for voice” will be important for years to come.
That’s not to say that either of the examples above is not without merit. Artificial intelligence may crank out a 36-second Wordsworth/Coke version intended to be 15 seconds, but it can crank out 50 versions of that prompt in about 30 minutes. That ends up being a lot of raw material for creatives to work with, as well as a lot of options for testing. Is Wordsworth not testing well? How about poetry in the style of Maya Angelou? Childish Gambino? Taylor Swift? The same is true of things like long-format content.
The keys to using AI for marketing
Right now, the keys to using AI in marketing copy and graphics is to:
- Understand what AI does well.
- Have processes in place for testing inputs and outputs.
- Have skilled marketing strategy and creative people who can craft inputs and test and perfect outputs more efficiently.
So regardless of whether you use Google Gemini, ChatGPT, both, or another entrant into the AI space, the result will be more creative options, better testing, and better outcomes. So, while AI is poised to provide you and your team with a higher volume of executional options, it will be critical to keep one hand on the steering wheel to ensure quality.
Get Expert Help Understanding the Future of AI in Marketing
We love talking about how evolving technology intersects with the world of marketing and how to be great stewards of that new technology. If you want to talk AI, give us a call at 502-499-4209 or drop us a note here, and we’ll set up a convenient time to chat.
Our Articles Delivered
Signup to receive our latest articles right in your inbox.