How AI Powers our Firm's Podcast Strategy and Production
"Our company should start a podcast" today is what "our company should start a blog" was ten years ago. On the surface, it's a decent idea - but there's more to doing it in a way that helps drive long-term success.
Similar to a blog, podcasts need:
A content calendar
A steady cadence of content to build up a helpful content library (and to keep people coming back)
A subject matter area to focus on
A consistent voice
Talent dedicated to producing it
Our Womble Bond Dickinson insights portal has plenty of content. We have hundreds of lawyers guiding some of the world's largest companies in dozens of different practice areas. Having a steady flow of content topics wouldn't be an issue.
Producing it would.
We had taken a couple of high-quality swings at the podcast series. We had one In-house Roundhouse that focused on interviewing in-house counsel folks. It was a great niche, and our host was great at driving a good conversation.
We started another one focused on white-collar crime with some fantastic guests, very timely discussion topics, and a couple of our firm's Partners in that practice area doing an excellent job facilitating most of it.
So what's the problem?
Consistency is the main challenge. We had a quality product we were putting out. What's trickier with podcasts that are easier to navigate on written content are the logistics around making something go live. With a podcast, especially an interview, you must align schedules with multiple stakeholders that are usually very crunched for time.
Once everyone lines up a time to talk, you need to prep. You also have to hope the audio quality is good enough for production; everyone has a good mic or minimal background noise.
All that is on top of ensuring you have a content calendar and keeping a consistent brand voice - like someone's literal voice on a podcast.
It was a challenge.
How could we publish content frequently in an audio format in a way that allowed us to bypass the hurdles most podcast production faces?
Artificial Intelligence.
Chris Mammen, a Partner at our firm and a legal thought leader in the AI space right now, introduced me to ElevenLabs. It's a browser-based voice lab that would narrate any text you put in there. He was looking at using it to showcase his insights to launch the firm's new Artificial Intelligence and Machine Learning practice and wanted to set up a podcast as part of that launch.
I took his initial idea and expanded on it. We could re-launch our WBD podcast and have it lead with AI for a while but slowly introduce more insights from other parts of the firm. Chris agreed and was gracious enough to let me use the launch of his AI team as a sandbox for other firm initiatives.
So, Womble Perspectives was born.
What do we use to produce Womble Perspectives?
We can now set up a podcast episode in less than an hour using the following tools:
Buzzsprout
LEGA Creates the Script
LEGA is our firm's internal generative AI tool that uses OpenAI's platform. It gives our firm the ability to use generative AI but reduces the risk of putting client data at risk in the process.
We'll take an insights piece from our website that is either 1) very timely or 2) getting a lot of traffic. Most of our insights pieces are written to be read and not necessarily listened to.
We copy and paste the complete insights piece into LEGA and prompt it to rewrite the article in the format of a radio broadcast (or something along those lines). It spits out a new audio script. We then send that script to our content marketing team to double-check that what the AI tool produced still maintains all the key messages of the piece.
Just because we are using AI doesn't mean we don't need a human element involved.
Recording the Episode
Once everyone aligns on the script, we copy and paste the content into ElevenLabs to create the audio read-out of the script. We take that audio, download it, and then upload it to our podcast editing tool, Kapwing.
Kapwing for Production
We have a pre-cut intro and outro for every episode (also using ElevenLabs) that uses royalty-free music in Kapwing's library. We paste the new "recorded" audio between the intro and outro and then export. That's it. Episode done.
Side note: Kapwing is fantastic. It has a lot of excellent generative AI features to create videos and other visual creative based on prompts. If you want a quick and fun way to spin up creative content without spending much money, it's worth checking out. I've even used it personally to help my kids with school projects.
Buzzsprout for Distribution
We use Buzzsprout for our podcast distribution. After initial setup, it is easy to syndicate your podcasts through any player that people would reasonably use. We upload our new audio file from Kapwing, write a podcast description with links to the complete insights article and relevant author bios, and then hit publish. Buzzsprout does the rest.
One feature Buzzsprout has that is pretty cool is making visually appealing sound clips for social use. You can upload a background image standard for all podcast sound clips and create your custom soundbites. The result looks something like this:
Do We Use Humans?
While 3/4 of what you hear on our podcasts are purely computer generated and not real people, we add a human element. On the weekends, we'll publish an interview-style podcast that humanizes what we do more. This AI approach helps give us more grace periods on podcast episodes that require schedule coordination, sound quality, and more preparation. If scheduling conflicts arise, we at least have a content queue rolling so as not to fall totally off folks' radars.
So there you have it—our internal process on how we go about making podcasts. There's a lot we want to do that we need to do, but it's a good start and a fun way for our whole team to dive into generative AI in a way that is manageable but also useful to our audience.
Any suggestions on how we could get better? Any tools we should look into?