Using AI as Your Personal Podcast Analyst

Most podcasters are sitting on a treasure trove of information.

Episode titles. Descriptions. Publish dates. Obtain numbers. Transcripts. Listener suggestions. Years of conversations, subjects, codecs, experiments, and choices.

The issue is that almost all of that information is difficult to see all of sudden.

You may have a look at your stats and see which episodes have probably the most downloads. You may scroll via outdated episode titles. You may bear in mind which conversations felt sturdy while you recorded them. However that doesn’t all the time inform the complete story.

Older episodes naturally have extra time to gather downloads. Some titles underperform as a result of they’re unclear. Some subjects work higher than you realized. Some segments could really feel necessary to you, however not join as strongly with listeners.

That’s the place AI will help. Not as a alternative to your creativity. Not as a shortcut round doing the work. However as a mirror.

A superb AI evaluation can replicate your individual present again to you in methods which are exhausting to see when you’re the one making it.

I attempted it with Buzzcast, and inside an hour I used to be seeing patterns in our present that I had missed for years.

This began with a easy query.

I used to be taking a look at Buzzcast episode titles and questioning if there have been patterns within the titles that carried out greatest. So I copied our episode titles, publish dates, durations, and obtain numbers from Buzzsprout and dropped them into ChatGPT.

Nothing fancy. No completely formatted spreadsheet. Simply the essential data from our episode listing.

Then I requested a easy query:

It did a surprisingly good job. Not good. Not magical. However adequate to make me surprise what else may be hiding in our personal information.

Buzzcast has been round for years. We have now a whole bunch of episodes, full transcripts, listener suggestions, and sufficient historical past to start out seeing actual patterns.

I saved going.

I downloaded our transcript file, added it to the dialog, and requested ChatGPT to attach the episode information with the precise content material of the episodes.

Now it was not simply taking a look at titles and downloads. It may see what we really talked about.

The primary lesson: clarify your information

Podcast stats want context.

For those who give AI a listing of episodes and obtain numbers, it’d assume probably the most downloaded episodes are the strongest episodes. However that’s not all the time true.

Older episodes have had extra time to gather downloads. A again catalog episode may decide up a couple of downloads day by day for years. That doesn’t imply it was a stronger episode than one thing you printed three weeks in the past.

So I informed ChatGPT precisely that.

That one instruction made the evaluation significantly better.

It began accounting for episode age. It in contrast newer episodes extra pretty towards older episodes. It seemed for patterns past uncooked obtain totals.

That was the primary actual unlock. Just a few years in the past, this sort of evaluation would have required a programmer or Excel knowledgeable and lots of handbook work. Now you can provide the software a plain-English correction and ask it to regulate.

That doesn’t imply you must belief each reply blindly. You continue to want to take a look at the outcomes and use your judgment. However it means the place to begin is far simpler than it was once.

What AI helped us see

One helpful factor it did shortly was categorize the present. It seemed throughout Buzzcast episodes and grouped our conversations into broad classes. Some had been apparent. Some weren’t.

Tactical podcasting recommendation. Trade information. Creator economics. Occasion recaps. Platform drama.

The primary few made sense. We discuss quite a bit about sensible podcasting recommendation and trade information. That’s core to the present. However a number of the different classes had been extra eye-opening.

If “occasion recaps” present up as a serious class, perhaps we’re spending extra time on occasion recaps than we realized. If “platform drama” reveals up as a recurring theme, perhaps we’re happening that highway extra typically than we must always.

That’s the mirror.

AI didn’t resolve what Buzzcast needs to be. It confirmed us what Buzzcast already was. Then we may ask the higher query: is that what we would like the present to be?

That’s the place this sort of evaluation turns into helpful. It’s not nearly discovering what carried out nicely. It’s about noticing the place your consideration goes and deciding whether or not that also matches the aim of your present.

For us, Buzzcast exists to assist podcasters hold podcasting. So when the evaluation confirmed that our strongest episodes typically answered the query, “How do I make my present higher?” that felt proper.

That’s one thing we must always lean into.

Your again catalog issues

One of many largest takeaways from this experiment is that your again catalog is extra priceless than you assume.

Each outdated episode continues to be an entry level into your present.

Somebody may discover it in a podcast app. They could discover it via Google. They could hear one other listener point out it. They could uncover your present via one episode, get pleasure from it, after which begin scrolling via all the things else you have got printed.

When that occurs, your outdated titles matter. Your descriptions matter. Your subjects matter.

Quite a lot of podcasters publish an episode and transfer on. That’s comprehensible. You all the time have the subsequent episode to make. However your again catalog retains working.

AI will help you look again via outdated episodes and ask helpful questions:

• Which older episodes nonetheless have sturdy discovery potential?
• Which episodes have good content material however weak titles?
• Which subjects appear to have the longest shelf life?
• Which episodes can be most helpful to a brand new listener?
• If somebody found my present right now, which again catalog episodes needs to be best to seek out?
• That may be a very totally different means of taking a look at outdated content material. Your again catalog isn’t just an archive. It’s a part of your discovery engine.

Titles are a very good place to start out

Episode titles had been the simplest place for us to start.

When ChatGPT checked out Buzzcast titles and efficiency, a couple of patterns stood out. Clear, sensible titles carried out nicely. Titles that promised rapid worth carried out nicely. Titles with a transparent impression carried out nicely.

That doesn’t imply each episode must sound like a listicle. You don’t want to show your podcast into clickbait. However readability issues.

A intelligent title that solely is sensible after somebody listens is often not doing sufficient work. A imprecise title may be significant to you, but it surely offers a possible listener little or no motive to press play.

Good titles assist individuals perceive why the episode issues.

That’s very true within the again catalog. If somebody is scrolling via outdated episodes, they’re making fast choices. They’re asking, “Is that this for me?

Your title wants to assist them reply that query.

A easy immediate can get you began:

Then observe up:

• Which titles are too imprecise?
• Which episodes may carry out higher with clearer titles?
• Which titles talk the strongest listener profit?
• Rewrite these titles to be clearer, however hold them sincere and pure.

The purpose is to not trick individuals into listening.

The purpose is to make the worth simpler to see.

Transcripts unlock deeper evaluation

Titles and obtain numbers can train you a large number. Transcripts take it a lot additional.

As soon as AI may have a look at our precise episode transcripts, it began figuring out patterns within the conversations themselves. What sorts of segments confirmed up most frequently. Which subjects created stronger discussions. How episodes had been structured. How we moved from one thought to a different.

It additionally observed one thing I had not anticipated.

Some episodes that seemed common by downloads had generated lots of listener suggestions within the subsequent episode. That turned one other sign of engagement.

An episode may not be your largest obtain spike, but it surely may generate probably the most listener response. It’d result in extra Fan Mail. It’d reply a query your viewers has been quietly wrestling with. It’d turn out to be the episode individuals advocate to pals. That’s price taking note of.

For Buzzcast, the strongest episodes typically mixed a couple of issues: an trade subject, sensible recommendation, and a deeper dialogue about why it issues.

That was useful as a result of it gave us a construction we may use once more. Not a inflexible method. Only a sample.

After we cowl one thing, we must always clarify what occurred, give podcasters one thing helpful to do with it, after which zoom out to why it issues.

That seems like Buzzcast at its greatest.

AI will help you perceive your function

This was most likely probably the most stunning a part of the evaluation.

I requested ChatGPT to research the talking distribution throughout our transcripts.

That seems like a novelty stat, and at first it was. It informed us how a lot every host talks, how typically every individual takes a flip, and the way these patterns have modified over time. A few of it was simply enjoyable.

However then it obtained extra helpful.

I began asking what function every of us appeared to play within the present.

It described Jordan because the listener’s voice. She asks clarifying questions, retains the dialog transferring, and notices when the listener may want extra context.

It described me because the big-picture voice. I are inclined to zoom out, join concepts, and clarify why one thing issues.

It described Alban as extra analytical and technical. He will get into how issues work, what tradeoffs exist, and what’s taking place underneath the floor.

That felt correct and it jogged my memory that every of us has a job to do on the present.

If I don’t convey the big-picture perspective, Jordan and Alban most likely won’t convey it in the identical means. If Jordan doesn’t convey the listener perspective, Alban and I could hold speaking previous the purpose the place a listener wants clarification. If Alban doesn’t convey the sensible and technical evaluation, we could miss necessary tradeoffs.

That’s helpful.

AI didn’t inform us to turn out to be totally different individuals. It helped us see the roles we already play so we will be extra intentional about them.

This might be useful for any present with a number of hosts, visitors, or recurring segments.

You might ask:

• Analyze the talking distribution throughout my podcast transcripts.
• What function does every host seem to play?
• The place do the perfect conversations occur?
• Are there locations the place one host dominates an excessive amount of?
• Are there recurring moments the place the dialog loses momentum?
• What ought to every host lean into extra?

A few of the suggestions could really feel proper. Some could really feel off. Each will be helpful.

If it feels proper, you have got one thing to lean into. If it feels incorrect, you have got one thing to look at.

Use AI for future planning, however solely after it understands the previous

As soon as I had labored via the older episodes, title patterns, transcripts, and efficiency information, I began asking about future episodes.

That order issues.

For those who begin by asking AI for brand spanking new episode concepts, you’ll most likely get generic concepts. Some may be superb, however they won’t be deeply related to your present.

The higher strategy is to let it research your present first. Give it your titles. Give it your stats. Give it your transcripts. Right the assumptions it will get incorrect. Ask follow-up questions. Be certain the evaluation feels grounded in actuality.

Then ask it that can assist you plan.

Then hold pushing:

1. Which of those are too generic?
2. Which concepts greatest match what my viewers already responds to?
3. Which subjects have we not coated not too long ago?
4. Which older subjects are price revisiting from a special approach?
5. Give me 5 episode concepts that reply the core query my viewers appears to care about most.

The outcomes had been significantly better after the AI had context.

It knew what Buzzcast was about. It knew which subjects had labored. It knew which buildings appeared strongest. It knew the roles we naturally play as hosts. That made the strategies extra helpful.

Once more, not good. You continue to want style. You continue to want judgment. You continue to must know your viewers.

You do not want to make this difficult.

Begin with no matter information you have already got. Copy your episode listing. Export your stats if you need extra element. Obtain your transcripts when you’ve got them. Put the data into ChatGPT or one other AI software with a big sufficient context window.

Then begin easy.

Then add context:

Then go deeper:

• What subjects carry out greatest?
• What titles appear clearest?
• Which episodes could also be underperforming due to weak titles?
• What do my strongest episodes have in frequent?
• What does my again catalog counsel I ought to do extra of?
• What recurring segments or codecs appear to work greatest?
• Based mostly on the transcripts, what makes this present distinct?

For those who have no idea what to ask, say that.

That may be a completely good immediate. You would not have to be an AI knowledgeable. You simply must be curious and keen to maintain asking higher questions.

The encouraging a part of this was not that AI may do one thing flashy. It was that AI helped us see our personal present extra clearly.

It helped us discover what was working. It helped us see what we may be overdoing. It helped us take into consideration titles, construction, again catalog, host dynamics, and future subjects with slightly extra intention.

Most podcasters would not have a analysis crew. They don’t have a full-time analyst. They don’t have somebody combing via years of episodes searching for patterns. However now you will get a few of that assist.

You continue to must make the present. You continue to must convey the attitude. You continue to must care in regards to the listener.

AI can level out patterns. You resolve what to do with them.

And if it helps you perceive your present higher, serve your listeners higher, and make the subsequent episode stronger, that is price exploring.

Preserve Podcasting!
Kevin Finn
Buzzsprout Co-Founder