Insights
ChatGPT-3.5 vs GPT-4o: Which is better for marketing?
Insights
Share to
For marketers, the idea of using AI for campaign creation and ideation has become a game-changer since the public launch of ChatGPT in 2022. Since then, OpenAI, the company behind ChatGPT, has launched a few different AI models that include GPT-3.5, GPT-4, and most recently, a new, faster AI model called GPT-4o. Right off the bat from OpenAI’s announcement, we can tell that this new model has huge potential to help marketers generate high-quality content due to its better understanding of written and spoken content, which includes multilingual capabilities.
But wait! Before you launch right into switching to GPT-4o for your next marketing campaign, it’s crucial to understand how much you can and should be relying on these tools. AI is not a replacement for human creativity and judgement, and using AI responsibly requires a thoughtful approach that’ll require some oversight on your part. This can include things like monitoring for tone, verifying accuracy of generated content, and abiding by best practices.
With these precautions in mind and new, freely available generative AI versions of ChatGPT popping up every few months, it’s challenging to know which are best suited to help you with your marketing campaigns. That’s why we tested GPT-3.5 and GPT-4o to see how they might work in real-world marketing scenarios. Let’s compare the two AI models first.
GPT-4, which came out in March 2023, already provides a leg up against GPT-3.5 with advanced reasoning capabilities and more accurate results. According to OpenAI, GPT-4o provides “GPT-4-level intelligence but is much faster,” with improved text, voice, and vision capabilities.
“GPT-4o is twice as fast, costs half as much, and is more accurate than GPT-4. This makes it ideal for enterprise use cases that require high-accuracy language models, which until now were either too slow or expensive to make sense.”
– Joachim Jonkers (JJ), Director of Product – Conversational AI at Sinch
One of the most exciting developments with GPT-4o is its potential for handling voice tasks, which previous models could only do by using significantly more time and energy. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds – matching human response time in a conversation.
And the technology behind these voice prompts is particularly exciting, because before GPT-4o, there were three AI models that made this happen: One that turned speech into text, one that created responses, and another that turned text back into speech. This setup required a lot of processing power, time, and money, and relied on transcribed text to work. But now, GPT-4o has the potential to do by itself what these three versions had to do together, with eventual capabilities to detect emotions in speech, writing, and visuals.
But given the clear use case for AI in marketing, we decided to test its limitations. To do this, we compared GPT-4o with GPT-3.5 in a few common scenarios.
If you’re like many marketers, you’ve probably already tried using AI to speed up your process for creating content that resonates and converts. To see how each free version could help create marketing copy for a mobile marketing campaign, we gave ChatGPT 4o and GPT-3.5 the same prompt:
We want to send out a green marketing campaign about our new zero-waste sunscreen. We want to send it as an MMS campaign. Create a fitting prompt for an audience that’s adventurous, likes to travel, and is between 25 and 45 years old.
Here’s what each AI model gave us:
GPT-4o had a better grasp as to what you can include in an MMS marketing message than GPT-3.5. Its answer recommended specifics to MMS messages like a subject line, a CTA button, emojis, and more than 160 characters. It also suggested personalization by incorporating the recipient’s first name, and added specific aspects that our target audience would care about (ingredients, packaging, reef-safe formula) instead of just saying “eco-friendly” like the GPT-3.5 version did.
GPT-4o also integrated hashtags as part of a broader social media campaign instead of in the text of the message itself, setting up the message to be part of a more engaging, omnichannel campaign.
We noticed, however, that neither message included a recommendation for opt-out language, despite this being a requirement for MMS marketing. This makes the case that it’s still important for you to fact-check and apply responsible best practices to AI-generated campaign content no matter which free GPT model you use..
Next, we wanted to test how and if each AI model could adapt social media copy based on channel and language, which might help if a company is marketing to both English-speaking and non-English speaking demographics.
To test the way that ChatGPT-3.5 and GPT-4o might write English and Spanish marketing copy for LinkedIn and Facebook, we gave each model the same prompt:
We’re a large telecom company launching a new phone plan in Texas and California where new users who sign up get free unlimited inbound and outbound calling and texting to Mexico, the U.S., and Canada for the first month, with no setup fees. Write our announcement social media post for LinkedIn and Facebook in Spanish and English that will maximize re-shares.
The models gave us text in both Spanish and English, but here are the English versions of the social copy:
These answers were the most similar of the three prompts we gave GPT-3.5 and GPT-4o. The Spanish versions from both models (not pictured above) didn’t have any obvious mistakes. GPT-4o’s tone sounded slightly more conversational than GPT-3.5, though they both came across as somewhat pompous.
Looking at the examples above, we found the wording of the actual offer slightly clearer in the GPT-4o version than in GPT-3.5. Our offer promised new users free unlimited inbound and outbound calling and texting capabilities without setup fees. While this includes many free features, we believe the language would need to be clearer than “completely free” to avoid misleading new customers who might expect everything to be free rather than specific aspects of their new plan.
Beyond that, neither model included hashtags in the actual social post itself, nor did they suggest adding images, links, or tagging other profiles. This would have made the post more engaging for re-shares, a key part of the prompt.
And honestly? After reading through these examples, they felt impersonal and generic. This might work for a post or two, but it highlights the importance of editing and adding your own brand voice to any AI-generated copy. If you use similar prompts and copy as your competitors, you risk blending in. And as marketers, we know it’s crucial to emphasize your brand value in all communications – otherwise your message will get lost in the crowd.
Omnichannel e-commerce plays into the idea that customers should be able to have a seamless shopping experience across messaging channels. Whether they’re shopping online, via a WhatsApp click-to-chat ad, or clicking on a link in an email newsletter, everything should feel interconnected and easy.
To see how ChatGPT-3.5 and GPT-4o would handle omnichannel marketing, we asked each to adapt campaign copy that you could use on different marketing channels. Ideally, each channel would be used strategically to keep the conversation going across channels, but since we’re using free versions of this AI, we focused on its ability to adapt its message.
Here’s the prompt we gave ChatGPT-3.5 and GPT-4o to test its knowledge of omnichannel marketing:
We want to send out a special promotion to loyal customers. For this re-engagement campaign, we’re addressing customers that have been with us for over two years and we want to offer them a 30% discount on anything in our shop, and if they recommend us to a friend on social media, they can get 40% off. Create a separate version of this campaign for each of the following channels: SMS, RCS, WhatsApp, and email.
When it came to creating an omnichannel campaign, there was a noticeable improvement in GPT-4o compared to GPT-3.5. GPT-4o adjusted the length of each message for each channel, used emojis, and even included special features like buttons and footers for its RCS messaging campaign.
We found GPT-4o’s tone to be slightly more conversational than GPT-3.5’s. For example, its email copy was longer and slightly more formal, matching what customers expect from this channel.
However, there were some limitations. WhatsApp as a marketing channel allows for buttons, images, automatic replies, and more. To best engage customers, we recommend adding more interactive features and personalization than either AI model suggested. Part of creating an effective omnichannel marketing strategy is about using the unique features of each messaging channel to engage customers effectively.
Additionally, neither GPT included space for the brand’s social media handles. We assume that there might be buttons in the email footer, for example, but it goes to show that you can’t just copy and paste AI-generated content – you need to apply marketing best practices to really make the most of it.
As marketers, we often wear a lot of hats. We can be responsible for creating copy for many different audiences, implementing campaigns, tracking those campaigns, analyzing how they’re contributing to business growth, and then pivoting to make the next campaign even bigger and better. It’s a lot of work, and there’s a clear use case for using AI to help us speed up the process.
And both ChatGPT-3.5 and GPT-4o have clear use cases for marketing: They can generate engaging content for social media, advertisements, emails, and mobile messaging channels to help you increase brand visibility. They can also help you with editing projects, or projects in multiple languages.
But these tools also have their limitations. For one, while GPT-4o can help you better adapt your tone for different marketing channels, it’s only free for short periods of time each day before it reverts to GPT-3.5 (unless you have a paid subscription).
It’s also important to balance your use of AI-generated copy in your marketing campaigns to avoid being impersonal or sending out the same campaigns as your competitors that are also using ChatGPT. As we’ve seen, AI-generated responses can help immensely with first drafts of creative, engaging campaigns, but it’s not the AI’s responsibility to make sure that best practices are abided by – it’s yours! That’s why we recommend making sure that any AI tool you use has a high degree of human touch as well to make sure your marketing remains on-brand, relatable, and authentic.
At Sinch, we’re incredibly excited about the future of AI and already have several applications available for our customers. To make this work, we ensure these tools have necessary guardrails to keep them on track, and always supplement AI-generated content with best practices for optimal results.
Ready to launch your own AI project like a chatbot or AI-powered knowledge base? Reach out to our team – we’re excited to help you build AI experiences that your customers will love.
