Here at WhatTheyThink, we’ve tried to provide a realistic look at artificial intelligence and its impact on print and marketing, including podcasts. With more than 100 million Americans listening to podcasts each week, we thought we’d take a look what it’s like to use AI to create one. Is it really as easy as inputting a Word file and letting AI do the work for you?
To find out, we look at an advertising campaign created by Wise, an Alpharetta, Ga.-based manufacturer of prime and durable labels for the resale trade. This campaign was for cover-up labels, which are applied over existing labels or packaging to cover up typos, update product information without reprinting the packaging, change pricing, and so on. The finished AI-generated audio can be found at the bottom of the landing page here.
This is a very, very simple audio. So what were the steps used to create it? What did that process look like?
- Notebook LM
The first step was to develop the content from which the script would be based. For this, Wise used Notebook LM, an AI research and learning tool created by Google. Wise input the source materials from which it wanted Notebook LM to generate its output. For this project, Wise input hyperlinks and uploaded copy from Word documents.
Wise hit “generate,” then Notebook LM analyzed all of the inputs and created a two-person discussion on cover-up labels that covered the content nicely. But Wise wasn’t looking for a two-person discussion (which Notebook LM calls an “audio overview”). It was looking for landing page copy. Instead of downloading the audio overview, it downloaded the text summary instead.
- ChatGPT
Next, Wise wanted to convert the summary into something that looked like web copy. To do this, it input the summary into ChatGPT. Wise gave the tool very specific instructions about the length and formatting of the content to provide the best starting point.
- Add the human.
Now it was time for more in-depth involvement by a human being. Wise used the ChatGPT output as a starting point, then refined the copy to speak in the company’s voice, reflect the company’s marketing goals, and speak directly to its reseller audience.
- Use the AI checker.
Although AI was used as a starting point, Wise did not want the landing page to read or sound like AI content. So after reworking the copy, Wise put the text into Grammarly’s AI checker and tweaked it until the percent of text that “appears to be generated by AI” was zero.
- Speechify to create the audio.
Now it was time to create an audio version of the copy that would allow people to listen to the same content if they preferred to do so. For this, Wise used Speechify Studio, which creates AI-generated podcasts/audio files from the content you provide. Speechify Studio also allows users to clone their own voices so that the audiocast sounds like it’s coming from you or your own people.
- More human touches.
Of course, people speak differently than they write, so this again required the expertise of a human being to tweak the script to sound natural when “read” by the AI. The copy was approved, then input into the platform. Rather than selecting one of the many AI voices offered, however, Wise used the option within Speechify Studio to create and use a voice clone of one of its team members instead.
- Even more human touches.
Once the audio was generated, a human being (again) had to listen to it to make sure it sounded natural. This required manual tweaks here and there, both to ensure a natural read and to soften and adjust the AI-generated text.
There is a learning curve to any AI tool, including Speechify Studio, and it required some time for Wise to figure out how to tweak the output to sound the way it wanted it to. For example, using text conventions like hyphens and ellipses to influence the nuance of the output, as well as insert pauses when things felt rushed.
- Download and good to go!
Once approved, the audio file was downloaded and added to the landing page that you can access here.
AI is a terrific tool, and it opens many new doors in marketing to allow companies to create new types of content, and more content more quickly, than ever before. But can it replace humans? You be the judge.
